Expert Systems with Applications 36(2009)12520-12528 Contents lists available at Science Direct Expert Systems with Applications ELSEVIER journalhomepagewww.elsevier.com/locate/eswa A multi-disciplinar recommender system to advice research resources in University Digital Libraries C Porcel ,J.M. Moreno, E. Herrera-Viedma University of jaen, Department of Computer Science, Jaen, Spain University of Murcia, Department of Computer Science, Murcia, Spain University of granada, Department of Computer Science and AL, Granada, Spain ARTICLE INFO ABSTRACT Web is recommender systen other media, as for example, newspapers, journal d libraries. In this paper, we analyze the log- Fuzzy linguistic modeling ical extensions of traditional libraries in the Information Society. In Information Society people want to University Digital Libraries communicate and collaborate. So, libraries must develop services for connecting people together in infor- nation environments. Then, the library staff need automatic techniques to facilitate so that a great num ber of users can access to a great number of resources. Recommender systems are tools whose objective is to evaluate and filter the great amount of information available on the Web to assist the in their information access processes. We present a model of a fuzzy linguistic recommender system to he the University Digital Libraries users to access for their research resources. This system recommend researchers specialized and complementary resources in order to discover collaboration possibilities to form multi-disciplinar groups In this way. this system increases social collaboration possibilities in a uni versity framework and contributes to improve the services provided by a University Digital Library. e 2009 Elsevier Ltd. All rights reserved. 1 Introduction or areas of interest. The library staff searches, evaluates, selects, catalogues, classifies, preserves and schedules the digital docu- In the last years the new concept of digital library is growing. ments access(Goncalves et al., 2004). Some of the main digital li- Digital libraries are information collections that have associated braries functions are the following: services delivered to user communities using a variety of technol- gies. The information collections can be scientific, business or per- To evaluate and select digital materials to add in its repository. video, or other media. This information can be displayer audio sonal data, and can be represented as digital text, image, audio, of the ma To describe and index the new digital materials (catalogue and digitalized paper or born digital material and the services offered classify). on such information can be varied and can be offered to individuals to deliver users the material stored in the libra or user communities(Callan et al., 2003: Goncalves, Fox, Watson, Other managerial tasks Kipp, 2004: Renda Straccia, 2005) Digital libraries are the logical extensions of physical libraries in Libraries offer different types of references and referral services the electronic information society. These extensions amplify exist -(e.g, ready reference, exhaustive search, and selective dissemina- ing resources and services. As such, digital libraries offer new levels tion of information), instructional services (e. g, bibliographi of access to broader audiences of users and new opportunities for instruction and database searching ) added value services(e. g, bib- the library. In practice, a digital library makes its contents and ser- liography preparation, and language translation) and promotional vices remotely accessible through networks such as the Web or services(e.g, literacy and freedom of expression ). As digital li- limited-access intranets(Marchionini, 2009). braries become commonplace and as their contents and services The digital libraries are composed of human resources(staff) become more varied, the users expect more sophisticated services that take over handle and enable the users to access the documents from their digital libraries (Callan et al, 2003: Goncalves et al. that are more interesting for them, taking into account their needs 2004: Renda Straccia, 2005). A service that is particularly important is the selective dissem ination of information or filtering(Morales del Castillo, Pedraza- E-mail addresses: cporcele (C Porce)). jmmorenoeumes (M. Moreno). Jimenez, Ruiz, Peis, Herrera-Viedma, 2009: Morales del Castillo viedma@desai. ugres(E. Herrera-Viedma) Peis, Moreno, Herrera-Viedma, in press). Users develop profiles 0957-4174/- see front matter o 2009 Elsevier Ltd. All rights reserved. do:101016eswa200904.038
A multi-disciplinar recommender system to advice research resources in University Digital Libraries C. Porcel a,*, J.M. Moreno b , E. Herrera-Viedma c aUniversity of Jaen, Department of Computer Science, Jaen, Spain bUniversity of Murcia, Department of Computer Science, Murcia, Spain cUniversity of Granada, Department of Computer Science and A.I., Granada, Spain article info Keywords: Recommender systems Fuzzy linguistic modeling University Digital Libraries abstract The Web is one of the most important information media and it is influencing in the development of other media, as for example, newspapers, journals, books, and libraries. In this paper, we analyze the logical extensions of traditional libraries in the Information Society. In Information Society people want to communicate and collaborate. So, libraries must develop services for connecting people together in information environments. Then, the library staff need automatic techniques to facilitate so that a great number of users can access to a great number of resources. Recommender systems are tools whose objective is to evaluate and filter the great amount of information available on the Web to assist the users in their information access processes. We present a model of a fuzzy linguistic recommender system to help the University Digital Libraries users to access for their research resources. This system recommends researchers specialized and complementary resources in order to discover collaboration possibilities to form multi-disciplinar groups. In this way, this system increases social collaboration possibilities in a university framework and contributes to improve the services provided by a University Digital Library. 2009 Elsevier Ltd. All rights reserved. 1. Introduction In the last years the new concept of digital library is growing. Digital libraries are information collections that have associated services delivered to user communities using a variety of technologies. The information collections can be scientific, business or personal data, and can be represented as digital text, image, audio, video, or other media. This information can be displayed on the digitalized paper or born digital material and the services offered on such information can be varied and can be offered to individuals or user communities (Callan et al., 2003; Gonçalves, Fox, Watson, & Kipp, 2004; Renda & Straccia, 2005). Digital libraries are the logical extensions of physical libraries in the electronic information society. These extensions amplify existing resources and services. As such, digital libraries offer new levels of access to broader audiences of users and new opportunities for the library. In practice, a digital library makes its contents and services remotely accessible through networks such as the Web or limited-access intranets (Marchionini, 2009). The digital libraries are composed of human resources (staff) that take over handle and enable the users to access the documents that are more interesting for them, taking into account their needs or areas of interest. The library staff searches, evaluates, selects, catalogues, classifies, preserves and schedules the digital documents access (Gonçalves et al., 2004). Some of the main digital libraries functions are the following: To evaluate and select digital materials to add in its repository. To preserve the security and conservation of the materials. To describe and index the new digital materials (catalogue and classify). To deliver users the material stored in the library. Other managerial tasks. Libraries offer different types of references and referral services (e.g., ready reference, exhaustive search, and selective dissemination of information), instructional services (e.g., bibliographic instruction and database searching), added value services (e.g., bibliography preparation, and language translation) and promotional services (e.g., literacy and freedom of expression). As digital libraries become commonplace and as their contents and services become more varied, the users expect more sophisticated services from their digital libraries (Callan et al., 2003; Gonçalves et al., 2004; Renda & Straccia, 2005). A service that is particularly important is the selective dissemination of information or filtering (Morales del Castillo, PedrazaJiménez, Ruíz, Peis, & Herrera-Viedma, 2009; Morales del Castillo, Peis, Moreno, & Herrera-Viedma, in press). Users develop profiles 0957-4174/$ - see front matter 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2009.04.038 * Corresponding author. E-mail addresses: cporcel@ujaen.es (C. Porcel), jmmoreno@um.es (J.M. Moreno), viedma@decsai.ugr.es (E. Herrera-Viedma). Expert Systems with Applications 36 (2009) 12520–12528 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
C Porcel et aL /Expert Systems with Applications 36(2009)12520-12528 12521 that reveals their areas of interest and as new materials are added A variety of techniques have been proposed as the basis for to the collection, they are compared to the profiles and relevant ommender systems. We can distinguish four different classes of items are sent to the users(Marchionini, 2009). ecommendation techniques based on the source of knowledge One interesting extension of this concept is to use the connec -(Burke, 2007: Hanani et al., 2001; Reisnick Varian, 1997): tivity inherent in digital libraries to support collaborative filtering. here users rate or add value to information objects and these rat Content-based systems: They generate the recommendations ings are shared with a large community, so that popular items can king into account the terms used in the items representation be easily located or people can search for objects found useful by and the ratings that a user has given to them(Basu, Hirsh, others with similar profiles(hanani, Shapira, Shoval, 2001: Mar- Cohen, 1998: Claypool, Gokhale, Miranda, 1999). These rec- hionini, 2009: Reisnick Varian, 1997). ommender systems tend to fail when little is known about the Digital libraries have been applied in many contexts but in this ser information needs paper we focus on an academic environment. University Digital Li Collaborative systems: The system generates recommendations braries(UDls) provide information resources and services to stu- using explicit or implicit preferences from many users, ignor ents, faculty and staff in an environment that supports learning. items representation. Collab teaching and research( Chao, 2002). peer users with a rating history similar to the current user In this paper we propose a fuzzy linguistic recommender sys- and they generate recommendations using this neighborhood tem to achieve major advances in the activities of UDL in order (Good et aL, 1999: Renda& Straccia, 2005 ). to improve their performance. The system is oriented to research Demographic systems: A demographic recommender system ers and it recommends two types of resources: in the first place. provides recommendations based on a demographic profile specialized resources of the user research area, and in the second of the user. Recommended items can be generated for differ place, complementary resources in order to include resources of ent demographic niches, by combining the ratings of users related areas that could be interesting to discover collaboration in those niches(Pazzani, 1999) possibilities with other researchers and to form multi-disciplina Knowledge-based systems: These systems generate the recom- groups. As in( Porcel, Lopez-Herrera, Herrera-Viedma, 2009) we endations based on the inferences about items that satisfy combine a recommender system, to filter out the information, with the users from the information provided by each user regard a multi-granular Fuzzy Linguistic Modeling(FLm), to represent and ing his her knowledge about items that can be recommended handle flexible information by means of linguistic labels(Chang Wang, Wang, 2007; Chen& Ben-Arieh, 2006: Herrera Martinez 2001: Herrera-Viedma, Cordon, Luque, Lopez, Munoz, 2003: All these techniques have benefits and disadvantages. However, Herrera-Viedma, Martinez, Mata, Chiclana, 2005: Herrera, we can use a hybrid approach to smooth out the disadvantages of Herrera-Viedma, Martinez, 2008). each one of them and to exploit their benefits(basu et al, 1998 The paper is structured as follows. Section 2 revises some pre- Claypool et al, 1999: Good et al., 1999). In these kind of systems. iminaries, i.e., the concept and main aspects about recommender the users' information preferences can be used to define user pro- systems and the approaches of FLM that we use to the system de- files that are applied as filters to streams of documents. Therefore, sign, the 2-tuple FLM and the multi-granular FLM. In Section 3 we the construction of accurate profiles is a key task and the system present a multi-disciplinar fuzzy linguistic recommender systems success will depend on a large extent on the ability of the learned to advice research resources in UDL Section 4 reports the system profiles to represent the user's preferences (Quiroga Mostafa evaluation and some experimental results. Finally, some conclud- 2002). narks are pointed out. The recommendation activity is followed by a relevance feed back phase Relevance feedback is a cyclic process whereby the user feeds back into the system decisions on the relevance of retrieved 2. Preliminaries documents and the system then uses these evaluations to auto- matically update the user profile( Hanani et al, 2001; Reisnick 2.1. Recommender systems Recommender systems could be defined as systems that pro- 2. 2. Fuzzy linguistic modeling ndividualized recommendations as output or has the effect hiding the user in a personalized way to interesting or useful The use of fuzzy sets theory has given very good results for cts in a large space of possible options(Burke 2002). They modeling qualitative information(Zadeh, 1975)and it has pro- becoming popular tools for reducing information overload ven to be useful in many problems, e.g., in decision making and for improving the sales in nerce web sites(Burke,(Cabrerizo, Alonso, Herrera-Viedma, 2009: Herrera, Herrera 2007: Cao Li, 2007: Hsu, 2008: Reisnick Varian, 1997) Viedma, Verdegay, 1996: Mata, Martinez, Herrera-Viedma It is a research area that offers tools for discriminating between 2009), quality evaluation(Herrera-Viedma, Pasi, Lopez-Herrera relevant and irrelevant information by providing personalized Porcel, 2006: Herrera-Viedma Peis, 2003 ). models of ssistance for continuous information accesses, filtering the infor- information retrieval(Herrera-Viedma, 2001 a, 2001 b: Herrera mation and delivering it to people who need it(Reisnick Varian, Viedma Lopez-Herrera, 2007: Herrera-Viedma, Lopez-Herrera 997). Automatic filtering services differ from retrieval services in Luque, Porcel, 2007: Herrera-Viedma, Lopez-Herrera, Porcel that in filtering the corpus changes continuously, the users have 2005), and political analysis(Arfi, 2005). It is a tool based on long time information needs( described by mean of user profiles the concept of linguistic variable proposed by Zadeh(1975 ). Next instead of to introduce a query into the system )and their objective we analyze the two approaches of FLM that we use in is to remove irrelevant data from incoming streams of data items system (Hanani et al., 2001; Marchionini, 2009: Reisnick varian, 1997). A result from a recommender system is understood com- 2. 2.1. The 2-tuple fuzzy linguistic mendation, an option worthy of consideration; a result from an The 2-tuple FLM(Herrera 2,2000)isa information retrieval system is interpreted as a match to the users of which allows query(Burke, 2007). he loss of information typical of other fuzzy linguisti
that reveals their areas of interest and as new materials are added to the collection, they are compared to the profiles and relevant items are sent to the users (Marchionini, 2009). One interesting extension of this concept is to use the connectivity inherent in digital libraries to support collaborative filtering, where users rate or add value to information objects and these ratings are shared with a large community, so that popular items can be easily located or people can search for objects found useful by others with similar profiles (Hanani, Shapira, & Shoval, 2001; Marchionini, 2009; Reisnick & Varian, 1997). Digital libraries have been applied in many contexts but in this paper we focus on an academic environment. University Digital Libraries (UDLs) provide information resources and services to students, faculty and staff in an environment that supports learning, teaching and research (Chao, 2002). In this paper we propose a fuzzy linguistic recommender system to achieve major advances in the activities of UDL in order to improve their performance. The system is oriented to researchers and it recommends two types of resources: in the first place, specialized resources of the user research area, and in the second place, complementary resources in order to include resources of related areas that could be interesting to discover collaboration possibilities with other researchers and to form multi-disciplinar groups. As in (Porcel, López-Herrera, & Herrera-Viedma, 2009) we combine a recommender system, to filter out the information, with a multi-granular Fuzzy Linguistic Modeling (FLM), to represent and handle flexible information by means of linguistic labels (Chang, Wang, & Wang, 2007; Chen & Ben-Arieh, 2006; Herrera & Martínez, 2001; Herrera-Viedma, Cordón, Luque, López, & Muñoz, 2003; Herrera-Viedma, Martínez, Mata, & Chiclana, 2005; Herrera, Herrera-Viedma, & Martínez, 2008). The paper is structured as follows. Section 2 revises some preliminaries, i.e., the concept and main aspects about recommender systems and the approaches of FLM that we use to the system design, the 2-tuple FLM and the multi-granular FLM. In Section 3 we present a multi-disciplinar fuzzy linguistic recommender systems to advice research resources in UDL. Section 4 reports the system evaluation and some experimental results. Finally, some concluding remarks are pointed out. 2. Preliminaries 2.1. Recommender systems Recommender systems could be defined as systems that produce individualized recommendations as output or has the effect of guiding the user in a personalized way to interesting or useful objects in a large space of possible options (Burke, 2002). They are becoming popular tools for reducing information overload and for improving the sales in e-commerce web sites (Burke, 2007; Cao & Li, 2007; Hsu, 2008; Reisnick & Varian, 1997). It is a research area that offers tools for discriminating between relevant and irrelevant information by providing personalized assistance for continuous information accesses, filtering the information and delivering it to people who need it (Reisnick & Varian, 1997). Automatic filtering services differ from retrieval services in that in filtering the corpus changes continuously, the users have long time information needs (described by mean of user profiles instead of to introduce a query into the system) and their objective is to remove irrelevant data from incoming streams of data items (Hanani et al., 2001; Marchionini, 2009; Reisnick & Varian, 1997). A result from a recommender system is understood as a recommendation, an option worthy of consideration; a result from an information retrieval system is interpreted as a match to the user’s query (Burke, 2007). A variety of techniques have been proposed as the basis for recommender systems. We can distinguish four different classes of recommendation techniques based on the source of knowledge (Burke, 2007; Hanani et al., 2001; Reisnick & Varian, 1997): Content-based systems: They generate the recommendations taking into account the terms used in the items representation and the ratings that a user has given to them (Basu, Hirsh, & Cohen, 1998; Claypool, Gokhale, & Miranda, 1999). These recommender systems tend to fail when little is known about the user information needs. Collaborative systems: The system generates recommendations using explicit or implicit preferences from many users, ignoring the items representation. Collaborative systems locate peer users with a rating history similar to the current user and they generate recommendations using this neighborhood (Good et al., 1999; Renda & Straccia, 2005). Demographic systems: A demographic recommender system provides recommendations based on a demographic profile of the user. Recommended items can be generated for different demographic niches, by combining the ratings of users in those niches (Pazzani, 1999). Knowledge-based systems: These systems generate the recommendations based on the inferences about items that satisfy the users from the information provided by each user regarding his/her knowledge about items that can be recommended (Burke, 2002). All these techniques have benefits and disadvantages. However, we can use a hybrid approach to smooth out the disadvantages of each one of them and to exploit their benefits (Basu et al., 1998; Claypool et al., 1999; Good et al., 1999). In these kind of systems, the users’ information preferences can be used to define user pro- files that are applied as filters to streams of documents. Therefore, the construction of accurate profiles is a key task and the system’s success will depend on a large extent on the ability of the learned profiles to represent the user’s preferences (Quiroga & Mostafa, 2002). The recommendation activity is followed by a relevance feedback phase. Relevance feedback is a cyclic process whereby the user feeds back into the system decisions on the relevance of retrieved documents and the system then uses these evaluations to automatically update the user profile (Hanani et al., 2001; Reisnick & Varian, 1997). 2.2. Fuzzy linguistic modeling The use of fuzzy sets theory has given very good results for modeling qualitative information (Zadeh, 1975) and it has proven to be useful in many problems, e.g., in decision making (Cabrerizo, Alonso, & Herrera-Viedma, 2009; Herrera, HerreraViedma, & Verdegay, 1996; Mata, Martínez, & Herrera-Viedma, 2009), quality evaluation (Herrera-Viedma, Pasi, López-Herrera, & Porcel, 2006; Herrera-Viedma & Peis, 2003), models of information retrieval (Herrera-Viedma, 2001a, 2001b; HerreraViedma & López-Herrera, 2007; Herrera-Viedma, López-Herrera, Luque, & Porcel, 2007; Herrera-Viedma, López-Herrera, & Porcel, 2005), and political analysis (Arfi, 2005). It is a tool based on the concept of linguistic variable proposed by Zadeh (1975). Next we analyze the two approaches of FLM that we use in our system. 2.2.1. The 2-tuple fuzzy linguistic approach The 2-tuple FLM (Herrera & Martínez, 2000) is a continuous model of representation of information which allows to reduce the loss of information typical of other fuzzy linguistic approaches C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528 12521
(classical and ordinal(Herrera Herrera-Viedma, 1997: Zadeh, that transform without loss of information numerical values 1975)). To define it we have to establish the 2-tuple representation into linguistic 2-tuples and viceversa, any of the existing model and the 2-tuple computational model to represent and aggregation operator can be easily extended for dealing with aggregate the linguistic information, respectively linguistic 2-tuples. Some examples are: Let s= Sgt be a linguistic term set with odd cardinali where the mid term represents a indifference value and the rest of Definition 3(Arithmetic mean). Let x=[( 1, 1),.(rn, Mn))be a the terms is symmetrically related to it. We assume that the set of linguistic 2-tuples, the 2-tuple arithmetic mean X is semantics of the labels is given by means of triangular membership computed as, strib 0nwha0edns1 In this fuzzy lin- X[,2)…(m=4(∑(,x)=4 guistic context, if a symbolic method(Herrera Herrera-Viedma 1997: Herrera et al., 1996) aggregating linguistic information ob- tains a value BE[0, g and B#(O, -.g, then an approximation Definition 4 (Weighted average operator). Let x=I(1, a1),.... function is used to express the result in S. (rn, an)I be a set of linguistic 2-tuples and W=WI, .. wn be their associated weights. The 2-tuple weighted average x is: Definition 1 Herrera and Martinez, 2000. Let b be the result of an aggregation of the indexes of a set of labels assessed in a linguistic term set S, i.e. the result of a symbolic aggregation operation. (n,21)…、(m,=4(2=10nx)-=4( BE[O, g. Let i= round(B)and a=B-i be two values, such that, iE [O g] and a E[-5.5 )then a is called a Symbolic Translation The 2-tuple fuzzy linguistic approach is developed from the Definition 5(Linguistic weighted average operator). Let concept of symbolic translation by representing the linguistic x=[(r1,1),.,(Ta, n)) be a set of linguistic 2-tuples and nformation by means of 2-tuples (S, M),S E S and a E[-5.5) weights. The 2-tuple linguistic weighted average xr is S represents the linguistic label of the information, and a, is a numerical value expressing the value of the translation xW((r1, a1).(w1, a").((m, an), (wn, a-M))=4 B1·Bw from the original result B to the closest index label, i, in the lin guistic term set(s∈S) with B,=4"(r, a4)and Bw,=4"(w, a) This model defines a set of transformation functions between numeric values and 2-tuples. 2. 2.2. The multi-granular fuzzy linguistic modeling Definition 2(Herrera and Martinez, 2000). LetS=(So,., s be a In any fuzzy linguistic approach, an important parameter to determine is the" granularity of uncertainty", i.e. the cardinality linguistic term set and Be [0 g] a value representing the result of a of the linguistic term set S. According to the uncertainty degre ymbolic aggregation operation, then the 2-tuple that expresses that an expert qualifying a phenomenon has on it. the linguistic the equivalent information to B is obtained with the following term set chosen to provide his knowledge will have more or less A:0.g→s×-0.50.5) ent granularity of uncertainty are necessary(herrera Martinez, 4(0=(s.x,wit5 i=round(e). 2001: Herrera-Viedma et al, 2005 ). The use of different labels sets la=B-i aE[-5.5) to assess information is also necessary when an expert has to as- sess different ts,as for example it happens in information where round()is the usual round operation, St has the closest index uate the importance of the query terms el to"p"and"a"is the value of the symbolic translation and the relevance of the retrieved documents(Herrera-Viedma et al., 2003). In such situations, we need tools to manage multi- For all 4 there exists 4", defined as 4"(st,a)=i+a. On the granular linguistic information. In(Herrera Martinez, 2001)a other hand, it is obvious that the conversion of a linguistic term multi-granular 2-tuple FLm based on the concept of linguistic hier into a linguistic 2-tuple consists of adding a symbolic translation archy is proposed. alue of o:s∈S→(S,0 A Linguistic Hierarchy, LH, is a set of levels I(t, n(t)), i.e., The computational model is defined by presenting the following LH=U l(t, n(t), where each level t is a linguistic term set with a different granularity n(t) from the remaining of levels of the hier- archy. The levels are ordered according to their granularity, i.e.a (1)Negation operator: Neg((s, ax))=4(g-(4"(5,, ax) level t 1 provides a linguistic refinement of the previous level t. (2)Comparison of 2-tuples(Sk, a,)and(s1, a2): We can define a level from its predecessor level as: I(t,n(t)) If k %2 then(Sk, %,)is bigger than(S, a2). (3)Aggregation operators. The aggregation of information con- Table 1 sists of obtaining a value that summarizes a set of values, Linguistic hierarchies. therefore, the result of the aggregation of a set of 2-tuples must be a 2-tuple. In the literature we can find many aggre- gation operators which allow us to combine the information l(t, n(t)) I(2.5) according to different criteria Using functions 4 and 4-l (,n(r)
(classical and ordinal (Herrera & Herrera-Viedma, 1997; Zadeh, 1975)). To define it we have to establish the 2-tuple representation model and the 2-tuple computational model to represent and aggregate the linguistic information, respectively. Let S ¼ fs0; ... ; sg g be a linguistic term set with odd cardinality, where the mid term represents a indifference value and the rest of the terms is symmetrically related to it. We assume that the semantics of the labels is given by means of triangular membership functions and we consider that all terms are distributed on a scale on which a total order is defined, si 6 sj () i 6 j. In this fuzzy linguistic context, if a symbolic method (Herrera & Herrera-Viedma, 1997; Herrera et al., 1996) aggregating linguistic information obtains a value b 2 ½0; g, and b R f0; ... ; gg; then an approximation function is used to express the result in S. Definition 1 Herrera and Martínez, 2000. Let b be the result of an aggregation of the indexes of a set of labels assessed in a linguistic term set S, i.e., the result of a symbolic aggregation operation, b 2 ½0; g. Let i ¼ roundðbÞ and a ¼ b i be two values, such that, i 2 ½0; g and a 2[.5,.5 )then a is called a Symbolic Translation. The 2-tuple fuzzy linguistic approach is developed from the concept of symbolic translation by representing the linguistic information by means of 2-tuples ðsi; aiÞ; si 2 S and ai 2[.5,.5): si represents the linguistic label of the information, and ai is a numerical value expressing the value of the translation from the original result b to the closest index label, i, in the linguistic term set ðsi 2 SÞ. This model defines a set of transformation functions between numeric values and 2-tuples. Definition 2 (Herrera and Martínez, 2000). Let S ¼ fs0; ... ; sg g be a linguistic term set and b 2 ½0; g a value representing the result of a symbolic aggregation operation, then the 2-tuple that expresses the equivalent information to b is obtained with the following function: D : ½0; g ! S ½0:5; 0:5Þ DðbÞ¼ðsi; aÞ; with si i ¼ roundðbÞ; a ¼ b i a 2 ½:5; :5Þ; where roundðÞ is the usual round operation, si has the closest index label to ‘‘b” and ‘‘a” is the value of the symbolic translation. For all D there exists D1 , defined as D1 ðsi; aÞ ¼ i þ a. On the other hand, it is obvious that the conversion of a linguistic term into a linguistic 2-tuple consists of adding a symbolic translation value of 0 : si 2 S ) ðsi; 0Þ. The computational model is defined by presenting the following operators: (1) Negation operator: Negððsi; aÞÞ ¼ Dðg ðD1 ðsi; aÞÞÞ. (2) Comparison of 2-tuples ðsk; a1Þ and ðsl; a2Þ: If k a2 then ðsk; a1Þ is bigger than ðsl; a2Þ. (3) Aggregation operators. The aggregation of information consists of obtaining a value that summarizes a set of values, therefore, the result of the aggregation of a set of 2-tuples must be a 2-tuple. In the literature we can find many aggregation operators which allow us to combine the information according to different criteria. Using functions D and D1 that transform without loss of information numerical values into linguistic 2-tuples and viceversa, any of the existing aggregation operator can be easily extended for dealing with linguistic 2-tuples. Some examples are: Definition 3 (Arithmetic mean). Let x ¼ fðr1; a1Þ; ... ;ðrn; anÞg be a set of linguistic 2-tuples, the 2-tuple arithmetic mean xe is computed as, xe ½ðr1; a1Þ; ... ;ðrn; anÞ ¼ D Xn i¼1 1 n D1 ðri; aiÞ ! ¼ D 1 n Xn i¼1 bi !: Definition 4 (Weighted average operator). Let x ¼ fðr1; a1Þ; ... ; ðrn; anÞg be a set of linguistic 2-tuples and W ¼ fw1; ... ; wng be their associated weights. The 2-tuple weighted average xw is: xw½ðr1; a1Þ; ... ;ðrn; anÞ ¼ D Pn i¼1D1 ðri P ; aiÞ wi n i¼1wi ! ¼ D Pn i¼1bi P wi n i¼1wi : Definition 5 (Linguistic weighted average operator). Let x ¼ fðr1; a1Þ; ... ;ðrn; anÞg be a set of linguistic 2-tuples and W ¼ fðw1; aw 1 Þ; ... ;ðwn; aw n Þg be their linguistic 2-tuple associated weights. The 2-tuple linguistic weighted average xw l is: xw l ½ððr1; a1Þ;ðw1; aw 1 ÞÞ ððrn; anÞ;ðwn; aw n ÞÞ ¼ D Pn i¼1bi P bWi n i¼1bWi !; with bi ¼ D1 ðri; aiÞ and bWi ¼ D1ðwi; aw i Þ. 2.2.2. The multi-granular fuzzy linguistic modeling In any fuzzy linguistic approach, an important parameter to determine is the ‘‘granularity of uncertainty”, i.e., the cardinality of the linguistic term set S. According to the uncertainty degree that an expert qualifying a phenomenon has on it, the linguistic term set chosen to provide his knowledge will have more or less terms. When different experts have different uncertainty degrees on the phenomenon, then several linguistic term sets with a different granularity of uncertainty are necessary (Herrera & Martínez, 2001; Herrera-Viedma et al., 2005). The use of different labels sets to assess information is also necessary when an expert has to assess different concepts, as for example it happens in information retrieval problems, to evaluate the importance of the query terms and the relevance of the retrieved documents (Herrera-Viedma et al., 2003). In such situations, we need tools to manage multigranular linguistic information. In (Herrera & Martínez, 2001) a multi-granular 2-tuple FLM based on the concept of linguistic hierarchy is proposed. A Linguistic Hierarchy, LH, is a set of levels l(t,n(t)), i.e., LH ¼ S tlðt; nðtÞÞ, where each level t is a linguistic term set with a different granularity nðtÞ from the remaining of levels of the hierarchy. The levels are ordered according to their granularity, i.e., a level t þ 1 provides a linguistic refinement of the previous level t. We can define a level from its predecessor level as: lðt; nðtÞÞ ! lðt þ 1; 2 nðtÞ 1Þ. Table 1 shows the granularity needed in each linguistic term set of the level t depending on the value n(t) defined in the first level (3 and 7, respectively). A graphical example of a linguistic hierarchy is shown in Fig. 1. Table 1 Linguistic hierarchies. Level 1 Level 2 Level 3 lðt; nðtÞÞ lð1; 3Þ lð2; 5Þ lð3; 9Þ lðt; nðtÞÞ lð1; 7Þ lð2; 13Þ 12522 C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528
C Porcel et aL / Expert Systems with Applications 36(2009)12520-12528 tion managed by digital libraries increases. Digital libraries must move from being passive, with little adaptation to their users, to being more proactive in offering and tailoring information for indi- viduals and communities, and in supporting community efforts to capture, structure and share knowledge(Callan et al, 2003: Gong alves et al, 2004: Renda Straccia, 2005). So, the digital libraries should anticipate the users'needs and recommend about resources that could be interesting for them. We present a hybrid recommender system that combines both the content-based and collaborative approaches(Burke, 2007 Hanani et al, 2001; Lekakos Giaglis, 2006). The the incoming information stream and delivers it to the suitable researchers according to their research areas. It recommends users research resources of their own research areas and of complemen- ry areas. We use typical similarity functions based on threshold values to identify research resources of the own areas( Porcel et al., 2009). For example we could use the threshold semantic Fig. 1. Linguistic hierarchy of 3, 5 and 9 labels. unctions defined in Information Retrieval to evaluate weighted ueries(Bordogna Pasi, 1993: Korfhage 1997). On the other hand to identify research resources of the complementary areas. erhras and msaefuti te zpoese t mortstrated ah i the stiginstric r. 2007. use Gaussian similarity functions( Bordogna Pasi, 1993: Yag lation and allow to combine ulti-granular linguistic information To represent the linguistic information we use different label Car- functions between labels from different levels was defined allow a higher flexibility in the communication processes of the Definition 6. Let LH=U(t, n(t)be a linguistic hierarchy whose system. Therefore, the system uses different label sets(S1, S2,-) inguistic term sets are denoted as sn()=(som L. The transformation function between a 2-tuple that belongs to level t to represent the different concepts to be assessed in its filtering and another 2-tuple in level tyt is defined as activity. These label sets S are chosen from those label sets of LH, i.e. SI ELH. We should point out that the number of different label sets that we can use is limited by the number of levels of Fr: I(t, n(o)I(t', n(t)) LH, and therefore, in many cases the label sets Si and S, can be as TF,x)=/4-1(,x2,(m)-1 iated to a same label set of Lh but with different interpretations pending on the concept to be modeled In our system, we distin- guish between three concepts that can be assessed: As it was pointed out in Herrera and Martinez (2001) this family of (S1)of a discipline with respec transformation functions is bijective. This result guarantees that the Irce scope or user pr transformations between levels of a linguistic hierarchy are carried Relevance degree(S2)of a resource for a user. out without loss of information. To define the computational model Complementary degree(S3)between the resource scope and we select a level to make uniform the information(for instance, the er topics of interest. greatest granularity level)and then we can use the operators de- Following the linguistic hierarchy shown in Fig. 1, in our system we use the level 2(5 labels)to assign importance degree(S=S) and the level 3(9 labels)to assign relevance degrees(S2=S)and 3. A multi-disciplinar recommender system to advice research complementary degrees(S,=S). Using this Lh the linguistic ources in UDL terms in each level are In this section we present a fuzzy linguistic recommender sys-.S=bo= Null=N, bi=Low=L, b2= Medium=M, b3=High tem designed using a hybrid approach and assuming a multi-gran- H, b4= Total= T: ular FLM. This system is applied to advice users on the best S=(co=Null=N, 1=Very_Low=VL, C2=Low=L, C3 research resources that could satisfy their information needs in a More_Less_Low=MLL, Ca=Medium= M. cs= More-Less High UDL. Moreover, the system recommends complementary resources that could be used by the users to meet other researchers of related areas with the aim to discover collaboration possibilities and so, to The system has three main components: resources manage- form multi-disciplinar groups In this way, it improves the services ment, user profiles management and recommendation process de users (see Fig. 2). The UDL staff manages and spreads a lot of information re sources, such as electronic books, electronic papers, electronic journals, and official dailies( Callan et aL, 2003 Renda Straccia. 3.1. Resources management 2005).Nowadays, this amount of information is growing up and This module is responsible for the management and representa- they are in need of automated tools to filter and spread that info mation to the users in a simple and timely manner. tion of the research resources. to characterize a resource. the li- a traditional search function is normally an integral part of any brary staff must insert all the available information, such as the digital library but, however, users' frustrations are increased as title, author(s), kind of resource (if it is a book, or book chapter. their needs become more complex and as the volume of informa- or a paper, or a journal, or a conference, or an official daily). journal
Herrera and Martínez (2001) demonstrated that the linguistic hierarchies are useful to represent multi-granular linguistic information and allow to combine multi-granular linguistic information without loss of information. To do this, a family of transformation functions between labels from different levels was defined: Definition 6. Let LH ¼ S tlðt; nðtÞÞ be a linguistic hierarchy whose linguistic term sets are denoted as SnðtÞ ¼ fs nðtÞ 0 ; ... ; s nðtÞ nðtÞ1g. The transformation function between a 2-tuple that belongs to level t and another 2-tuple in level t 0 –t is defined as: TFt t0 : lðt; nðtÞÞ ! lðt 0 ; nðt 0 ÞÞ; TFt t0ðs nðtÞ i ; anðtÞ Þ ¼ D D1 ðs nðtÞ i ; anðtÞ Þðnðt 0 Þ 1Þ nðtÞ 1 !: As it was pointed out in Herrera and Martínez (2001) this family of transformation functions is bijective. This result guarantees that the transformations between levels of a linguistic hierarchy are carried out without loss of information. To define the computational model, we select a level to make uniform the information (for instance, the greatest granularity level) and then we can use the operators de- fined in the 2-tuple FLM. 3. A multi-disciplinar recommender system to advice research resources in UDL In this section we present a fuzzy linguistic recommender system designed using a hybrid approach and assuming a multi-granular FLM. This system is applied to advice users on the best research resources that could satisfy their information needs in a UDL. Moreover, the system recommends complementary resources that could be used by the users to meet other researchers of related areas with the aim to discover collaboration possibilities and so, to form multi-disciplinar groups. In this way, it improves the services that a UDL could provide users. The UDL staff manages and spreads a lot of information resources, such as electronic books, electronic papers, electronic journals, and official dailies (Callan et al., 2003; Renda & Straccia, 2005). Nowadays, this amount of information is growing up and they are in need of automated tools to filter and spread that information to the users in a simple and timely manner. A traditional search function is normally an integral part of any digital library but, however, users’ frustrations are increased as their needs become more complex and as the volume of information managed by digital libraries increases. Digital libraries must move from being passive, with little adaptation to their users, to being more proactive in offering and tailoring information for individuals and communities, and in supporting community efforts to capture, structure and share knowledge (Callan et al., 2003; Gonç- alves et al., 2004; Renda & Straccia, 2005). So, the digital libraries should anticipate the users’ needs and recommend about resources that could be interesting for them. We present a hybrid recommender system that combines both the content-based and collaborative approaches (Burke, 2007; Hanani et al., 2001; Lekakos & Giaglis, 2006). The system filters the incoming information stream and delivers it to the suitable researchers according to their research areas. It recommends users research resources of their own research areas and of complementary areas. We use typical similarity functions based on threshold values to identify research resources of the own areas (Porcel et al., 2009). For example, we could use the threshold semantic functions defined in Information Retrieval to evaluate weighted queries (Bordogna & Pasi, 1993; Korfhage, 1997). On the other hand, to identify research resources of the complementary areas, we use Gaussian similarity functions (Bordogna & Pasi, 1993; Yager, 2007). To represent the linguistic information we use different label sets, i.e. the communication among the users and the system is carried out by using multi-granular linguistic information, in order to allow a higher flexibility in the communication processes of the system. Therefore, the system uses different label sets ðS1; S2; ...Þ to represent the different concepts to be assessed in its filtering activity. These label sets Si are chosen from those label sets of LH, i.e., Si 2 LH. We should point out that the number of different label sets that we can use is limited by the number of levels of LH, and therefore, in many cases the label sets Si and Sj can be associated to a same label set of LH but with different interpretations depending on the concept to be modeled. In our system, we distinguish between three concepts that can be assessed: Importance degree ðS1Þ of a discipline with respect to a resource scope or user preferences. Relevance degree ðS2Þ of a resource for a user. Complementary degree ðS3Þ between the resource scope and the user topics of interest. Following the linguistic hierarchy shown in Fig. 1, in our system we use the level 2 (5 labels) to assign importance degree ðS1 ¼ S5 Þ and the level 3 (9 labels) to assign relevance degrees ðS2 ¼ S9 Þ and complementary degrees ðS3 ¼ S9 Þ. Using this LH the linguistic terms in each level are: S 5 ¼ fb0 ¼ Null ¼ N; b1 ¼ Low ¼ L; b2 ¼ Medium ¼ M; b3 ¼ High ¼ H; b4 ¼ Total ¼ Tg; S 9 ¼ fc0 ¼ Null ¼ N; c1 ¼ Very Low ¼ VL; c2 ¼ Low ¼ L; c3 ¼ More Less Low ¼ MLL; c4 ¼ Medium ¼ M; c5 ¼ More Less High ¼ MLH; c6 ¼ High ¼ H; c7 ¼ Very High ¼ VH; c8 ¼ Total ¼ Tg. The system has three main components: resources management, user profiles management and recommendation process (see Fig. 2). 3.1. Resources management This module is responsible for the management and representation of the research resources. To characterize a resource, the library staff must insert all the available information, such as the title, author(s), kind of resource (if it is a book, or book chapter, or a paper, or a journal, or a conference, or an official daily), journal Fig. 1. Linguistic hierarchy of 3, 5 and 9 labels. C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528 12523
C Porcel et aL/ Expert Systems with Applications 36(2009)12520-12528 Informati Resources representation profiles Process Relevant resources for users Fig. 2. Structure of the system. (if it is part of a journal, the system stores the journal name ),con- s about resources( the users choose the kind of desired ference name and dates (if it is a conference). book (if it is a book i.e. if they want only books, or papers, etc )and topics chapter, the system stores the book title), official daily (if it is part of an official daily, the system stores the daily title), date, source, o the vector model Korfhage 1997)to represent text, access link to the resource and its scope. topics of interest. Then, for a user x, we have a vector We use the vector model to represent the resource scope(Korth- VUx=(VUx, VUx,., VUx5) age, 1997). Thus, to represent a resource i, we use a classification composed of 25 disciplines(see Fig 3). In each position we store where each component VUxy E SI, with y=1,., 25, stores a lin- linguistic 2-tuple value representing the importance degree for guistic 2-tuple indicating the importance degree of the discipline y with regard to the user x topics of interest. These 2-tuples values R=(R1,WR2,…VRa5) are also assigned by the library staff. The system is based on a content-based approach, but this Then, each component VR E Su, with j=1...25. indicates the proach suffers the cold-start problem to handle new items or importance degree of the discipline j with regard to the resource new users(Burke, 2007). New items cannot be recommended to i. These importance degrees are assigned by the library staff when any user until they have been rated by some one Recommenda- tions for new resources are considerably weaker than those for more widely rated resources. To overcome this problem, in our sys 3. 2. User profiles management em, as it was done in other systems(for example in Movielens), when a new user is inserted, the first action to confirm his/her reg To characterize an user, the system stores the following basic ister is to access and assess more than 15 resources of all the re- information: nickname, password (necessary to access the system). sources in the system passport number, name and surname, department and center, ad Another aspect of our system is that users can modify the dress, phone number, mobile phone and fax, web, email (elemental threshold that defines the number of recommendations that they information to send the resources and recommendations ) research want to receive So, if the system sends a lot of recommendations. group(it is a string composed of six digits, three characters indicat- the users can limit this number to N and in the future they will re- ng the research area and three numbers identifying the group), ceive only the n most relevant resources D Agriculture, animal breeding and fishing E Vegetal and animal biology and ecology Biotechnology,molecular and cellular biology and genetics E Food science and techonology D Materials science and techonology EEarth science a Social science R Computers science and techonology Energy and combustibles E Philology and philosophy hysics and space sciences a History and art Clvll engineering, transportations, construction and Industrial, mechanics, naval and aeronautic engineering E Mathematics Medicine and veterinary Environment and environmental technology a Mult-disclplinar Sclentific policy [E Psychology and education sciences E Chemisty and chemistry technology a Telecommunications, electric e electronics and Fig 3. Disciplines to define the resource scope
(if it is part of a journal, the system stores the journal name), conference name and dates (if it is a conference), book (if it is a book chapter, the system stores the book title), official daily (if it is part of an official daily, the system stores the daily title), date, source, text, access link to the resource and its scope. We use the vector model to represent the resource scope (Korfhage, 1997). Thus, to represent a resource i, we use a classification composed of 25 disciplines (see Fig. 3). In each position we store a linguistic 2-tuple value representing the importance degree for the resource scope of the discipline in that position: VRi ¼ ðVRi1; VRi2; ... ; VRi25Þ: Then, each component VRij 2 S1, with j ¼ 1 ... 25, indicates the importance degree of the discipline j with regard to the resource i. These importance degrees are assigned by the library staff when they add a new resource. 3.2. User profiles management To characterize an user, the system stores the following basic information: nickname, password (necessary to access the system), passport number, name and surname, department and center, address, phone number, mobile phone and fax, web, email (elemental information to send the resources and recommendations), research group (it is a string composed of six digits, three characters indicating the research area and three numbers identifying the group), preferences about resources (the users choose the kind of desired resources, i.e. if they want only books, or papers, etc.) and topics of interest. We use also the vector model (Korfhage, 1997) to represent the topics of interest. Then, for a user x, we have a vector: VUx ¼ ðVUx1; VUx2; ... ; VUx25Þ; where each component VUxy 2 S1, with y ¼ 1; ... ; 25, stores a linguistic 2-tuple indicating the importance degree of the discipline y with regard to the user x topics of interest. These 2-tuples values are also assigned by the library staff. The system is based on a content-based approach, but this approach suffers the cold-start problem to handle new items or new users (Burke, 2007). New items cannot be recommended to any user until they have been rated by some one. Recommendations for new resources are considerably weaker than those for more widely rated resources. To overcome this problem, in our system, as it was done in other systems (for example in Movielens), when a new user is inserted, the first action to confirm his/her register is to access and assess more than 15 resources of all the resources in the system. Another aspect of our system is that users can modify the threshold that defines the number of recommendations that they want to receive. So, if the system sends a lot of recommendations, the users can limit this number to N, and in the future they will receive only the N most relevant resources. Information sources Resources insertion process Resources representation Matching Process Relevant resources for users Complementary recommendations Feedback User profiles Users Users insertion process Resources management Recommendation process User profile management Fig. 2. Structure of the system. Fig. 3. Disciplines to define the resource scope. 12524 C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528
C Porcel et aL / Expert Systems with Applications 36(2009)12520-12528 12525 3.3. Recommendation strategy 2007). In the proposed system we use a triangular function In this phase the system generates the recommendations to de- liver the information to the fitting users. We use the following rec- g(x) 2xfor0≤x≤1/2 ommendation strategies 2-2xfor1/28(linguistic threshold va- ∑k1(4-1(nhxn)2xV∑k=1(-1(x,xma)2 lue), the user y is chosen as nearest neighbor of x. Next, the system where g is the granularity of the used term set, n is the number of account the u he resources that satisfied these users and takes into terms used to define the vectors (i.e. the number of disciplines source or not. To obtain the relevance of a resource i for the user x and(vk, M ik)is the 2 nguistic value of term k in th the system aggregates (using the arithmetic mean defined in Def or resource vector(Vi). With this similarity measure we obtain a inition 3)the o(Vx, Vy)with the assessments previously provided about i by the nearest neighbors of x To aggregate the information. linguistic value in S, to assess the similarity among the two re- we need to transform the value on(Vx, Vy) in a linguistic label in Sz sources, two users, or a resource and a user. When a new ce has been inserted into the system, the lin- using the transformation function in Definition 6. guistic similarity measure ai(Vi, V) is computed among the new guistic threshold u, then, the system recommends to the new user esource scope vector(Vi against all the stored resources in the the resource information and its calculated linguistic relevance de- system(Vij=1, ., m where m is the number of resources). If ol(Vi,V)>a(linguistic threshold value to filter out the informa- gree (label of Sz). If not, the system proceeds to estimate if the re- tion). the resource is chosen. Next, the system searches for the is a co for the new user users which were satisfied with these chosen resources(previously they have rated the resource as good )and takes into account the Then the system calculates the linguistic similarity measure user preferences(kind of resources)to consider the user or n (Vx. Vi among the user x and the resource i(for all resources ). To obtain the relevance of the resource i for a selected user x. the Then, it applies the multi-disciplinar function g(x) previously system aggregates (using the arithmetic mean defined in Defini shown( Fig 4)to the value a(Vx, Vi). If the obtained multi-discipli tion 3)the ai(Vi, V) with the assessments previously provided by tem recommends the resource as complementary. To express multi-disciplinar value as a linguistic label in S3, the transform by others users. To aggregate the information we need to trans- tion function in Definition 6 is used ormation function in definition 6 Finally, the system sends to the new users the information of all identified resources that are interesting for them, and its estimate Finally, if the calculated relevance degree is greater than a lin- linguistic complementary degree(label of S3). guistic threshold u, then, the system sends the resource inform on and its calculated linguistic relevance degree(label of S2)to the selected users. If not, the system proceeds to estimate if the re- source could be interesting as a complementary recommendation. To obtain the complementary recommendations, the system calculates the linguistic similarity measure o(Vi, vx)among the resource i and the user x( for all users). Then, it applies a multi-dis- iplinar function to the value o(Vi, V). This function must give greatest weights to similarity middle values (near 0.5 ,because values of total similarity contribute with efficient recommenda- tions but are pre own for the users. Same null values of similarity show a tionship between areas. To establish this function we can centered OWA operators in which the OWA weights are generated from a Gaussian type function (Yager, Fig. 4. Triangular function
3.3. Recommendation strategy In this phase the system generates the recommendations to deliver the information to the fitting users. We use the following recommendation strategies: When a new resource is inserted into the system, it recommends this information to the users. In this case, the system follows the content-based approach. When a new user is inserted into the system, he/she receives information about resources, previously inserted, interesting for him/her. Now, the system follows the collaborative approach. Both the processes are based on a Matching Process among the terms used in the users and resources representations (Hanani et al., 2001; Korfhage, 1997). We use the vector model (Korfhage, 1997) to represent both the resource scope and the users topics of interest. This vector model uses similarity calculations to do the matching process, such as Euclidean Distance or Cosine Measure. Exactly we use the standard cosine measure (Korfhage, 1997). However, as we have linguistic values, we need to introduce a new linguistic similarity measure: rlðV1; V2Þ ¼ D g Pn k¼1ðD1 ðv1k; av1kÞ D1 ðv2k; av2kÞÞ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn k¼1ðD1 ðv1k; av1kÞÞ2 q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn k¼1ðD1 ðv2k; av2kÞÞ2 q 0 B@ 1 CA; where g is the granularity of the used term set, n is the number of terms used to define the vectors (i.e. the number of disciplines) and ðvik; avikÞ is the 2-tuple linguistic value of term k in the user or resource vector ðViÞ. With this similarity measure we obtain a linguistic value in S1 to assess the similarity among the two resources, two users, or a resource and a user. When a new resource has been inserted into the system, the linguistic similarity measure rlðVi; VjÞ is computed among the new resource scope vector ðViÞ against all the stored resources in the system (Vj; j ¼ 1; ... ; m where m is the number of resources). If rlðVi; VjÞ P a (linguistic threshold value to filter out the information), the resource j is chosen. Next, the system searches for the users which were satisfied with these chosen resources (previously they have rated the resource as good) and takes into account the user preferences (kind of resources) to consider the user or not. To obtain the relevance of the resource i for a selected user x, the system aggregates (using the arithmetic mean defined in Definition 3) the rlðVi; VjÞ with the assessments previously provided by x about the similar resources and with the assessments provided by others users. To aggregate the information we need to transform the value rlðVi; VjÞ in a linguistic label in S2, using the transformation function in Definition 6. Finally, if the calculated relevance degree is greater than a linguistic threshold l, then, the system sends the resource information and its calculated linguistic relevance degree (label of S2) to the selected users. If not, the system proceeds to estimate if the resource could be interesting as a complementary recommendation. To obtain the complementary recommendations, the system calculates the linguistic similarity measure rlðVi; VxÞ among the resource i and the user x (for all users). Then, it applies a multi-disciplinar function to the value rlðVi; VxÞ. This function must give greatest weights to similarity middle values (near 0.5), because values of total similarity contribute with efficient recommendations but are probably known for the users. Same, null values of similarity show a null relationship between areas. To establish this function we can use the centered OWA operators in which the OWA weights are generated from a Gaussian type function (Yager, 2007). In the proposed system we use a triangular function (Fig. 4): gðxÞ ¼ 2x for 0 6 x 6 1=2; 2 2x for 1=2 < x 6 1: Next, if the obtained multi-disciplinar value is greater than that of a previously defined linguistic threshold c, the system recommends the complementary resource. To express multi-disciplinar values as a linguistic label in S3, the transformation function in Definition 6 is used. Finally, the system sends the resource information and its estimated linguistic complementary degree (label of S3) to the appropriate users. In the following, we describe the process followed when a new user is inserted into the system. A new user gives few information about the items that satisfied his/her topics of interest, so we use the collaborative approach to generate the recommendations. We follow a memory-based algorithm, which generates the recommendations according to the preferences of nearest neighbors, also known as nearest-neighbor algorithms. These algorithms present good performance as related research reported (Symeonidis, Nanopoulos, Papadopoulos, & Manolopoulos, 2008). The first step is to identify the users most similar to the new user, using a similarity function. We use the linguistic similarity measure rlðVx; VyÞ between the topics of interest vectors of the new user ðVxÞ against all users in the system (Vy; y ¼ 1; ... ; n where n is the number of users). If rlðVx; VyÞ P d (linguistic threshold value), the user y is chosen as nearest neighbor of x. Next, the system searches for the resources that satisfied these users and takes into account the user preferences (kind of resources) to consider the resource or not. To obtain the relevance of a resource i for the user x, the system aggregates (using the arithmetic mean defined in Definition 3) the rlðVx; VyÞ with the assessments previously provided about i by the nearest neighbors of x. To aggregate the information, we need to transform the value rlðVx; VyÞ in a linguistic label in S2, using the transformation function in Definition 6. Finally if the calculated relevance degree is greater than the linguistic threshold l, then, the system recommends to the new user the resource information and its calculated linguistic relevance degree (label of S2). If not, the system proceeds to estimate if the resource could be interesting as a complementary recommendation for the new user. Then the system calculates the linguistic similarity measure rlðVx; ViÞ among the user x and the resource i (for all resources). Then, it applies the multi-disciplinar function gðxÞ previously shown (Fig. 4) to the value rlðVx; ViÞ. If the obtained multi-disciplinar value is greater than the that of linguistic threshold c, the system recommends the resource as complementary. To express multi-disciplinar value as a linguistic label in S3, the transformation function in Definition 6 is used. Finally, the system sends to the new users the information of all identified resources that are interesting for them, and its estimated linguistic complementary degree (label of S3). 0 0.5 1 1g(x) Fig. 4. Triangular function. C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528 12525
C Porcel et aL/ Expert Systems with Applications 36(2009)12520-12528 3.4. Feedback phase P In this phase the recommender system recalculates and updates the recommendations of the accessed resources. This feedback Recall is calculated as the ratio of the selected relevant items to the activity is developed in the following steps elevant items, that is, it represents the probability of a relevant items being selected (1)The system recommends the user U a resource R, and then it R asks the user his/her opinion or evaluation judgements R about it Fl is a combination metric that gives equal weight to both precision (2)The user communicates the linguistic evaluation judge- and recall(Cao& Li, 2007: Sarwar et al., 2000): Sy (3) This evaluation is registered in the system for future F1=2XRX endations. The system recalculates the linguistic P endations of R by aggregating the opinions provi other users together with rcy provided by U. This can be 4.2. Experimental result done using the 2-tuple aggregation operator x given in he purpose of the experiments is to test the performance of the proposed recommender system, so we compared the recommen- 4. Experiment and evaluation dations made by the system and the recommendations made by the library staff. In this section we present the evaluation of the proposed sys- We considered a data set with 50 research resources of different tem. The main focus in evaluating the system is to determine if it areas, collected by the library staff from different information ulfills the proposed objectives, that is, the recommended informa sources. These resources were included in the system following tion is useful and interesting for the users. At the moment. we have the indications described above. We limited these experiments to plemented a trial version, in which the system works only with 10 users; all of them completed the registration process and eval- a few researchers. In a later version we will include the system in a uated 15 resources. The resources and the provided evaluations UDL. To evaluate this trial version we have designed experiments in other 20 resources that constituted the test data set. The system which the system is used to recommend research resources that filtered this 20 resources and recommends them to the suitable best satisfy the preferences of 10 users. users. Then, we compared the recommendations the systems with the recommendations provided by the library staff, and the obtained contingency table for all users is shown in 4.1. Evaluation metrics Table 5 From this contingency table, the corresponding precision, recall For the evaluation of recommender systems precision, recall and Fl are shown in Table 5.3. The average of precision, recall and and FI are measures widely used to evaluate the quality of the rec- F1 metrics is 63.52%, 67.94% and 65.05%, respectively. Fig. 5 shows ommendations( Cao& Li, 2007: Cleverdon Keen, 1966; Sarwar, a graph with the precision, recall and F1 values for each user.These Karypis, Konstan, Riedl, 2000). To calculate these metrics we values reveal a good performance of the proposed system and need a contingency table to categorize the items with respect to therefore a great satisfaction by the users. the information needs. The items are classified as both relevant or irrelevant and selected(recommended to the user)or not se- lected. The contingency table(Table 5.1)is created using these four categories Precision is defined as the ratio of the selected relevant items to Detailed experimental result. le selected items, that is, it measures the probability of a selected Precision (%6 Recall(%) Fl item being relevant: User1 66.67 571 75.00 Table 5.1 Selected Not selected Total Users Relevant User9 8000 72.73 User10 relevant 63.52 Table 5.2 Experimental contingency table. User User5
3.4. Feedback phase In this phase the recommender system recalculates and updates the recommendations of the accessed resources. This feedback activity is developed in the following steps: (1) The system recommends the user U a resource R, and then it asks the user his/her opinion or evaluation judgements about it. (2) The user communicates the linguistic evaluation judgements, rcy 2 S2. (3) This evaluation is registered in the system for future recommendations. The system recalculates the linguistic recommendations of R by aggregating the opinions provided by other users together with rcy provided by U. This can be done using the 2-tuple aggregation operator xe given in Definition 3. 4. Experiment and evaluation In this section we present the evaluation of the proposed system. The main focus in evaluating the system is to determine if it fulfills the proposed objectives, that is, the recommended information is useful and interesting for the users. At the moment, we have implemented a trial version, in which the system works only with a few researchers. In a later version we will include the system in a UDL. To evaluate this trial version we have designed experiments in which the system is used to recommend research resources that best satisfy the preferences of 10 users. 4.1. Evaluation metrics For the evaluation of recommender systems precision, recall and F1 are measures widely used to evaluate the quality of the recommendations (Cao & Li, 2007; Cleverdon & Keen, 1966; Sarwar, Karypis, Konstan, & Riedl, 2000). To calculate these metrics we need a contingency table to categorize the items with respect to the information needs. The items are classified as both relevant or irrelevant and selected (recommended to the user) or not selected. The contingency table (Table 5.1) is created using these four categories. Precision is defined as the ratio of the selected relevant items to the selected items, that is, it measures the probability of a selected item being relevant: P ¼ Nrs Ns : Recall is calculated as the ratio of the selected relevant items to the relevant items, that is, it represents the probability of a relevant items being selected: R ¼ Nrs Nr : F1 is a combination metric that gives equal weight to both precision and recall (Cao & Li, 2007; Sarwar et al., 2000): F1 ¼ 2 R P R þ P : 4.2. Experimental result The purpose of the experiments is to test the performance of the proposed recommender system, so we compared the recommendations made by the system and the recommendations made by the library staff. We considered a data set with 50 research resources of different areas, collected by the library staff from different information sources. These resources were included in the system following the indications described above. We limited these experiments to 10 users; all of them completed the registration process and evaluated 15 resources. The resources and the provided evaluations constituted our training data set. After this, we took into account other 20 resources that constituted the test data set. The system filtered this 20 resources and recommends them to the suitable users. Then, we compared the recommendations provided by the systems with the recommendations provided by the library staff, and the obtained contingency table for all users is shown in Table 5.2. From this contingency table, the corresponding precision, recall and F1 are shown in Table 5.3. The average of precision, recall and F1 metrics is 63.52%, 67.94% and 65.05%, respectively. Fig. 5 shows a graph with the precision, recall and F1 values for each user. These values reveal a good performance of the proposed system and therefore a great satisfaction by the users. Table 5.2 Experimental contingency table. User1 User2 User3 User4 User5 User6 User7 User8 User9 User10 Nrs 5431556444 Nrn 3311123212 Nis 2222332123 Nr 8742679656 Ns 7653888567 Table 5.3 Detailed experimental result. Precision (%) Recall (%) F1 (%) User1 71.43 62.50 66.67 User2 66.67 57.14 61.54 User3 60.00 75.00 66.67 User4 33.33 50.00 40.00 User5 62.50 83.33 71.43 User6 62.50 71.43 66.67 User7 75.00 66.67 70.59 User8 80.00 66.67 72.73 User9 66.67 80.00 72.73 User10 57.14 66.67 61.54 Average 63.52 67.94 65.05 Table 5.1 Contingency table. Selected Not selected Total Relevant Nrs Nrn Nr Irrelevant Nis Nin Ni Total Ns Nn N 12526 C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528
C Porcel et aL /Expert Systems with Applications 36(2009)12520-12528 Precision■ Recall OF1 9000 8000 700 6000 50.00 4000 30.00 2000 Users Fig. 5. Experimental result. 5 Conclusions Cao, Y.& Li, Y.(2007). An intelligent fuzzy-base Chang, S. L, Wang. R C,& Wang, S.Y(2007) Applying a direct multigranularity Internet access has resulted in digital libraries that are increas- nguistic and strategy-oriented aggregation approach on the assessment of ingly used by diverse communities for diverse pur and in 1015 performance. European Journal of operational Research, 177(2), ments. Users of UDL need tools to assist them in their processes of cha development as testig f enterit ibr marionsSaie te eserc 24 information gathering because of the large amount of information available on these systems. We have presented a multi-disciplinar Chen, Z& Ben-Anieh, D(2006). On the fusion of multi-granularity linguistic lab fuzzy linguistic recommender system to spread research resources in UDL The proposed system is oriented to researchers who re- Claypool, M. Gokhale, A,& Miranda, T(1999). Combining content-based and ive recommendations about resources that could be interesting for them. In particular, it is a hybrid recommender system that Cleverdon. c ow. men, E. M(1966)D e TnwaMmehn E y" Proj. incorporates complementary recommendations. The system filters the incoming information stream to spread the information to the fitting users, and when new users are inserted into the system, A, Fox, E A, Watson, L T,& Kipp, N. A (2004)Streams, structures S: A formal model for digital libraries. ACM ey receive interesting information for them to improve the ser vices that a UDL provides, it additionally recommends complemen- Good, N- Schafer, JB, Konstan J.A 1. Herlocker, J. L, et aL tary resources that allow researchers to discover collaboration possibilities with other colleagues and to form multi-disciplinar conference ce(pp.439-44 groups. The multi-granular fuzzy linguistic modeling has been ap- Hanani, U, Shapira, B.& Shoval, P(2001). Information filterit plied in order to improve the users-system interaction and the interpretability of the system activities. The experimental results Herrera, F.& Herrera-Viedma, E.(1997). Aggregatio weighted information. IEEE Transactions on Systems, Man and Cybernetics, Part A: how great user satisfaction with the received recommendations ethodology to deal with unbalanced linguistic term sets. IEEE Transactions cknowledgements Fuzzy Systems, 16(2), 354-370. n group decision making using linguistic OwA operators. Fuzzy Sets and This paper has been ed with the financing of SAINFO WEB Project(TIC00602 DER funds in FUZZYLING Project Herrera, F, Martinez, L(2000).A 2-tuple fuzzy linguistic representation model for (TN2007-61079), PETRI (PET2007-0460). and Project of computing with words. IEEE Transactions on Fuzzy Systems, 8(6). 740 Ministry of Public Works(90/07). Herrera, F.& Martinez, L(2001) A model based on linguistic 2-tuples for dealing References aking. IEEE Transactions on Systems, Man and Cybernetics. Part B: Cybernetics. 31(2)227-234. Herrera-Viedma, E.(2001a) g the retrieval process of an information Arf, B(2005). Fuzzy decision making in politics. A linguistic fuzzy-set approach Basu, C. Hirsh, H,& Cohen, w.(1998) Recom ation as classification: Us Herrera-Viedma, E (2001b). An information retrieval system with ordinal linguistic and content-based information in recommendation. In Proc. of th weighted queries based on two weighting elements. International Joumal of pp.714-720) Bordogna, G,& Pasi, G(1993) A fuzzy linguistic approach generalizing boolean Herrera-Viedma, E, Cordon, o M, Lopez, A G.& Munoz, A M.(2003). A information retrieval: A model and its evaluation. Journal of the American Society on Science. 44. 70-82. der systems. User Modeling and User-Adapted Herrera-Viedma, E,& L6pez-Herrera, A G(2007). A model of information retrieval 12(4).331-370 stem with unbalanced fuzzy linguistic information. International Journal of Burke, R(2007) Hybrid web recommender systems. In P Brusilovsky, A Kobsa, w. Herrera-viedma, E, Lopez-Herrera, A G ,& Porcel, meaton, A Beaulieu, M, Borlund, P, Brusilovsky, P Chalmers, M, et al WorkingGroupReportAvailablefrom. system model for group decision-making
5. Conclusions Internet access has resulted in digital libraries that are increasingly used by diverse communities for diverse purposes, and in which sharing and collaboration have become important social elements. Users of UDL need tools to assist them in their processes of information gathering because of the large amount of information available on these systems. We have presented a multi-disciplinar fuzzy linguistic recommender system to spread research resources in UDL. The proposed system is oriented to researchers who receive recommendations about resources that could be interesting for them. In particular, it is a hybrid recommender system that incorporates complementary recommendations. The system filters the incoming information stream to spread the information to the fitting users, and when new users are inserted into the system, they receive interesting information for them. To improve the services that a UDL provides, it additionally recommends complementary resources that allow researchers to discover collaboration possibilities with other colleagues and to form multi-disciplinar groups. The multi-granular fuzzy linguistic modeling has been applied in order to improve the users-system interaction and the interpretability of the system activities. The experimental results show great user satisfaction with the received recommendations. Acknowledgements This paper has been developed with the financing of SAINFOWEB Project (TIC00602) and FEDER funds in FUZZYLING Project (TIN2007-61079), PETRI project (PET2007-0460), and Project of Ministry of Public Works (90/07). References Arfi, B. (2005). Fuzzy decision making in politics. A linguistic fuzzy-set approach (LFSA). Political Analysis, 13(1), 23–56. Basu, C., Hirsh, H., & Cohen, W. (1998). Recommendation as classification: Using social and content-based information in recommendation. In Proc. of the fifteenth national conference on artificial intelligence (pp. 714–720). Bordogna, G., & Pasi, G. (1993). A fuzzy linguistic approach generalizing boolean information retrieval: A model and its evaluation. Journal of the American Society for Information Science, 44, 70–82. Burke, R. (2002). Hybrid recommender systems. User Modeling and User-Adapted Interaction, 12(4), 331–370. Burke, R. (2007). Hybrid web recommender systems. In P. Brusilovsky, A. Kobsa, W. Nejdl (Eds.), The Adaptive Web, LNCS (Vol. 4321) (pp. 377–408). Cabrerizo, F. J., Alonso, S., & Herrera-Viedma, E. (2009). A consensus model for group decision making problems with unbalanced fuzzy linguistic information. International Journal of Information Technology and Decision Making, 8, 1–23. Callan, J., Smeaton, A., Beaulieu, M., Borlund, P., Brusilovsky, P., Chalmers, M., et al. (2003). Personalization and recommender systems in digital libraries. Joint NSF-EU DELOS Working Group Report. Available from: . Cao, Y., & Li, Y. (2007). An intelligent fuzzy-based recommendation system for consumer electronic products. Expert Systems with Applications, 33, 230–240. Chang, S. L., Wang, R. C., & Wang, S. Y. (2007). Applying a direct multigranularity linguistic and strategy-oriented aggregation approach on the assessment of supply performance. European Journal of Operational Research, 177(2), 1013–1025. Chao, H. (2002). Assessing the quality of academic libraries on the Web: The development and testing of criteria. Library and Information Science Research, 24, 169–194. Chen, Z., & Ben-Arieh, D. (2006). On the fusion of multi-granularity linguistic label sets in group decision making. Computers and Industrial Engineering, 51(3), 526–541. Claypool, M., Gokhale, A., & Miranda, T. (1999). Combining content-based and collaborative filters in an online newpaper. In Proc. of the ACM SIGIR-99 workshop on recommender systems-implementation and evaluation. Cleverdon, C. W., & Keen, E. M. (1966). Factors determining the performance of indexing systems (Vol. 2), Test Results. ASLIB Cranfield Res. Proj., Cranfield, Bedford, England. Gonçalves, M. A., Fox, E. A., Watson, L. T., & Kipp, N. A. (2004). Streams, structures, spaces, scenarios, societies (5s): A formal model for digital libraries. ACM Transactions on Information Systems, 22(2), 270–312. Good, N., Schafer, J. B., Konstan, J. A., Borchers, A., Sarwar, B. M., Herlocker, J. L., et al. (1999). Combining collaborative filtering with personal agents for better recommendations. In Proc. of the sixteenth national conference on artificial intelligence (pp. 439–446). Hanani, U., Shapira, B., & Shoval, P. (2001). Information filtering: Overview of issues, research and systems. User Modeling and User-Adapted Interaction, 11, 203–259. Herrera, F., & Herrera-Viedma, E. (1997). Aggregation operators for linguistic weighted information. IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems, 27, 646–656. Herrera, F., Herrera-Viedma, E., & Martínez, L. (2008). A fuzzy linguistic methodology to deal with unbalanced linguistic term sets. IEEE Transactions on Fuzzy Systems, 16(2), 354–370. Herrera, F., Herrera-Viedma, E., & Verdegay, J. L. (1996). Direct approach processes in group decision making using linguistic OWA operators. Fuzzy Sets and Systems, 79, 175–190. Herrera, F., & Martínez, L. (2000). A 2-tuple fuzzy linguistic representation model for computing with words. IEEE Transactions on Fuzzy Systems, 8(6), 746– 752. Herrera, F., & Martínez, L. (2001). A model based on linguistic 2-tuples for dealing with multigranularity hierarchical linguistic contexts in multiexpert decisionmaking. IEEE Transactions on Systems, Man and Cybernetics. Part B: Cybernetics, 31(2), 227–234. Herrera-Viedma, E. (2001a). Modeling the retrieval process of an information retrieval system using an ordinal fuzzy linguistic approach. Journal of the American Society for Information Science and Technology, 52(6), 460–475. Herrera-Viedma, E. (2001b). An information retrieval system with ordinal linguistic weighted queries based on two weighting elements. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 9, 77–88. Herrera-Viedma, E., Cordón, O., Luque, M., López, A. G., & Muñoz, A. M. (2003). A model of fuzzy linguistic irs based on multi-granular linguistic information. International Journal of Approximate Reasoning, 34(3), 221–239. Herrera-Viedma, E., & López-Herrera, A. G. (2007). A model of information retrieval system with unbalanced fuzzy linguistic information. International Journal of Intelligent Systems, 22(11), 1197–1214. Herrera-Viedma, E., López-Herrera, A. G., Luque, M., & Porcel, C. (2007). A fuzzy linguistic irs model based on a 2-tuple fuzzy linguistic approach. International Journal of Uncertainty, Fuzziness and Knowledge-based Systems, 15(2), 225–250. Herrera-Viedma, E., López-Herrera, A. G., & Porcel, C. (2005). Tuning the matching function for a threshold weighting semantics in a linguistic information retrieval system. International Journal of Intelligent Systems, 20(9), 921–937. Herrera-Viedma, E., Martínez, L., Mata, F., & Chiclana, F. (2005). A consensus support system model for group decision-making problems with multi-granular Fig. 5. Experimental result. C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528 12527
2528 C Porcel et aL/ Expert Systems with Applications 36(2009)12520-12528 EE Zzy Movielens.Movierecommendations. Accessed April 200 Papadopulo Manolopoulos, Y. (2008). Mata, F, Martinez, L,& dma, E(2009). An adaptive consens Collaborative recommender syste ng effectiveness and efficiency Morales del Castillo, JM Peis, E, Moreno, J M.& Herrera-Viedma, E(in press). D- Zadeh, L A (1975). The concept of a lines. Soft Computing. 11,631-639 Yager, RR(2007). Centered oWA operat Fussion: A semantic SDI service for the research community in digital libraries. formation Research. 975). The concept of a linguistic variable and its Morales del Castillo, J. M, Pedraza-Jimenez, R, Ruiz, AA, Peis, E,& Herrera- easoning. Part Il Information Sciences, 8, 301-357: Zadeh, L A(1975) The Viedma, E(2009). A semantic model of selective dissemination of information ncept of a linguistic variable and its applications to approximate reasoning. for digital libraries. information Technology and Libraries, 28(1) 22-31 Part llL Information Sciences, 9, 43-80
linguistic preference relations. IEEE Transactions on Fuzzy Systems, 13(5), 644–658. Herrera-Viedma, E., Pasi, G., López-Herrera, A. G., & Porcel, C. (2006). Evaluating the information quality of web sites: A methodology based on fuzzy computing with words. Journal of the American Society for Information Science and Technology, 57(4), 538–549. Herrera-Viedma, E., & Peis, E. (2003). Evaluating the informative quality of documents in SGML-format using fuzzy linguistic techniques based on computing with words. Information Processing and Management, 39(2), 233–249. Hsu, M. H. (2008). A personalized English learning recommender system for ESL students. Expert Systems with Applications, 34, 683–688. Korfhage, R. R. (1997). Information Storage and Retrieval. New York: Wiley Computer Publishing. Lekakos, G., & Giaglis, G. M. (2006). Improving the prediction accuracy of recommendation algorithms: Approaches anchores on human factors. Interacting with Computers, 18, 410–431. Marchionini, G. (2009). Research and development in digital libraries. . Accessed April 2009. Mata, F., Martínez, L., & Herrera-Viedma, E. (2009). An adaptive consensus support model for group decision making problems in a multi-granular fuzzy linguistic context. IEEE Transactions on Fuzzy Systems, 17(2), 279–290. Morales del Castillo, J. M., Peis, E., Moreno, J. M., & Herrera-Viedma, E. (in press). DFussion: A semantic SDI service for the research community in digital libraries. Information Research. Morales del Castillo, J. M., Pedraza-Jiménez, R., Ruíz, A. A., Peis, E., & HerreraViedma, E. (2009). A semantic model of selective dissemination of information for digital libraries. Information Technology and Libraries, 28(1), 22–31. Movielens. Movie recommendations. . Pazzani, M. (1999). A framework for collaborative, content-based and demographic filtering. AI Review, 13(5/6), 393–408. Porcel, C., López-Herrera, A. G., & Herrera-Viedma, E. (2009). A recommender system for research resources based on fuzzy linguistic modeling. Expert Systems with Applications, 36, 5173–5183. Quiroga, L. M., & Mostafa, J. (2002). An experiment in building profiles in information filtering: the role of context of user relevance feedback. Information Processing and Management, 38, 671–694. Reisnick, P., & Varian, H. R. (1997). Recommender Systems. Special Issue of Communications of the ACM, 40(3), 56–59. Renda, M. E., & Straccia, U. (2005). A personalized collaborative digital library environment: A model and an application. Information Processing and Management, 41, 5–21. Sarwar, B., Karypis, G., Konstan, J., & Riedl, J. (2000). Analysis of recommendation algorithms for e-commerce. In Proceedings of ACM e-commerce 2000 conference (pp. 158–167). Symeonidis, P., Nanopoulos, A., Papadopoulos, A. N., & Manolopoulos, Y. (2008). Collaborative recommender systems: Combining effectiveness and efficiency. Expert Systems with Applications, 34, 2995–3013. Yager, R. R. (2007). Centered OWA operators. Soft Computing, 11, 631–639. Zadeh, L. A. (1975). The concept of a linguistic variable and its applications to approximate reasoning. Part I. Information Sciences, 8, 199–249; Zadeh, L. A. (1975). The concept of a linguistic variable and its applications to approximate reasoning. Part II. Information Sciences, 8, 301–357; Zadeh, L. A. (1975). The concept of a linguistic variable and its applications to approximate reasoning. Part III. Information Sciences, 9, 43–80. 12528 C. Porcel et al. / Expert Systems with Applications 36 (2009) 12520–12528