2009 World Congress on Computer Science and Information Engineering Semantic-enhanced case-based reasoning for intelligent recommendation Huimin Wang Guihua nie Donglin chen School of Economics, Wuhan School of Economics, Wuhan School of Economics, Wuhan University of University of University of Technology, China Technology, China Technology, China Abstract The content-based filtering techniques rely on products descriptions and generate recommendations Case-based reasoning techniques have been applied from products that are similar to those the target user recommender systems. In this paper, we have has liked in the past [2]. In content-based systems only presented a new intelligent recommendation approach the data of the current user are exploited in building a that combines semantic web techniques with case- recommendation, but often have difficulties in based recommendation techniques to improve the identifying related interests of the same user. The performance of recommender systems. The proposed collaborative filtering techniques depend on the case model integrates both content information and availability of user ratings information and generate rating information. Instec ntactic recommendations by comparing user profiles of rated techniques, case similarity between the current case items, according to similarities and differences in the and a retrieved case is measured based on semantic item ratings[3]. The collaborative filtering techniques similarity algorithm in order to understand and reuse have two serious drawbacks. One is unable to handle cases well stored in distributed case bases. The domain new items that do not have existing usage information ontologies provide a formal representation whicl entation which The other is that in order to obtain users' rating includes semantic descriptions of users and products information, a large number of previous user/system The proposed approach that considers semantic interactions are required to build reliable commendations [4] and the users preferences overcomes the limitations of the traditional recommender systems. recem meader s seemsques ave dense-ppased recommender systems are knowledge-based 1. Introduction recommender systems that exploit case-based reasoning to generate personalized recommendations With the growth of e-commerce and the present of [6]. Case-based recommenders borrow heavily from abundant information, users are confronted with more based reasoning. In principle, a recommender system he core concepts of retrieval and similarity difficult decisions about how to select products more suitable to their needs. In order to overcome the must possess knowledge about the user's requirement Information overload problem, intelligent and preferences and their relation to the offered recommender systems have become a popular field of products [7]. However, in many case-based research. Today, many e-Commerce sites are already recommender systems, only product description are ploying recommender systems to support the elr users represented as cases and recommendations are during the selection of a product that best matches their generated by retrieving those cases that are most requirements and preferences similar to a user's query that is essentially a partial The recommendation systems are to recommend recommendation reduces the user's effort and time in recommendation problem. Besides the product-based making decisions. Modern recommender systems case modeling approach, there have been a few typically employ collaborative filtering, content-based attempts to exploit more extensively the CBr filtering approaches, and hybrid approaches mixing methodology in recommender systems. In the approach both techniques 978-0-7695-3507-408525.00c2008IEE DOI10.1109CSIE2009634
Semantic-enhanced case-based reasoning for intelligent recommendation Huimin Wang Guihua Nie Donglin Chen School of Economics, Wuhan University of Technology,China School of Economics,Wuhan University of Technology,China School of Economics,Wuhan University of Technology,China Abstract Case-based reasoning techniques have been applied to recommender systems. In this paper, we have presented a new intelligent recommendation approach that combines semantic web techniques with casebased recommendation techniques to improve the performance of recommender systems. The proposed case model integrates both content information and rating information. Instead of using syntactic techniques, case similarity between the current case and a retrieved case is measured based on semantic similarity algorithm in order to understand and reuse cases well stored in distributed case bases. The domain ontologies provide a formal representation which includes semantic descriptions of users and products. The proposed approach that considers semantic information of both the products’ content descriptions and the user’s preferences overcomes the limitations of the traditional recommender systems. 1. Introduction With the growth of e-commerce and the present of abundant information, users are confronted with more difficult decisions about how to select products more suitable to their needs. In order to overcome the information overload problem, intelligent recommender systems have become a popular field of research. Today, many e-Commerce sites are already deploying recommender systems to support their users during the selection of a product that best matches their requirements and preferences. The recommendation systems are to recommend items that users may be interested in based on their predefined preferences or access histories [1]. Effective recommendation reduces the user’s effort and time in making decisions. Modern recommender systems typically employ collaborative filtering, content-based filtering approaches, and hybrid approaches mixing both techniques. The content-based filtering techniques rely on products descriptions and generate recommendations from products that are similar to those the target user has liked in the past [2]. In content-based systems only the data of the current user are exploited in building a recommendation, but often have difficulties in identifying related interests of the same user. The collaborative filtering techniques depend on the availability of user ratings information and generate recommendations by comparing user profiles of rated items, according to similarities and differences in the item ratings[3]. The collaborative filtering techniques have two serious drawbacks. One is unable to handle new items that do not have existing usage information. The other is that in order to obtain users’ rating information, a large number of previous user/system interactions are required to build reliable recommendations [4]. In recent years CBR techniques have been applied to recommender systems [4-6]. Case-based recommender systems are knowledge-based recommender systems that exploit case-based reasoning to generate personalized recommendations [6]. Case-based recommenders borrow heavily from the core concepts of retrieval and similarity in casebased reasoning. In principle, a recommender system must possess knowledge about the user’s requirement and preferences and their relation to the offered products [7]. However, in many case-based recommender systems, only product description are represented as cases and recommendations are generated by retrieving those cases that are most similar to a user’s query that is essentially a partial description of her desired product[8-9]. In this approach, the case does not store the user needs/preferences or the context that originates the recommendation problem. Besides the product-based case modeling approach, there have been a few attempts to exploit more extensively the CBR methodology in recommender systems. In the approach presented by Ricci [10], a case models a user’s 2009 World Congress on Computer Science and Information Engineering 978-0-7695-3507-4/08 $25.00 © 2008 IEEE DOI 10.1109/CSIE.2009.634 697
nteraction with the system in the recommendation n many case-based recommender systems, the session authors assume that the case base is the product All of these approaches have a common drawback catalogue, and the"problem"is the user's query that is that the recommendations are made by syntactic essentially a partial description of her desired mechanisms of both the items'content descriptions product[ 8-9]. This product-based case modeling and the users preferences. In the paper we propose a approach does not capture the link between the semantic-enhanced case-based reasoning approach in problem and the solution as in traditional CBR order to overcome the limitations of the traditional systems recommender systems. Semantic approach is required Some case-based recommender systems appear quit to understand and reuse case well stored in distributed similar to collaborative filtering based recommender system. It works on data organized around users and assets that might be considered case descriptions [11] 2 Related work The learning ability of CBr was utilized to improve the performance of collaborative filtering based 2.1 case-based reasoning recommender systems. In the approach presented by Case-based reasoning is a problem solving Ricci, a case models a user's interaction with the methodology that tries to solve new problems by system in the recommendation session [3] reusing the specific knowledge of previously All of these approaches don't consider the case experienced, concrete problem situations stored in the sharing in distributed case bases cases [ll]. A case usually denotes a problem situation a past case models a previously experienced situation 3 Semantic-enhanced case-based storing both the problem description and the solution, recommender system and a new case or unsolved case is the description of a new problem to be solved. An important feature is that Taking into account the limitations of traditional case-based reasoning is an approach to incremental, recommendation approaches we have designed a novel sustained learning since a new experience is retained semantic-enhanced case-based recommendation each time a problem has been solved, making it method. The user model structure provides knowledge immediately available for future problems [12] about the user's preferenc ces and can general case-based reasoning cycle has four main products characterized by a set of features, ratings description of a problem defines a new case that is and user's personal information. So in our approach used to retrieve similar cases from the collection of the features of a new case can be divided into three previous cases. The information and knowledge in the categories: product description given by the user retrieved case were reused to solve the initial problem. rating information of selected products and personal Through the revising process the system adapts the nformation solution to fit the specific constraint of the new In the paper, case similarity between the current roblem. Finally, useful experience is retained for case and a retrieved case was computed based on future reuse, and the case base is update semantic similarity algorithm instead of using syntactic 2. 2 case-based reasoning recommendation techniques A case-based recommender system maintains a set 3.1 Semantic-enhanced case-based of cases of previously solved recommendation recommender system structure problems and their solutions. A simple case-based The conventional case-based reasoning approaches recommender system can be described by the have their limitations handling semantics of following processes knowledge, thus decreasing possibilities for knowledge (1) When a user is looking for some product dissemination. In this paper we would like to propose which describes the desire Semantic Web approach to understand and product properties becomes as the description of a new well stored in distributed case bases. Semantic Web is a technology to add well-defined meaning to (2)The system retrieves the most similar products information on the Web to enable computers as well as from the case base according to the user query. people to understand meaning of the documents easily (3) The retrieved solution( roducts)is shown to the user or is revised to better solve the new semantic information of cases, our approach needs a formal representation which includes semantic (4) The new case is retained in case base descriptions of users and products. Ontology is a
interaction with the system in the recommendation session. All of these approaches have a common drawback that the recommendations are made by syntactic mechanisms of both the items’ content descriptions and the user’s preferences. In the paper we propose a semantic-enhanced case-based reasoning approach in order to overcome the limitations of the traditional recommender systems. Semantic approach is required to understand and reuse case well stored in distributed case bases. 2. Related work 2.1 case-based reasoning Case-based reasoning is a problem solving methodology that tries to solve new problems by reusing the specific knowledge of previously experienced, concrete problem situations stored in the cases [11]. A case usually denotes a problem situation. A past case models a previously experienced situation storing both the problem description and the solution, and a new case or unsolved case is the description of a new problem to be solved. An important feature is that case-based reasoning is an approach to incremental, sustained learning since a new experience is retained each time a problem has been solved, making it immediately available for future problems [12]. A general case-based reasoning cycle has four main processes: retrieve, reuse, revise and retain. An initial description of a problem defines a new case that is used to retrieve similar cases from the collection of previous cases. The information and knowledge in the retrieved case were reused to solve the initial problem. Through the revising process the system adapts the solution to fit the specific constraint of the new problem. Finally, useful experience is retained for future reuse, and the case base is updated. 2.2 case-based reasoning recommendation A case-based recommender system maintains a set of cases of previously solved recommendation problems and their solutions. A simple case-based recommender system can be described by the following processes: (1) When a user is looking for some product to purchase, user query which describes the desired product properties becomes as the description of a new problem. (2) The system retrieves the most similar products from the case base according to the user query. (3) The retrieved solution (some products) is shown to the user or is revised to better solve the new problem. (4) The new case is retained in case base. In many case-based recommender systems, the authors assume that the case base is the product catalogue, and the “problem” is the user's query that is essentially a partial description of her desired product[8-9]. This product-based case modeling approach does not capture the link between the problem and the solution as in traditional CBR systems. Some case-based recommender systems appear quit similar to collaborative filtering based recommender system. It works on data organized around users and assets that might be considered case descriptions [11]. The learning ability of CBR was utilized to improve the performance of collaborative filtering based recommender systems. In the approach presented by Ricci, a case models a user’s interaction with the system in the recommendation session [13]. All of these approaches don’t consider the case sharing in distributed case bases. 3. Semantic-enhanced case-based recommender system Taking into account the limitations of traditional recommendation approaches we have designed a novel semantic-enhanced case-based recommendation method. The user model structure provides knowledge about the user’s preferences and can be expressed as products characterized by a set of features, ratings referred to each product defined in the user’s profile and user’s personal information. So in our approach the features of a new case can be divided into three categories: product description given by the user, rating information of selected products and personal information. In the paper, case similarity between the current case and a retrieved case was computed based on semantic similarity algorithm instead of using syntactic techniques. 3.1 Semantic-enhanced case-based recommender system structure The conventional case-based reasoning approaches have their limitations in handling semantics of knowledge, thus decreasing possibilities for knowledge dissemination. In this paper we would like to propose Semantic Web approach to understand and reuse cases well stored in distributed case bases. Semantic Web is a technology to add well-defined meaning to information on the Web to enable computers as well as people to understand meaning of the documents easily. In order to understand and make full use of the semantic information of cases, our approach needs a formal representation which includes semantic descriptions of users and products. Ontology is a 698
formal, explicit specification ofa Given a users query, the system first integrates it conceptualization in terms of classes, properties and and users history ratings and personal preferences to relations [14]. Among the ontology knowledge, the build a new case model. Second, the system retrieves most important information to the personalized the most similar case from case bases. Third, the recommendation system is category of users and the products from the retrieved cases are shown to the category of products. Besides the category feature, user. If they can't meet user's requirement, the users and products have other features that can be retrieved solution is revised. Finally, the current ntegrated into the domain ontology to provide more recommendation case is retained in the case base. The accurate recommendation. For example, in the movie structure of the recommender system is shown in domain there are other features such as actors directors Figure 2 besides the movie category character, can be combined nto the ontology User query user’ s model 2 Product ratings Feedback Profiles 3 Personal in formation ordnet Ontology Case retrieve Ontology Bases Retrieved Case Base Case bases Case Reuse Retrieved Solution Case revise Revised solution Figure 2. Structure of semantic-enhanced case-based recommender system intuitive that objects in the same domain o 3.2 Semantic retrieval of cases domain may have some similarity within ead s The descriptions of case features in multi-case bases Similarity between two objects in the can have different semantic information. Moreover, hierarchy can be measured according to their sometime although the descriptions of new case feature correlation in the hierarchy model have different terms to the descriptions of case features In our approach, a recommendation case feature is in multi-case bases, they can have same semantic modeled as nformation. In order to overcome the problem, the Case feature =(OR,PR,Pn) paper has focused on finding semantic relevance between similar case features by semantic similarity Where, IOR stores the systems initial user query; PR measure in ontologies. The semantic relevance is stores the users product selections or the user's ratings inspired by category theory and conceptual graph. It is on selected products; PI stores users personal
formal, explicit specification of a shared conceptualization in terms of classes, properties and relations [14]. Among the ontology knowledge, the most important information to the personalized recommendation system is category of users and the category of products. Besides the category feature, users and products have other features that can be integrated into the domain ontology to provide more accurate recommendation. For example, in the movie domain there are other features such as actors, directors besides the movie category character, can be combined into the ontology. Given a user’s query, the system first integrates it and user’s history ratings and personal preferences to build a new case model. Second, the system retrieves the most similar case from case bases. Third, the products from the retrieved cases are shown to the user. If they can’t meet user’s requirement, the retrieved solution is revised. Finally, the current recommendation case is retained in the case base. The structure of the recommender system is shown in Figure 2. 1 user’s query (Product description) 2 Product ratings 3 Personal information User Profiles user’s model User Ontology Case Base Product Ontology Retrieved Case Wordnet Revised Solution Case Retrieve Case Base Ontology Bases Case Bases Case Reuse Retrieved Solution Case Revise Feedback User query 1 user’s query (Product description) 2 Product ratings 3 Personal information User Profiles user’s model User Ontology Case Base Product Ontology Retrieved Case Wordnet Revised Solution Case Retrieve Case Base Ontology Bases Case Bases Case Reuse Retrieved Solution Case Revise Feedback User query Figure 2. Structure of semantic-enhanced case-based recommender system 3.2 Semantic Retrieval of cases The descriptions of case features in multi-case bases can have different semantic information. Moreover, sometime although the descriptions of new case feature have different terms to the descriptions of case features in multi-case bases, they can have same semantic information. In order to overcome the problem, the paper has focused on finding semantic relevance between similar case features by semantic similarity measure in ontologies. The semantic relevance is inspired by category theory and conceptual graph. It is intuitive that objects in the same domain or related domain may have some similarity within each other. Similarity between two objects in the category hierarchy can be measured according to their correlation in the hierarchy model. In our approach, a recommendation case feature is modeled as: Case _ feature = (IQR, PR, PI) . Where, IQR stores the system’s initial user query; PR stores the user’s product selections or the user’s ratings on selected products; PI stores user’s personal 699
ormation, for example, the preferences on product depth(lso(cu, c,)is the global depth inthe For the retrieval and comparison of ca to check whether the cases are semantic similar in If fi represents the j th feature of new case /,and of case features in order to understand the problem of J is the th feature of the i th case rl in the case measuring semantic similarity between cases, it is first bases, sim(, fR)that denotes the semantic similarity between two terms in ontology. Given a pair similarity between two case features can be computed of terms. cl and c2. a traditional method for me by Wu Palmer formula their similarity consists of calculating the distance Finally, we need to calculate the similarity between between the nodes associated with these terms and the new case and case R/ in case base R depth of nodes in the ontology. Such a semantic similarity model was proposed by Wu and Palmer [15] ∑"xsim(/},f) m(I, RI) sim(Cu, C,)= n(G, lso(G, C2))+len(c,, Iso(cI, C2 ))+2xdepth(lso(eu, c:)) Note that len(c, c2) ontology two nput user query, personal information and ratings of Multi-case bases Ontology bases Case base Case base Case features Features Mapping Ontology m(,A eatures Mapping Computing sim(I, RI) sim(L, RI)>thresho ks When igim gure 3. The retrieval process of semantic similar cases to new case of new problem don't include in the of semantic similar cases to new case ontology and product ontology feature Figure 3 matching is finished by WordNet. The retrieval process
information, for example, the preferences on product features. For the retrieval and comparison of cases, we need to check whether the cases are semantic similar in terms of each feature instead of syntactical similarities of case features. In order to understand the problem of measuring semantic similarity between cases, it is first necessary to describe approaches to calculating the similarity between two terms in ontology. Given a pair of terms, c1 and c2, a traditional method for measuring their similarity consists of calculating the distance between the nodes associated with these terms and the depth of nodes in the ontology. Such a semantic similarity model was proposed by Wu and Palmer [15]: ( , ( , )) ( , ( , )) 2 ( ( , )) 2 ( ( , )) ( , ) 1 1 2 2 1 2 1 2 1 2 1 2 len c lso c c len c lso c c depth lso c c depth lso c c sim c c + + × × = Note that ( , ) 1 2 len c c is the shortest path in ontology between two nodes and ( ( , )) 1 2 depth lso c c is the global depth in the hierarchy. If I j f represents the j th feature of new case I , and R ij f is the j th feature of the i th case RI in the case bases, ( , ) R ij I j sim f f that denotes the semantic similarity between two case features can be computed by Wu_Palmer formula. Finally, we need to calculate the similarity between new case I and case RI in case base R . ∑ ∑ = = × = n j j n j R ij I j j w w sim f f sim I RI 1 1 ( , ) ( , ) . >threshold? Input user query, personal information and ratings of selected products as new case Case features the old case as semantic similar case WordNet Case base Case base Multi-case bases Features Mapping Customer Ontology Product Ontology Ontology bases ( , )=0? R ij I j sim f f Yes Features Mapping Computing sim(I, RI) sim(I, RI) >threshold? Input user query, personal information and ratings of selected products as new case Case features the old case as semantic similar case WordNet Case base Case base Multi-case bases Features Mapping Customer Ontology Product Ontology Ontology bases ( , )=0? R ij I j sim f f Yes Features Mapping Computing sim(I, RI) sim(I, RI) Figure 3. The retrieval process of semantic similar cases to new case When the terms of new problem don’t include in the user ontology and product ontology, case feature matching is finished by WordNet. The retrieval process of semantic similar cases to new case is shown in Figure 3. 700
4. Conclusions and future work preferences in knowledge-based recommender systems Knowledge-Based Systems, 2008, 21, Pp 305-320 In this paper, we have presented a new intelligent 3]OS Derry, avid, and S Barry. "Improving Case recommendation approach that combines semantic web Based Recommendation a Collaborative Filtering techniques with case-based recommendation Approach. In Proceedings of European Conference on Case techniques in order to overcome limitations of Based Reasoning,, 2002, pp 278-291 4Y.H. Guo, G.S. Deng, G.Q. Zhang, "Using Case-Based traditional syntactic recommendation strategies Reasoning and Social Trust to Improve the Performance of Besides, in our approach the user model structure Computing, Information and Control, 2007, pp. 484-44 tive Recommender System In E-Commerce, Innov providing knowledge about the user's preferences is expressed as products characterized by a set of [5] Q.N. Nguyen, F. Ricci, Conversational Case-Based features, ratings referred to each product and user's Recommendations Exploiting a Structured Case Model personal information. The features of cases are also ECCBR 2008, Pp. 400-414 divided into three categories: product description given [6] D. Bridge, M. Goker, L.McGinty,"Case-based by the users, rating information of selected products 2005, 20(3), pp 315-320 Knowledge Engineering Review, [7S. Armin, " Combining case-based and similarity-based integrates both content information and rating product recommendation", ECCBR 2006, pp.355-369 information and make recommendation overcome the [8]L McGinty B Smyth, "Adaptive Selection: An Analysis limitation of product-based case modeling approach of Critiquing and Preference based Feedback in and collaboration-based case modeling approach. In Conversational Recommender Systems', International the paper, instead of using syntactic techniques, case Journal of Electronic Commerce, 2006, 11(2),Pp 35-57 similarity between the current case and a retrieved case D. McSherry, " Completeness Criteria for Retrieval in is measured based on semantic similarity algorithm in Recommender Systems'", 8th European Conference on Case- order to understand and make full use of the semantic [10]F. Ricci, A. Venturini, D. Cavada, and et al, "Product information of cases stored in distributed case bases Recommendation with Interactive Query Management and representation which includes semantic descriptions of Based Reasoning, 2003, Pp 479-499 Conference on Case- User ontology and product ontology provide a formal users and products [I1 C. Hayes, P. Cunningham and B Smyth,"A Case The future work of this research includes the Based Reasoning view of Automated Collaborative search of the structure of domain ontology and case Filtering, Proceedings of 4th international conference on recommendation iving user a more accurate case-based reasoning.2001,pp 243-248 revision for [12 A. Agnar, P. Enric, Case-Based Reasoning semantic-enhance will apply case-based Foundational Issues, Methodological Variations, and System recommendation method to the real applications to test Approaches, Al Communications, 1994, 7(1), pp 39-59 3 H Shimazu, "Expertclerk: A Conversational Case-based its performance Reasoning Tool for Developing Salesclerk Agents in E Commerce Webshops", Artificial Intelligence Review, 2002 5. Acknowledgment 18(3-4),pp.223-24 [14]D. Fensel, "Ontologies: A Silver Bullet for Knowledge Management and Electronic Commerce" Springer-Verlag This paper is supported by National Natural Science New York, Inc. 2003 Fund(70572079) and Nati Science Support Plan(2006BAH02A08) [15 Z. Wu, M. Palmer, "Verb semantics and lexical selection,, In Proceedings of the 32nd Annual Meeting of the cation Ior Computational Linguistics, 1994, pp 133- 6. References n Music and User Grouping,, Journal of Information Systems, 2005, 24, pp. 1 13-132 [2B. F. Yolanda, J. P Jose, G.s. Alberto, and et al,"A flexible semantic inference methodology to reason about user
4. Conclusions and future work In this paper, we have presented a new intelligent recommendation approach that combines semantic web techniques with case-based recommendation techniques in order to overcome limitations of traditional syntactic recommendation strategies. Besides, in our approach the user model structure providing knowledge about the user’s preferences is expressed as products characterized by a set of features, ratings referred to each product and user’s personal information. The features of cases are also divided into three categories: product description given by the users, rating information of selected products and personal information. The proposed case model integrates both content information and rating information and make recommendation overcome the limitation of product-based case modeling approach and collaboration-based case modeling approach. In the paper, instead of using syntactic techniques, case similarity between the current case and a retrieved case is measured based on semantic similarity algorithm in order to understand and make full use of the semantic information of cases stored in distributed case bases. User ontology and product ontology provide a formal representation which includes semantic descriptions of users and products. The future work of this research includes the search of the structure of domain ontology and case revision for giving user a more accurate recommendation. Furthermore, we will apply our proposed semantic-enhance case-based recommendation method to the real applications to test its performance. 5. Acknowledgment This paper is supported by National Natural Science Fund(70572079) and National Science Support Plan(2006BAH02A08). 6. References [1] H.C. Chen, A.P. Chen, “A Music Recommendation System Based on Music and User Grouping”, Journal of Intelligent Information Systems, 2005, 24, pp.113–132 [2] B.F. Yolanda, J. P. Jose′, G.S. Alberto, and et al, “A flexible semantic inference methodology to reason about user preferences in knowledge-based recommender systems”, Knowledge-Based Systems, 2008, 21, pp.305–320 [3] O. S Derry, W. David, and S. Barry. “Improving CaseBased Recommendation : A Collaborative Filtering Approach. In Proceedings of European Conference on CaseBased Reasoning”, 2002, pp.278-291 [4] Y.H. Guo, G.S.Deng, G.Q. Zhang, “Using Case-Based Reasoning and Social Trust to Improve the Performance of Recommender System In E-Commerce”, Innovative Computing, Information and Control, 2007, pp. 484-484 [5] Q.N. Nguyen, F. Ricci, “Conversational Case-Based Recommendations Exploiting a Structured Case Model”, ECCBR 2008, pp.400–414 [6] D. Bridge, M. Göker, L. McGinty, “Case-based Recommender Systems”, Knowledge Engineering Review, 2005, 20(3), pp.315–320 [7] S. Armin, “Combining case-based and similarity-based product recommendation”, ECCBR 2006, pp. 355–369 [8] L. McGinty, B. Smyth, “Adaptive Selection: An Analysis of Critiquing and Preference based Feedback in Conversational Recommender Systems”, International Journal of Electronic Commerce, 2006, 11(2), pp. 35–57 [9] D. McSherry, “Completeness Criteria for Retrieval in Recommender Systems”, 8th European Conference on CaseBased Reasoning, Springer, Heidelberg, 2006, pp. 9–29. [10] F. Ricci, A. Venturini, D. Cavada, and et al, “Product Recommendation with Interactive Query Management and Twofold Similarity”, 5th International Conference on CaseBased Reasoning, 2003, pp. 479–493 [11] C. Hayes, P. Cunningham and B. Smyth, “A CaseBased Reasoning View of Automated Collaborative Filtering”, Proceedings of 4th international conference on case-based reasoning, 2001, pp.243-248 [12] A. Agnar, P. Enric, “Case-Based Reasoning: Foundational Issues, Methodological Variations, and System Approaches”, AI Communications, 1994, 7(1), pp.39-59. [13] H. Shimazu, “Expertclerk: A Conversational Case-based Reasoning Tool for Developing Salesclerk Agents in ECommerce Webshops”, Artificial Intelligence Review, 2002, 18(3-4), pp. 223–244 [14] D. Fensel, “Ontologies: A Silver Bullet for Knowledge Management and Electronic Commerce” Springer-Verlag New York, Inc. 2003. [15] Z. Wu, M. Palmer, “Verb semantics and lexical selection”, In Proceedings of the 32nd Annual Meeting of the Association for Computational Linguistics, 1994, pp.133– 138 701