2010 IEEE 24th International Conference on Advanced Information Networking and Applications Workshops A Framework for Tag-Based Research Paper Recommender system: An IR Approach Pijitra Jomsri, Siripun Sanguansintukul, Ph. D Worasit Choochaiwattana Ph D Department of Mathematics Faculty of Information Technology Faculty of Science, Chulalongkorn University Dhurakij Pundit University Bangkok, Thailand Bangkok, Thailand PijitraJ@Student chula ac th, siripun. s(@chula. ac th Worasit. cha @dpu ac th Abstract-The Internet and the World wide Web provide a way find interesting items easily, quickly and efficiently for and share information in academie fields. academic papers. Thus, how to recommend research papers to Community-based research pap g systems, such as users based on tagging information becomes a central focus of CiteULike, have become popular archers. This paper the research question and the research is just start proposes a framework for a tag based research pape recommender system. The proposed approach exploits the use of These papers strive to propose a new framework for a sets of tags for recommending research papers to each user. The community-based research paper recommender system using preliminary evaluation shows that user self-defined tags could social tagging. The preliminary evaluation uses a set of t used as a profile for each individual user. This recommender recommend research paper to each user is illustrated system demonstrated an encouraging preliminary result with the The paper is structured as follows: related works are discussed overall accuracy percentage up to 91.66%. in section Il; a framework for a tag based research paper recommendation system is discussed in section Ill; section IV Keywords- recommender system, research paper, tag based presents a preliminary evaluation of the tag based paper . INTRODUCTION recommendation mechanism; finally, section V contains the conclusion and future work. The usages of Internet search engines are growing rapidly Search engines are important tools to facilitate document searches. Generally, search engines are capable to return IL RELATED WORK search results according to user queries. Recently, social This section describes background of the framework. The tagging has been widely adopted by various web services such as social bookmarking systems. These systems provide section is divided into two parts: a community-based search functions that allow users to share content with one another engine and academic social bookmarking In academia, there are many works employing search engines A. Community-Based Search Engine for research papers and investigating literature reviews such as In recent years, many studies of community-based search CiteULike [1]. Scientists, researchers and academics are able engines have been carried out.The maIn to store, organize, share and discover links to academic community-based search engines include recommendation research papers relevance feedback, personalization, and any of their Within the information retrieval community there has been combinations. The pioneering work on social recommendation considered an alternative approach for retrieving the was by Ringo [30]. Additionally, Brusilovsky studied the nformation based on community of users in the system. Many origins of personalization[34], [35] recommender systems have been designed and implemented Researchers who presented a framework for community for various types of items including newspaper, research based search engines include: Thomson [6], who introduced papers, and emails standardized cross-site personalization frameworks for the Much of the research in recommender systems have focused web: Mislove [32] suggested a Web search framework on improving the accuracy of the recommendations. Recent enhanced by social networks, and studied the mechanisms for work advises a general set of goals, such as trust, user content publishing and location in social networks,a satisfaction, and transparency. The key to achieving this framework that led to considerable improvements in general set of goals is to explain recommendations to users. effectiveness; Beydoun [40] who presented a"semantic While recommendations tell users what items they might like, annotation approach"to support search in a social network explanations reveal why they might like them [38] Researchers who studied and improved folksonomies or social Presently, there are several research works agging are: Suchanek [25], who analyzed tags and found that collaborative filtering on tags to recommend tags are "meaningful"where the tagging process is influenced works on utilizing tagging information for assisting users to y tas information to users. Unfortunately, there are only a few 978-0-7695-4019-/10526.00◎2010 Crown Copyright DOI10.1109 WAINA.2010.35
A Framework for Tag-Based Research Paper Recommender System: An IR Approach Pijitra Jomsri, Siripun Sanguansintukul,Ph.D. Department of Mathematics Faculty of Science, Chulalongkorn University Bangkok, Thailand Pijitra.J@Student.chula.ac.th, siripun.s@chula.ac.th Worasit Choochaiwattana Ph.D. Faculty of Information Technology Dhurakij Pundit University Bangkok, Thailand Worasit.cha@dpu.ac.th Abstract— The Internet and the World Wide Web provide a way to store and share information, especially in academic fields. Community-based research paper sharing systems, such as CiteULike, have become popular among researchers. This paper proposes a framework for a tag-based research paper recommender system. The proposed approach exploits the use of sets of tags for recommending research papers to each user. The preliminary evaluation shows that user self-defined tags could be used as a profile for each individual user. This recommender system demonstrated an encouraging preliminary result with the overall accuracy percentage up to 91.66%. Keywords- recommender system, research paper, tag-based I. INTRODUCTION The usages of Internet search engines are growing rapidly. Search engines are important tools to facilitate document searches. Generally, search engines are capable to return search results according to user queries. Recently, social tagging has been widely adopted by various web services such as social bookmarking systems. These systems provide functions that allow users to share content with one another. In academia, there are many works employing search engines for research papers and investigating literature reviews such as CiteULike [1]. Scientists, researchers and academics are able to store, organize, share and discover links to academic research papers. Within the information retrieval community there has been considered an alternative approach for retrieving the information based on community of users in the system. Many recommender systems have been designed and implemented for various types of items including newspaper, research papers, and emails. Much of the research in recommender systems have focused on improving the accuracy of the recommendations. Recent work advises a general set of goals, such as trust, user satisfaction, and transparency. The key to achieving this general set of goals is to explain recommendations to users. While recommendations tell users what items they might like, explanations reveal why they might like them [38]. Presently, there are several research works concerning collaborative filtering on tags to recommend personalized information to users. Unfortunately, there are only a few works on utilizing tagging information for assisting users to find interesting items easily, quickly and efficiently for academic papers. Thus, how to recommend research papers to users based on tagging information becomes a central focus of the research question and the research is just starteded. These papers strive to propose a new framework for a community-based research paper recommender system using social tagging. The preliminary evaluation uses a set of tags to recommend research paper to each user is illustrated. The paper is structured as follows: related works are discussed in section II; a framework for a tag based research paper recommendation system is discussed in section III; section IV presents a preliminary evaluation of the tag based paper recommendation mechanism; finally, section V contains the conclusion and future work. II.RELATED WORK This section describes background of the framework. The section is divided into two parts: a community-based search engine and academic social bookmarking. A. Community-Based Search Engine In recent years, many studies of community-based search engines have been carried out. The main techniques involving community-based search engines include recommendation, relevance feedback, personalization, and any of their combinations. The pioneering work on social recommendation was by Ringo [30]. Additionally, Brusilovsky studied the origins of personalization [34], [35]. Researchers who presented a framework for communitybased search engines include: Thomson [6], who introduced standardized cross-site personalization frameworks for the web; Mislove [32] suggested a Web search framework enhanced by social networks, and studied the mechanisms for content publishing and location in social networks, a framework that led to considerable improvements in effectiveness; Beydoun [40] who presented a “semantic annotation approach” to support search in a social network. Researchers who studied and improved folksonomies or social tagging are: Suchanek [25], who analyzed tags and found that tags are “meaningful” where the tagging process is influenced by tag suggestions. 2010 IEEE 24th International Conference on Advanced Information Networking and Applications Workshops 978-0-7695-4019-1/10 $26.00 © 2010 Crown Copyright DOI 10.1109/WAINA.2010.35 103
Thom-Santelli [26] explored tags for communication in directions for presenting characterizations of CiteULike and social tagging, Budura [15] presented HAMLET to promote Bibsonomy that target the management of scientific literature an efficient and precise reuse of shared metadata in highly Noel [24] looked at the tagging behavior of people who were dynamic content where tags are scarce; Li [23] used the self describing four frequently entered references. The techniques rganizing characteristics of SOM neural networks to classify from CiteULike have been applied to other academic search popular tags within the "delicious"website; Choochaiwattana such as Farooq et al. [191,[21] presenting four novel 17 examined the use of social annotations to improve the implications for designing Cite Seer quality of web searches; Gelernter [20 compared th Toine Bogers [27] employed CiteULike to generate reading information retrieval value of the cloud format tags and the tag lists for scientific articles based on the user's online reference words themselves as found in the Library Thing catalog Schmitz [10] utilized association rules mining to analyze and library. They applied three different CF algorithms and found structure folksonomies. The results can be used for ontology that user-based filtering performs the best Ms. g and supporting emergent semantics IIL. A FRAMEWORK FOR TAG-BASED RESEARCH PAPER Many researchers tried to improve the recommender RECOMMENDER SYSTEM system: Vig [38] introduced"tagsplanations", which are the explanations-based community tags. They also examined which types of tags are the most useful for tagsplanations illustrated in Figure 1. The framework is divided into a two Forsati [39] presented the HITs algorithm for part system: research paper sharing and a research paper recommender recommendation sets and showed how this method can improve the overall quality of web recommendations and other A. Research paper sharing system researchers[12],[13],[14,[22l,[29].[3land33J. A research paper sharing system provides users with new Researchers who tried to improve search results on the ways to share their research interests. They can post and research paper websites using indexes that were also applied comment on papers. They can also discover interesting papers integrating the results into e-learning resources such as [3 or Sed by other users who share the same interests. This kind to abstracts include [5], [7,[8], and 19]. There are works or ers to create their own keywords for and into research paper from libraries such as [2] [4] attaching to the posted papers. These keywords are known as s. Tags provide user-defined terms for a paper. The interesting part is that these tags can be used to create a profile B. Academic social bookmarking for each user. A research paper recommender mechanism Web searches based on social bookmarking, which let users could take advantage of this created user profile in research pecify their keywords of interest, or tags on web resources have become increasingly popular. One of the most famous (www.Citeulike.Org),aweb-basedsocialbookmarking B. A Research Paper Recommender system 1) Crawler: A research paper crawler is a small been available as a free web service since november 2004 computer program that browses directly to the research paper Like many successful software tools, CiteULike has a flexible sharing systems of the www in a predetermined manner. The filing system based on tags. Tags provide an open, quick and research paper crawler is responsible for gathering research user-defined classification model that can produce interesting paper information such as paper author, tags used, etc. This useful information helps the system to determine a user's Benefits from these innovative and time-saving scholarly interests and also helps the system to create index for each bookmarking services includes: one-click extraction of paper. Java programming is used to implement a crawler on others are reading and for sharing resources with peers, export 2) Research Paper corpus: Paper corpus is a collection of libraries in various citation formats, import existing libraries, research papers extracted from the research paper sharing access from any computer with Internet connections. It can be easily seen that CiteULike is more than just your personal 3)Indexer: TF-IDF (term frequency-inverse document research library. Additionally, it is also capable to frequency) will be used for creating indices. TF-IDF is a weight often used in information retrieval and text mining tag papers into categories This weight is a statistical measure used to evaluate how add your own comments important a word is to a document in a collection or corpus allow others to see your library The importance increases proportionally to the number of o The components of CiteULike are registered and add times a word appears in the document but is offset by the bookmarks frequency of the word in the corpus. [36] and [37] showed that Research works related to academic search engines include: research paper information like"tag, title and abstract " could Capocci [16] analyzed the small-world properties of the be a useful source for creating indices for research papers CiteULike folksonomy. Santos-Neto [28] explored three main
Thom-Santelli [26] explored tags for communication in social tagging; Budura [15] presented HAMLET to promote an efficient and precise reuse of shared metadata in highly dynamic content where tags are scarce; Li [23] used the self organizing characteristics of SOM neural networks to classify popular tags within the “delicious” website; Choochaiwattana [17] examined the use of social annotations to improve the quality of web searches; Gelernter [20] compared the information retrieval value of the cloud format tags and the tag words themselves as found in the Library Thing catalog. Schmitz [10] utilized association rules mining to analyze and structure folksonomies. The results can be used for ontology learning and supporting emergent semantics. Many researchers tried to improve the recommender system: Vig [38] introduced ‘‘tagsplanations’’, which are the explanations-based community tags. They also examined which types of tags are the most useful for tagsplanations. Forsati [39] presented the HITS algorithm for recommendation sets and showed how this method can improve the overall quality of web recommendations and other researchers [12], [13], [14], [22], [29], [31], and [33]. Researchers who tried to improve search results on the research paper websites using indexes that were also applied to abstracts include [5], [7], [8], and [9]. There are works on integrating the results into e-learning resources such as [3], and into research paper from libraries such as [2], [4]. B. Academic social bookmarking Web searches based on social bookmarking, which let users specify their keywords of interest, or tags on web resources, have become increasingly popular. One of the most famous social bookmarking websites in academia is CiteULike (www.CiteULike.org), a web-based social bookmarking service and traditional bibliographic management tool. It has been available as a free web service since November 2004. Like many successful software tools, CiteULike has a flexible filing system based on tags. Tags provide an open, quick and user-defined classification model that can produce interesting new categorizations. Benefits from these innovative and time-saving scholarly bookmarking services includes : one-click extraction of bibliographic references, tag and rate user references on what others are reading and for sharing resources with peers, export libraries in various citation formats, import existing libraries, access from any computer with Internet connections. It can be easily seen that CiteULike is more than just your personal research library. Additionally, it is also capable to: • ‘tag’ papers into categories • add your own comments on papers • allow others to see your library The components of CiteULike are registered and add bookmarks. Research works related to academic search engines include: Capocci [16] analyzed the small-world properties of the CiteULike folksonomy. Santos-Neto [28] explored three main directions for presenting characterizations of CiteULike and Bibsonomy that target the management of scientific literature. Noël [24] looked at the tagging behavior of people who were describing four frequently entered references. The techniques from CiteULike have been applied to other academic search such as Farooq et al. [18], [19], [21] presenting four novel implications for designing CiteSeer. Toine Bogers [27] employed CiteULike to generate reading lists for scientific articles based on the user’s online reference library. They applied three different CF algorithms and found that user-based filtering performs the best. III. A FRAMEWORK FOR TAG-BASED RESEARCH PAPER RECOMMENDER SYSTEM A framework for tag-based research paper search engine is illustrated in Figure 1. The framework is divided into a two part system: research paper sharing and a research paper recommender. A. Research paper sharing system A research paper sharing system provides users with new ways to share their research interests. They can post and comment on papers. They can also discover interesting papers posed by other users who share the same interests. This kind of system allows users to create their own keywords for attaching to the posted papers. These keywords are known as tags. Tags provide user-defined terms for a paper. The interesting part is that these tags can be used to create a profile for each user. A research paper recommender mechanism could take advantage of this created user profile in research paper recommendations. B. A Research Paper Recommender system 1) Crawler: A research paper crawler is a small computer program that browses directly to the research paper sharing systems of the WWW in a predetermined manner. The research paper crawler is responsible for gathering research paper information such as paper author, tags used, etc. This useful information helps the system to determine a user's interests and also helps the system to create index for each paper. Java programming is used to implement a crawler on this framework. 2) Research Paper corpus: Paper corpus is a collection of research papers extracted from the research paper sharing system. 3) Indexer: TF-IDF (term frequency–inverse document frequency) will be used for creating indices. TF-IDF is a weight often used in information retrieval and text mining. This weight is a statistical measure used to evaluate how important a word is to a document in a collection or corpus. The importance increases proportionally to the number of times a word appears in the document but is offset by the frequency of the word in the corpus. [36] and [37] showed that research paper information like “tag , title and abstract” could be a useful source for creating indices for research papers. 104
s research work has emphasis on the 6) User preference crawler and user prefer rence recommender system This crawler is responsible for crawling user preference data, meas. 9) Search Function: Cosine similarity is a similarity which detail the research papers posted by each individual urement between two vectors of n dimensions. The user including a set of personally defined tags concept is finding the cosine of the angle between two vectors 7A Profiler: A profiler is a mechanism that exploits the This measurement is often used to compare documents in text use of preference data in the recommender mechanism to mining. Given two vectors of attributes, A and B, the cosine suggest a research paper that matches with user preferences similarity, 0, is calculated by the attributes dot product divided 8)User profile: A collection of personal data related to a by the magnitude as Equation (1) specific user. A profile refers to the explicit digital similarity= cos()=.4.B representation of a person's identity. User profiles can be JAB (1) considered as the computer representation of a user model Research Paper sharing System Po Search pape World Wide Web Post a paper Server Searc Seed l Research Paper Recommender System User preference Get data crawler Crawler Indexer research paper aper User preference data Keyword Search Function QI Index Profiler Search Ranking Function User profile d Research Paper Recommender System For text matching, the attribute vectors A and B are usually delivering personalized information. The prototype of the quency vectors of the documents. The cosine system and preliminary results are presented similarity can be seen as a method of normalizing document length during comparison. systems use collaborative filtering methods. However, our 5) Ranking: The number of people bookmarking each proposed method takes advantage of a set of user defined tags paper, the number of appearance of each paper in multiple of the posted paper. The recommender function system group users, a user defined priority of the paper, and consists of these two stages date/time/year of posted paper
However, this research work has emphasis on the recommender system. 4) Search Function: Cosine similarity is a similarity measurement between two vectors of n dimensions. The concept is finding the cosine of the angle between two vectors. This measurement is often used to compare documents in text mining. Given two vectors of attributes, A and B, the cosine similarity, θ, is calculated by the attributes dot product divided by the magnitude as Equation (1). A B A B similarity . = cos(θ ) = For text matching, the attribute vectors A and B are usually term frequency vectors of the documents. The cosine similarity can be seen as a method of normalizing document length during comparison. 5) Ranking: The number of people bookmarking each paper, the number of appearance of each paper in multiple group users, a user defined priority of the paper, and date/time/year of posted paper can be used for ranking mechanism. 6) User preference crawler and user preference data: This crawler is responsible for crawling user preference data, which detail the research papers posted by each individual user including a set of personally defined tags. 7) A Profiler: A profiler is a mechanism that exploits the use of preference data in the recommender mechanism to suggest a research paper that matches with user preferences. 8) User profile: A collection of personal data related to a specific user. A profile refers to the explicit digital representation of a person's identity. User profiles can be considered as the computer representation of a user model, delivering personalized information. The prototype of the system and preliminary results are presented. 9) Recommendation mechanism: Most recommender systems use collaborative filtering methods. However, our proposed method takes advantage of a set of user defined tags of the posted paper. The recommender function system consists of these two stages: Figure 1. A Framework for Tag-Based Research Paper Recommender System Research Paper sharing System Post a paper Server Search paper World Wide Web Post a paper Search paper Group1 Group2 Community Research Paper Recommender System Get data User Query Search Seed URL Crawler research paper Indexer Search Function Ranking Keyword Result Paper Corpus Index Recommend Function User preference crawler Profiler User preference data User profile (1) 105
Analyze user's posted papers and post paper corpus contains: title ID, title name, abstract, tag of each histories to extract user preference from posting patterns paper, and link for viewing full text article, book title within 2. Recommend papers based on user which the paper was published, posted date, posted time and prefe 2) Experimental setting: The informal evaluation was The details of the recommendation mechanism are conducted. Three Ph d. students were invited as the shown in Figure 2. First step: individual profiles from users experiment participants. Their user profiles were created from are stored. The second step: cosine similarity is used to their own CiteULike accounts. User profiling is used to mode ompare research paper indexes with user profiles. The third a users features or preferences. Approaches for profiling users step, the similarity value between a user profile and the with tag vectors are used in the recommender systems to research paper index which is greater than a threshold value describe user tagging behavior. The subjects were asked to will be selected. Then, the research paper will be evaluate the research papers recommended byour recommended to the user recommendation mechanism The cosine similarity to compare a user profile and research paper index is shown in Equation (2) Paper Recommendation Mechanism sim(UT, PT)= sim(UT,P≥a Read user profile Where U is the set individual user profile. U=/up, Ur.., u/, contains all users in the system. D is the set of research papers. Profile contains all documents in the randomly research paper index Measurement the set of user self defined tags. DT. is the set of tags assigned to research papers D. PTy is the set of tags assigned randomly classification of the items Definition [User Profile For a user ui, i=1, . n, let UT, be the relationship between uis tag and item set, UT=(<uj,ut, 吨t paper to user We conduct this informal evaluatie threshold value at (-0.15. Then. the values of cosine similarity are compared with this threshold value. Note that 10 papers are randomly selected from the community that users are currently participated 3)Experimental result: The similarities between each user tag and tags from other papers in the community are Figure 2. Recommendation mechanism compared. Figure 3 shows the accuracy of the recommender system. The accuracy percentages are 75%, 100%, and 100% for users 1, 2, and 3, respectively. The overall average of IMINARY EVALUATION OF TAG BASED PAPER accuracy is 91.66% RECOMMENDATION MECHANISM I) Data Set: To evaluate the proposed research paper recommendation mechanism, the crawler collects data from CiteULike during March to May 2009. The collected ocuments consist of 62, 192 research papers. There are 103 groups that relate to computer science. Each record in the
1. Analyze user's posted papers and post histories to extract user preference from posting patterns. 2. Recommend papers based on user preferences. The details of the recommendation mechanism are shown in Figure 2. First step: individual profiles from users are stored. The second step: cosine similarity is used to compare research paper indexes with user profiles. The third step, the similarity value between a user profile and the research paper index which is greater than a threshold value will be selected. Then, the research paper will be recommended to the user. Figure 2. Recommendation mechanism IV. PRELIMINARY EVALUATION OF TAG BASED PAPER RECOMMENDATION MECHANISM 1) Data Set: To evaluate the proposed research paper recommendation mechanism, the crawler collects data from CiteULike during March to May 2009. The collected documents consist of 62,192 research papers. There are 103 groups that relate to computer science. Each record in the paper corpus contains: title ID, title name, abstract, tag of each paper, and link for viewing full text article, book title within which the paper was published, posted date, posted time and paper priority. 2) Experimental setting: The informal evaluation was conducted. Three Ph.d. students were invited as the experiment participants. Their user profiles were created from their own CiteULike accounts. User profiling is used to model a user’s features or preferences. Approaches for profiling users with tag vectors are used in the recommender systems to describe user tagging behavior. The subjects were asked to evaluate the research papers recommended by our recommendation mechanism. The cosine similarity to compare a user profile and research paper index is shown in Equation (2). ∑ ∑ ∑ = = = • • = k j k j xj ij k j i j xj ut pt ut pt sim UT PT 1 1 2 2 1 ( , ) sim(UT,PT) ≥α Where U is the set individual user profile. U= {u1 , u2 …,un }, contains all users in the system. D is the set of research papers. D= {d1 , d2 …,d m}, includes all documents in the document collection. T is the set of tag, T= { t 1 , t 2 …,t p }. P is the set of research paper where random selection P= {p1 , p2 ,…,po }, contains all documents in the randomly research papers. UTij is the set of user self defined tags. DTkj is the set of tags assigned to research papers D. PTxj is the set of tags assigned randomly to research papers. A tag is a relevant keyword assigned to one or more items by a user, describing the items and enabling classification of the items. Definition [User Profile]: For a user ui, i=1, .., n, let UTi be the relationship between ui’s tag and item set, UTi = {| utjj ∈UT, ui ∈U, and E(ui, utij, dk)=1}. We conduct this informal evaluation by setting the heuristic threshold value at α=0.15. Then, the values of cosine similarity are compared with this threshold value. Note that 10 papers are randomly selected from the community that users are currently participated. 3) Experimental result: The similarities between each user tag and tags from other papers in the community are compared. Figure 3 shows the accuracy of the recommender system. The accuracy percentages are 75%, 100%, and 100% for users 1, 2, and 3, respectively. The overall average of accuracy is 9l.66%. (2) Paper Recommendation Mechanism Yes Start cos(Θ) >= Threshold End Recommend paper to user Profile Calculate Similarity Measurement index Read user profile No (3) 106
http://www.eric.ed.gov/ericwebPortAl/cUsTom/pOrtlets/recoRddeta jsp? nfpb=true&&ERICExtSearch Search Value 0=EJ42169 Search SearchType_O-no&accnoFEJ421691 18 M. Pinto. (2008, Oct). Cyberabstracts: a portal information literacy ski ACcuracy Information Science(JIS). Volume(34), Pp.667-679. Avai http://jis.sagepub.com/cgi/content/abstract/345/667 9 M. Pinto.(2008, Oct). The role of ing to abstract Joumal of Information Science (JIS). Volume(34), 99-815.Available:http://jis.sagepub.com/cgi/content/abstract/34/6/799 Figure3. Accuracy of recommender system 10]C. Schmitz, A. Hotho, R. Jaschke, and G Stumme.(2008, Oct). Mining Association rule in Folksonomies. Journal of Information Science(JIS). 11]J Li, O Zaiane: Combining Usage, Content, and Structure Data to V. CONCLUSION workshop on Web Mining and Web Usage, (2004) This paper presents a framework for the development of a [12]P Kazienko, M Kiewra: Integration of relational databases and esearch paper recommender system. The framework portrays content for product and page recommendation, Database Engineer recommender system. The objective of this framework is to [13N Golovin, E Rahm Reinforcement leaming Architecture for Web indicate what components and what modules they have to Information, 2004) include in their systems. From this informal experiment [14]A Mathes -Retrieved: Evaluating collaborative filtering recommender setting, the recommender system illustrated encouraging ystems ACM Transactions on Information Systems, (2004) results with an overall accuracy percentage up to 91.66%. [15]T. Budura,S. Michel, P. Cudre-Mauroux, and K. Aberer, "To Tag or Therefore, tagging created by users has a potential to be used Not to tag-Harvesting Adjacent Metadata in Large-Scale Tagging Systems, as a representative of the user profile and paper in a [16]A. Capocci and G. Caldarelli, "Folksonomies and Clust recommendation. However, the number of subjects is collaborative System CiteULike", eprint ar XiV: 0710.2835, 2007. considered to be small in the experiment. In order to confirm [17]w. Choochaiwattana, and M.B. Spring, "Applying Social Annotations to the findings, more subjects are needed in the experiments In the future, a more comprehensive formal evaluation will nference on Information management and Engineering(ICIME 2009) uala Lumpur, Malaysia 3-5 April 2009 be conducted according to the proposed framework. The 8]U. Farooq, C.H. Ganoe, J.M. Carroll, and C.L. Giles, distributed scientific collaboration ations fordesigning the CiteSeer approach to an improved research paper recommender system collaborator of the Hawaii Int'l Conferenc will be determined Sciences), IEEE Compute Society, Waikoloa, Hawaii, 3-6 January 2007, 26c. 19]U. oq, T.G. Kannampallil, Y. Song, C.H. Canoe nd C. Lee Giles, "Evalating Tagging Behavior in Social Bookmarkir ACKNOWLEDgMENT stems Metrics and design heuristie ACM conference The authors would like to thank Suan Sunandha rajah land,Florida,USA,4-7November 200,pp. work(GROUP"07)Sanibel University for scholarship support. The study is not possible 201J. Gelemter, "A Quantitative Analysis of Collaborative Tags:Evaluation without the data from CiteULike Networking, Applications and Worksh Collaborate Com 2007_, 12-15 Nov 2007, New York, NY. Pp 376-381 21]C L. Giles, K. Bollacker, and S. Lawrence, "CiteSeer: An Automatic REFERENCES Libraries, ACM Press, Pittsburg, Pennsylvania, 23-26 June 1998, pp 89 sital [1 CiteULike, Access on 1/08/2009 [Online).Available http://www.citeulike.org. 22]R. J'aschke, L. B Marinho, A. Hotho, L. Schmidt-Thieme, and G 2 S Lee Giles Lawrence, C Bollacker, K NEC Res. Inst and NJ. 200 Stumme, "Tag Recommendations in Folksonomies", In Proceedings of PKDD n, "Digital libraries and auton citation indexing, " in Conf. Rec. 2007, pp. 506-51 999 IEEE Int Conf. Computer Publication, pp 67-71 23 B. LI, and Q. Zhu, "The Determination of 3 Z. Wu, Y. Mao, and H. Chen,"Subontology-BasedResource Tagging System Based on SOM Model, Second Intemational Symposium on Knowled nent for Web-Based e-Leaming."in Conf. Rec. 1999 IEEE Int Conf. Intelligent Information Technology Application 2008(IITA08),20-22 Dec 2008, Shanghai,pp.909-913 4 M. Pinto, A. Fermandez-Ramos, and A.Doucet,"Measu 24]S. Noel, and R. Beale, "Sharing vocabularies: Tag Usage in mation Literacy Skills through Abstracting: Case Study CiteULike "Proceedings of the 22nd Annual Conference of Interaction Information Science Perspective, "in Conf Rec. 2008 research libraries, pp. 132-154 1-5 September 1=Ck2m包电 [5]DH na,“ Social Tags: fornia. USA. 26-3 October 2008 "A Standard Framework for Web Personalization, "in [26 J. Thom-Santelli, and M. J. Muller, Conf. Rec. 2008 Int. Conf College research libraries, pp. 132-154 Roles: Publishers, Evangelists, Leaders 1208.H50 7 M. C. Sievert and M. J. Andrews. ( 1991, Jan). Indexing Consistency in Information Science Abstracts. " Joumal of the American Society for
0 20 40 60 80 100 user1 user2 user3 Accuracy Accuracy Figure3. Accuracy of recommender system V.CONCLUSION This paper presents a framework for the development of a research paper recommender system. The framework portrays the structural components and architecture for a research paper recommender system. The objective of this framework is to indicate what components and what modules they have to include in their systems. From this informal experiment setting, the recommender system illustrated encouraging results with an overall accuracy percentage up to 91.66%. Therefore, tagging created by users has a potential to be used as a representative of the user profile and paper in a recommendation. However, the number of subjects is considered to be small in the experiment. In order to confirm the findings, more subjects are needed in the experiments. In the future, a more comprehensive formal evaluation will be conducted according to the proposed framework. The approach to an improved research paper recommender system will be determined. ACKNOWLEDGMENT The authors would like to thank Suan Sunandha Rajabhat University for scholarship support. The study is not possible without the data from CiteULike. REFERENCES [1] CiteULike, Access on 1/08/2009 [Online]. Available: http://www.CiteULike.org. [2] S. Lee Giles Lawrence, C. Bollacker, K. NEC Res. Inst and NJ. Princeton, “Digital libraries and autonomous citation indexing,” in Conf. Rec. 1999 IEEE Int. Conf. Computer Publication, pp. 67–71. [3] Z. Wu, Y. Mao, and H. Chen, “Subontology-Based Resource Management for Web-Based e-Learning,” in Conf. Rec. 1999 IEEE Int. Conf. Knowledge and Data Engineering, pp. 67–71. [4] M. Pinto, A. Fernandez-Ramos, and A. Doucet, “Measuring Students' Information Literacy Skills through Abstracting : Case Study from a Library & Information Science Perspective,” in Conf. Rec. 2008 Int. Conf. College & research libraries , pp. 132–154. [5] D.H. Widyantoro, and J. Yen, “A fuzzy ontology-based abstract search engine and its user studies,” in Conf. Rec. 2001 IEEE Int. Conf. Fuzzy System , pp. 1291-1294. [6] L. Thomson, “A Standard Framework for Web Personalization,” in Conf. Rec. 2008 Int. Conf. College & research libraries , pp. 132–154. [7] M. C. Sievert and M. J. Andrews. (1991, Jan). Indexing Consistency in "Information Science Abstracts.". Journal of the American Society for Information Science. Volume (42), pp.1-6. Available: http://www.eric.ed.gov/ERICWebPortal/custom/portlets/recordDetails/detailm ini.jsp?_nfpb=true&_&ERICExtSearch_SearchValue_0=EJ421691&ERICExt Search_SearchType_0=no&accno=EJ421691 [8] M. Pinto. (2008, Oct). Cyberabstracts: a portal on the subject of abstracting designed to improve information literacy skills. Journal of Information Science (JIS). Volume (34), pp.667-679. Available: http://jis.sagepub.com/cgi/content/abstract/34/5/667 [9] M. Pinto. (2008, Oct). The role of information competencies and skills in learning to abstract. Journal of Information Science (JIS). Volume (34), pp. 799-815. Available: http://jis.sagepub.com/cgi/content/abstract/34/6/799 [10]C. Schmitz, A. Hotho, R. Jäschke, and G. Stumme. (2008, Oct). Mining Association rule in Folksonomies. Journal of Information Science (JIS). [11] J Li, O Zaiane : Combining Usage, Content, and Structure Data to Improve Web Site Recommendation, Proceedings of“ WebKDD-2004 workshop on Web Mining and Web Usage,(2004). [12] P Kazienko, M Kiewra : Integration of relational databases and Web site content for product and page recommendation, Database Engineering and Applications Symposium, 2004.IDEAS,(2004). [13] N Golovin, E Rahm : Reinforcement Learning Architecture for Web Recommendations, Proceedings of the International Conference on Information,(2004). [14] A Mathes - Retrieved : Evaluating collaborative filtering recommender systems ACM Transactions on Information Systems,(2004). [15] T. Budura, S. Michel, P. Cudre-Mauroux, and K. Aberer, “To Tag or Not to tag-Harvesting Adjacent Metadata in Large-Scale Tagging Systems,” SIGIS’08, Singapore, 20-24 July 2008. [16] A. Capocci and G. Caldarelli, “Folksonomies and Clustering in the Collaborative System CiteULike”, eprint arXiv: 0710.2835, 2007. [17]W. Choochaiwattana, and M.B. Spring, “Applying Social Annotations to Retrieve and Re-rank Web Resources,” Proceedings of 2009 International Conference on Information Management and Engineering (ICIME 2009), Kuala Lumpur, Malaysia 3 – 5 April 2009. [18] U. Farooq, C.H. Ganoe, J.M. Carroll, and C.L. Giles, “Supporting distributed scientific collaboration: Implications fordesigning the CiteSeer collaborator,” Proceedings of the Hawaii Int’l Conference on System Sciences), IEEE Compute Society,Waikoloa, Hawaii, 3-6 January 2007, 26c. [19] U. Farooq, T.G. Kannampallil, Y. Song ,C.H. Ganoe ,John M. Carroll ,and C. Lee Giles, “Evalating Tagging Behavior in Social Bookmarking Systems: Metrics and design heuristics,” Proceedings of the 2007 international ACM conference on Supporting group work (GROUP’07)Sanibel Island,Florida, USA,4-7 November 2007, pp.351-360. [20]J. Gelernter, “A Quantitative Analysis of Collaborative Tags: Evaluation for Information Retrieval—a Preliminary Study,” International Conference on Collaborative Computing: Networking, Applications and Worksharing, 2007. CollaborateCom 2007. ,12-15 Nov. 2007, New York, NY. pp.376-381 [21]C.L. Giles, K. Bollacker, and S. Lawrence, “CiteSeer: An Automatic Citation Indexing System,” Proceedings of the Conference on Digital Libraries,ACM Press, Pittsburg, Pennsylvania, 23-26 June 1998, pp. 89-98. [22]R. J¨aschke, L. B. Marinho, A. Hotho, L. Schmidt-Thieme, and G. Stumme, “Tag Recommendations in Folksonomies”, In Proceedings of PKDD 2007, volume 4702 of Lecture Notes in Computer Science, Springer Verlag, 2007, pp. 506–514. [23]B. LI, and Q. Zhu, “The Determination of Semantic Dimension in Social Tagging System Based on SOM Model,” Second International Symposium on Intelligent Information Technology Application 2008(IITA’08), 20-22 Dec. 2008, Shanghai, pp. 909-913 [24] S. Noel, and R. Beale, “Sharing vocabularies : Tag Usage in CiteULike,” Proceedings of the 22nd Annual Conference of Interaction, a Specialist group of the British Computer Society(HCI2008), Liverpool, UK, 1-5 September. [25] F. M. Suchanek, M. Vojnovi´c, and D. Gunawardena, “Social Tags: Meaning and Suggestions,” CIKM’08, Napa Valley, California, USA. 26–30 October 2008. [26]J. Thom-Santelli, and M. J. Muller, David R. Millen, “Social Tagging Roles: Publishers, Evangelists, Leaders”, CHI 2008, Florence, Italy, 5-10April 2008. 107
[27]T. Toine 8E Santos-Neto, M. Ripeanu, and A. lamnitchi, "Tracking usage in collaborative tagging communities [oNline).available:http://www.csee.usfedu/-anda/papers/camao7ready 29]. Haruechaiyasak, M Shyu, and S. Chen, "A Data mining Framework for Building A Web-Page Recommender System ardanand and P. Maes. Social information filtering Algd automating"word of mouth". In CHI, 1995 Advanced Information Networking and Applications Workshops (WAINA 09), Bradford, United Kingdom, 26-29 May 2009, p 654-660 32A. Mislove, K. P. Gummadi, and P. Druschel. Exploiting social networks for internet search. In HotNets. 2006 33]M. Gorgoglione, and U. Panniello, "Including Context in a Transactional Recommender System Using a Pre-filtering Approach: Two Real E-commerce 2009 Intemational Conference on Advanced Information Kingdom, 26-29 May 2009, P667-672ps(WAINA'09),Bradford,United 34]Brusilovsky, P (1996). Methods and techniques of adaptive hypermedia. 35] Brusilovsky, P(2001). Adaptive hypermedia. In A. Kobsa(Ed),User odeling and user-adapted interaction, ten year anniversary, 11, 87-110 B36 P. Jomsri S. Sanguansintukul, and W. Choochaiwattana, "Improve Research paper Searching with social tagging-A Preliminary Investigation 20-21 October 2009 37]P.JomsriSSang “A of Search Engine Using"Tag Title and Abstract "with CiteULike- An Initial ecured Transactions(ICITST-2009)London, UK, 9-12 November 2009 38R. Forsati, M.R. Meybodi, A Ghari Neiat, "Tagsplanations: Explaining Computer technology 2009, pp. 1-10. mational Conference on Electronic 39]R. Forsati, M.R. Meybod, A. Rahbar, "An efficient algorithm for web recommendation systems", Computer Systems and Applications, D09(AICCSA 2009), Rabat, Morocco, May, 2009, p. 579-586 140 G. Beydoun, R. Kultchitsky, and G. Manasseh. Evolving semantic web with social navigation. Expert Syst. App., 32(2): 265-276, 2007
[27] T. Toine Bogers, and A. van den Bosch, “Recommending Scientific Articles Using CiteULike”, RecSys’08, Switzerland, 23-25 October 2008. [28] E. Santos-Neto, M. Ripeanu, and A. Iamnitchi, “Tracking usage in collaborative tagging communities,” [Online].Available:http://www.csee.usf.edu/~anda/papers/CAMA07_ready_v 2.pdf [29]C. Haruechaiyasak, M. Shyu, and S. Chen, “A Data mining Framework for Building A Web-Page Recommender System” [30] U. Shardanand and P. Maes. Social information filtering:Algorithms for automating ”word of mouth”. In CHI, 1995. [31] K. Takano, and K. F. Li, “An Adaptive Personalized Recommender Based on Web-Browsing Behavior Learning”. 2009 International Conference on Advanced Information Networking and Applications Workshops (WAINA’09), Bradford, United Kingdom, 26-29 May 2009,p 654 – 660. [32] A. Mislove, K. P. Gummadi, and P. Druschel. Exploiting social networks for internet search. In HotNets,2006 [33] M. Gorgoglione, and U. Panniello, “Including Context in a Transactional Recommender System Using a Pre-filtering Approach: Two Real E-commerce Applications”. 2009 International Conference on Advanced Information Networking and Applications Workshops (WAINA’09), Bradford, United Kingdom, 26-29 May 2009,p 667 - 672 [34]Brusilovsky, P. (1996). Methods and techniques of adaptive hypermedia. Journal of User Modeling and User-Adaptive Interaction, 6(2-3), 87-129. [35]Brusilovsky, P. (2001). Adaptive hypermedia. In A. Kobsa (Ed.), User modeling and user-adapted interaction, ten year anniversary, 11, 87-110. [36] P. Jomsri S. Sanguansintukul, and W. Choochaiwattana, “ Improve Research paper Searching with social tagging-A Preliminary Investigation” , the Eight International Symposium on Natural Language Processing, bangkok, 20-21 October 2009. [37] P. Jomsri S. Sanguansintukul, and W. Choochaiwattana ,“A Comparison of Search Engine Using “Tag Title and Abstract” with CiteULike – An Initial Evaluation” The 4th International Conference for Internet Technology and Secured Transactions (ICITST-2009) London , UK, 9-12 November 2009 [38]R. Forsati, M.R. Meybodi, A. Ghari Neiat, “Tagsplanations: Explaining Recommendations Using Tags”, International Conference on Electronic Computer technology 2009, pp. 1-10. [39]R. Forsati, M.R. Meybodi, A. Rahbar, “An efficient algorithm for web recommendation systems”, Computer Systems and Applications, 2009(AICCSA 2009), Rabat, Morocco ,May, 2009, p. 579 – 586. [40] G. Beydoun, R. Kultchitsky, and G. Manasseh. Evolving semantic web with social navigation. Expert Syst. Appl., 32(2):265–276, 2007. 108
