Expert Systems with Applications 38(2011)8488-8496 Contents lists available at Science Direct Expert Systems with Applications ELSEVIER journalhomepagewww.elsevier.com/locate/eswa Collaborative user modeling with user-generated tags for social recommender systems Heung-Nam Kim,, Abdulmajeed Alkhaldi, Abdulmotaleb El Saddik., Geun-Sik Jo KI b School of Computer and Information Engineering. Inha University, 253 Younghyun-dong. Nam-gu, Incheon 402-751. Republic of Korea College of Computer and Information Sciences, King Saud University, Riyadh, Saudi arabia ARTICLE INFO ABSTRACT With the popularity of social media services, the sheer amount of content is increasing exponentially on the Social Web that leads to attract considerable attention to recommender systems. Recommender sys tems provide users with recommendations of items suited to their needs To provide proper recommen Recommender dations to users, recommender systems require an accurate user model that can reflect a users ocial media characteristics, preferences and needs. In this study, by leveraging user-generated tags as preference indi cators, we propose a new collaborative approach to user modeling that can be exploited to recommender systems. Our approach first discovers relevant and irrelevant topics for users, and then dual user model with collaboration from other similar users. In order to evaluate the our model, we compare experimental results with a user model based on coll pproaches and a vector space model. The experimental results have shown the proposed model provides a better representation in user interests and achieves better recommendation results in terms of accuracy e 2011 Elsevier Ltd. All rights reserved. 1 Introduction best use of"word-of-mouth"recommendations(Breese, man,& Kadie, 1998: Resnick, lacovo, Suchak, Bergstorm Social media has been changing the way people find information, 1994). Although the field of CF research has a large number share knowledge and communicate with each other. This social mation filtering problems, generally a typical CF domain starts with phenomenon has transformed the masses, who were only informa- rating information that maps user-item pairs on a set of numerical models in CF can be represented by ratings given 101m山m5hm able is increasing exponentially with daily additions. In turn, it is tagging. Consequently, a user model can profit by those tags in addi ecoming increasingly more difficult for users to find the most tion to ratings. Therefore, in the last few years, a number of studies attractive content and users often struggle with a great challenge have tried to combine recommender systems with social tagging in in terms of information overload( Siersdorfer Sizov, 2009). a way that can be highly beneficial to both areas(Milicevic, Nanop- Recommender systems that have emerged in response to the oulos, Ivanovic, 2010). Such systems generate automated recom- challenge provide users with recommendations of items that they mendations just as traditional recommender systems, but retain the would like the most(Adomavicius Tuzhilin, 2005). To provide flexibility of tagging information(Sen, vig, Riedl, 2009). In point of roper recommendations to users, the systems require information combining CF with tagging, they can be regarded as a new type of that includes a users characteristics, preferences, and needs, typi- hybrid recommender systems that utilize human perceptive con- cally referred to as a User Model(Godoy Amandi, 2005). Therefore, tent contained in items. The reason is that a set of aggregated tags building an accurate model for users is crucial to the success of rec on an item is rich and compact enough to characterize and describe ommender systems One of the most successful technologies among the same main concepts of the item although tag usage of users de- recommender systems is Collaborative Filtering(CF)that makes the ends on a type of media items(.g articles, music, videos, and phe tos)they annotate( Bischoff, Firan, Nejdl, Paiu, 2008: Li, Guo, Zhao, 2008: Wetzker, Zimmermann, Bauckhage, Albayrak, 2010). orresponding author. Tel. +1 613 562 5800x6248: fax: +1 613 562 5664 In this study, we introduce a new method of building a user E-mail address: hnkimemcrlab ottawa ca(H-N. Kim). model that can represent a user's diverse preferences, and thus 0957-4174 front matter o 2011 Elsevier Ltd. All rights reserved oi:10.1016eswa201101.048Collaborative user modeling with user-generated tags for social recommender systems Heung-Nam Kim a,⇑ , Abdulmajeed Alkhaldi a , Abdulmotaleb El Saddik a,c , Geun-Sik Jo b a School of Information Technology and Engineering, University of Ottawa, 800 King Edward, Ottawa, Ontario, Canada K1N 6N5 b School of Computer and Information Engineering, Inha University, 253 Younghyun-dong, Nam-gu, Incheon 402-751, Republic of Korea c College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia article info Keywords: User modeling Personalization Recommender systems Social tagging Social media filtering abstract With the popularity of social media services, the sheer amount of content is increasing exponentially on the Social Web that leads to attract considerable attention to recommender systems. Recommender sys￾tems provide users with recommendations of items suited to their needs. To provide proper recommen￾dations to users, recommender systems require an accurate user model that can reflect a user’s characteristics, preferences and needs. In this study, by leveraging user-generated tags as preference indi￾cators, we propose a new collaborative approach to user modeling that can be exploited to recommender systems. Our approach first discovers relevant and irrelevant topics for users, and then enriches an indi￾vidual user model with collaboration from other similar users. In order to evaluate the performance of our model, we compare experimental results with a user model based on collaborative filtering approaches and a vector space model. The experimental results have shown the proposed model provides a better representation in user interests and achieves better recommendation results in terms of accuracy and ranking. 2011 Elsevier Ltd. All rights reserved. 1. Introduction Social media has been changing the way people find information, share knowledge and communicate with each other. This social phenomenon has transformed the masses, who were only informa￾tion consumers via mass media, to be producers of information. However, as rich information is shared through social media sites, the huge amount of information that has not previously been avail￾able is increasing exponentially with daily additions. In turn, it is becoming increasingly more difficult for users to find the most attractive content and users often struggle with a great challenge in terms of information overload (Siersdorfer & Sizov, 2009). Recommender systems that have emerged in response to the challenge provide users with recommendations of items that they would like the most (Adomavicius & Tuzhilin, 2005). To provide proper recommendations to users, the systems require information that includes a user’s characteristics, preferences, and needs, typi￾cally referred to as a User Model (Godoy & Amandi, 2005). Therefore, building an accurate model for users is crucial to the success of rec￾ommender systems. One of the most successful technologies among recommender systems is Collaborative Filtering (CF) that makes the best use of ‘‘word-of-mouth’’ recommendations (Breese, Hecker￾man, & Kadie, 1998; Resnick, Iacovou, Suchak, Bergstorm, & Riedl, 1994). Although the field of CF research has a large number of infor￾mation filtering problems, generally a typical CF domain starts with rating information that maps user-item pairs on a set of numerical values. Thus, user models in CF can be represented by ratings given by users on a set of items. Besides ratings, a number of modern ser￾vices also allow the users to add tags to the items, known as social tagging. Consequently, a user model can profit by those tags in addi￾tion to ratings. Therefore, in the last few years, a number of studies have tried to combine recommender systems with social tagging in a way that can be highly beneficial to both areas (Milicevic, Nanop￾oulos, & Ivanovic, 2010). Such systems generate automated recom￾mendations just as traditional recommender systems, but retain the flexibility of tagging information (Sen, Vig, & Riedl, 2009). In point of combining CF with tagging, they can be regarded as a new type of hybrid recommender systems that utilize human perceptive con￾tent contained in items. The reason is that a set of aggregated tags on an item is rich and compact enough to characterize and describe the same main concepts of the item although tag usage of users de￾pends on a type of media items (e.g., articles, music, videos, and pho￾tos) they annotate (Bischoff, Firan, Nejdl, & Paiu, 2008; Li, Guo, & Zhao, 2008; Wetzker, Zimmermann, Bauckhage, & Albayrak, 2010). In this study, we introduce a new method of building a user model that can represent a user’s diverse preferences, and thus 0957-4174/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.01.048 ⇑ Corresponding author. Tel.: +1 613 562 5800x6248; fax: +1 613 562 5664. E-mail address: hnkim@mcrlab.uottawa.ca (H.-N. Kim). Expert Systems with Applications 38 (2011) 8488–8496 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa