TagiCoFi:Tag Informed Collaborative Filtering Yi Zhen Wu-Jun Li Dit-Yan Yeung Department of Computer Department of Computer Department of Computer Science and Engineering Science and Engineering Science and Engineering Hong Kong University of Hong Kong University of Hong Kong University of Science and Technology Science and Technology Science and Technology Hong Kong,China Hong Kong,China Hong Kong,China yzhen@cse.ust.hk liwujun@cse.ust.hk dyyeung@cse.ust.hk ABSTRACT available.Some representative examples include product Besides the rating information,an increasing number of mod- recommendation in Amazon.com 14,movie recommenda- ern recommender systems also allow the users to add per- tion in Netflix [3]and MovieLens'[16],reference recommen- sonalized tags to the items.Such tagging information may dation in CiteULike2,and bookmark recommendation in provide very useful information for item recommendation, Del.icio.us3.Existing recommender systems can be roughly because the users'interests in items can be implicitly re divided into two major categories 1.Content-based sys- flected by the tags that they often use.Although some tems 2,12,15 make use of profiles of the users or products content-based recommender systems have made preliminary to characterize their nature.On the other hand,systems attempts recently to utilize tagging information to improve based on collaborative filtering (CF)[4,9,16,17,19 do not the recommendation performance,few recommender systems exploit explicit user profiles but only past activities of the based on collaborative filtering (CF)have employed tagging users,such as their transaction history or product satisfac- information to help the item recommendation procedure. tion expressed in ratings,to predict the future activities of In this paper,we propose a novel framework,called tag the users.In recent years.CF-based systems have become informed collaborative filtering (TagiCoFi),to seamlessly in- more and more popular than content-based systems because tegrate tagging information into the CF procedure.Experi- it is much easier to collect the past activities of users than mental results demonstrate that TagiCoFi outperforms its their profiles due to privacy considerations. counterpart which discards the tagging information even In recent years,besides the ratings on the items given when it is available,and achieves state-of-the-art perfor- by the users.an increasing number of modern recommender systems also allow the users to add personalized tags4,in mance. the form of words or phrases,to the items.For example, users may add tags to movies in MovieLens,to web sites in Categories and Subject Descriptors Del.icio.us and to references in CiteULike.Such tagging in- H.3 Information Storage and Retrieval):Information formation may provide very useful information for item rec- Search and Retrieval-Information Filtering:H.2 [Database ommendation,because the users'interests in items can be Management:Database Application-Data Mining implicitly reflected by the tags that they often use 21.For example,if two users often use the tags "Oscar"and "Tom General Terms Hanks”,both of them may like the movie“Forrest Gump'” In fact,the effectiveness of tags in representing users'prefer- Algorithms ence or interests has been validated by Zanardi et al.in the CiteULike dataset [27.Very recently,some content-based Keywords systems,such as those in 6,22,23,have made some pre- Collaborative filtering,recommender systems,tag liminary attempts to utilize tagging information to improve the recommendation performance.However,there has been little work on improving CF-based systems with the help of 1.INTRODUCTION tagging information.Because CF-based systems have be- Since the amount of information on the Web is increasing come more popular than content-based systems,it would be at an astonishing rate that is much faster than our ability a very worthwhile endeavor to devise novel CF techniques to process it,recommendation plays a more and more im- which can also utilize tagging information for item recom- portant role for us to make effective use of the information mendation Existing CF methods can be divided into two main cat- http://movielens.umn.edu/ Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are http://www.citeulike.org/ not made or distributed for profit or commercial advantage and that copies http://delicious.com/ bear this notice and the full citation on the first page.To copy otherwise,to AIt should be emphasized that the setting in this paper is republish,to post on servers or to redistribute to lists,requires prior specific different from those about tag recommendation 7,24 in permission and or a tee. which the recommended objects are tags.The recommended RecSys'09.October 23-25.2009.New York.New York.USA. objects in this paper are called items,whereas tags are other Copyright2009ACM978-1-60558-435-5/09/10.$10.00. objects about the items added by users.TagiCoFi: Tag Informed Collaborative Filtering Yi Zhen Department of Computer Science and Engineering Hong Kong University of Science and Technology Hong Kong, China yzhen@cse.ust.hk Wu-Jun Li Department of Computer Science and Engineering Hong Kong University of Science and Technology Hong Kong, China liwujun@cse.ust.hk Dit-Yan Yeung Department of Computer Science and Engineering Hong Kong University of Science and Technology Hong Kong, China dyyeung@cse.ust.hk ABSTRACT Besides the rating information, an increasing number of mod￾ern recommender systems also allow the users to add per￾sonalized tags to the items. Such tagging information may provide very useful information for item recommendation, because the users’ interests in items can be implicitly re- flected by the tags that they often use. Although some content-based recommender systems have made preliminary attempts recently to utilize tagging information to improve the recommendation performance, few recommender systems based on collaborative filtering (CF) have employed tagging information to help the item recommendation procedure. In this paper, we propose a novel framework, called tag informed collaborative filtering (TagiCoFi), to seamlessly in￾tegrate tagging information into the CF procedure. Experi￾mental results demonstrate that TagiCoFi outperforms its counterpart which discards the tagging information even when it is available, and achieves state-of-the-art perfor￾mance. Categories and Subject Descriptors H.3 [Information Storage and Retrieval]: Information Search and Retrieval—Information Filtering; H.2 [Database Management]: Database Application—Data Mining General Terms Algorithms Keywords Collaborative filtering, recommender systems, tag 1. INTRODUCTION Since the amount of information on the Web is increasing at an astonishing rate that is much faster than our ability to process it, recommendation plays a more and more im￾portant role for us to make effective use of the information Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. RecSys’09, October 23–25, 2009, New York, New York, USA. Copyright 2009 ACM 978-1-60558-435-5/09/10 ...$10.00. available. Some representative examples include product recommendation in Amazon.com [14], movie recommenda￾tion in Netflix [3] and MovieLens1 [16], reference recommen￾dation in CiteULike2 , and bookmark recommendation in Del.icio.us3 . Existing recommender systems can be roughly divided into two major categories [1]. Content-based sys￾tems [2, 12, 15] make use of profiles of the users or products to characterize their nature. On the other hand, systems based on collaborative filtering (CF) [4, 9, 16, 17, 19] do not exploit explicit user profiles but only past activities of the users, such as their transaction history or product satisfac￾tion expressed in ratings, to predict the future activities of the users. In recent years, CF-based systems have become more and more popular than content-based systems because it is much easier to collect the past activities of users than their profiles due to privacy considerations. In recent years, besides the ratings on the items given by the users, an increasing number of modern recommender systems also allow the users to add personalized tags4 , in the form of words or phrases, to the items. For example, users may add tags to movies in MovieLens, to web sites in Del.icio.us and to references in CiteULike. Such tagging in￾formation may provide very useful information for item rec￾ommendation, because the users’ interests in items can be implicitly reflected by the tags that they often use [21]. For example, if two users often use the tags “Oscar” and “Tom Hanks”, both of them may like the movie “Forrest Gump”. In fact, the effectiveness of tags in representing users’ prefer￾ence or interests has been validated by Zanardi et al. in the CiteULike dataset [27]. Very recently, some content-based systems, such as those in [6, 22, 23], have made some pre￾liminary attempts to utilize tagging information to improve the recommendation performance. However, there has been little work on improving CF-based systems with the help of tagging information. Because CF-based systems have be￾come more popular than content-based systems, it would be a very worthwhile endeavor to devise novel CF techniques which can also utilize tagging information for item recom￾mendation. Existing CF methods can be divided into two main cat- 1 http://movielens.umn.edu/ 2 http://www.citeulike.org/ 3 http://delicious.com/ 4 It should be emphasized that the setting in this paper is different from those about tag recommendation [7, 24] in which the recommended objects are tags. The recommended objects in this paper are called items, whereas tags are other objects about the items added by users