Recommending Scientific Literatures in a Collaborative Tagging Environment* Ping Yin, Ming Zhang and Xiaoming li School of Electronics Engineering and Computer Science Peking University, Beijing, China pkufrankyegmail. com, mzhangenet. pku. edu. cn, lxm@pku. edu.cn Abstract. Recently, collaborative tagging has become popular in the web2.0 orld. Tags can be helpful if used for the recommendation since they reflect haracteristic content features of the resources. However there are few re- searches which introduce tags into the recommendation. This paper proposes a tag-based recommendation framework for scientific literatures which models the user interests with tags and literature keywords. A hybrid recommendation orithm is then applied which is similar to the user-user collaborative filtering algorithm except that the user similarity is measured based on the vector model of user keywords other than the rating matrix, and that the rating is not from the user but represented as user-item similarity computed with the dot-product based similarity instead of the cosine-based similarity. Experiments show that our tag-based algorithm is better than the baseline algorithm and the extension of user model and dot-product-based similarity computation are also helpful 1 Introduction Collaborative recommendation and content-based recommendation are widely used in recommendation systems. Due to advantages and flaws of both technologies, it's a hot research to combine them to achieve better results [1, 2 In recent years, collaborative tagging [3] becomes more and more popular. Tags can reflect both users opinion and content features of resources. The utilization of the tag content for recommendation is worthy of a further research This paper focuses on scientific literature recommendation in a collaborative e envI- onment, considering both collaborative tags and content information. There is much work related to ours. Digital libraries such as ACM list similar pa pers in the form of text search. Cite Seer provides content-based and citation-based recommendations. McNee etc. generate recommendations by mapping the web of citations between papers into the C user-item rating matrix [4, 5 The remainder of the paper is organized as follows. Section 2 describes the ke teps for scientific literature recommendation in a collaborative tagging environment Section 3 experimentally evaluates the algorithm. Section 4 summarizes this paper This work is supported by the National Natural Science Foundation of China under Grant No abs China under""On line course organ 2http:/www.acm.org/dl http://citeseer.ist.psu.edu D. H.-L. Goh et al.(Eds ) ICADL 2007, LNCS 4822, Pp. 478-481, 2007. o Springer-Verlag Berlin Heidelberg 200D.H.-L. Goh et al. (Eds.): ICADL 2007, LNCS 4822, pp. 478–481, 2007. © Springer-Verlag Berlin Heidelberg 2007 Recommending Scientific Literatures in a Collaborative Tagging Environment* Ping Yin, Ming Zhang, and Xiaoming Li School of Electronics Engineering and Computer Science Peking University, Beijing, China pkufranky@gmail.com, mzhang@net.pku.edu.cn, lxm@pku.edu.cn Abstract. Recently, collaborative tagging has become popular in the web2.0 world. Tags can be helpful if used for the recommendation since they reflect characteristic content features of the resources. However, there are few re￾searches which introduce tags into the recommendation. This paper proposes a tag-based recommendation framework for scientific literatures which models the user interests with tags and literature keywords. A hybrid recommendation algorithm is then applied which is similar to the user-user collaborative filtering algorithm except that the user similarity is measured based on the vector model of user keywords other than the rating matrix, and that the rating is not from the user but represented as user-item similarity computed with the dot-product￾based similarity instead of the cosine-based similarity. Experiments show that our tag-based algorithm is better than the baseline algorithm and the extension of user model and dot-product-based similarity computation are also helpful. 1 Introduction Collaborative recommendation and content-based recommendation are widely used in recommendation systems. Due to advantages and flaws of both technologies, it’s a hot research to combine them to achieve better results [1, 2]. In recent years, collaborative tagging [3] becomes more and more popular. Tags can reflect both user’s opinion and content features of resources. The utilization of the tag content for recommendation is worthy of a further research. This paper focuses on scientific literature recommendation in a collaborative envi￾ronment, considering both collaborative tags and content information. There is much work related to ours. Digital libraries such as ACM1 list similar pa￾pers in the form of text search. CiteSeer2 provides content-based and citation-based recommendations. McNee etc. generate recommendations by mapping the web of citations between papers into the CF user-item rating matrix [4, 5]. The remainder of the paper is organized as follows. Section 2 describes the key steps for scientific literature recommendation in a collaborative tagging environment. Section 3 experimentally evaluates the algorithm. Section 4 summarizes this paper. * This work is supported by the National Natural Science Foundation of China under Grant No. 90412010, HP Labs China under “On line course organization”. 1 http://www.acm.org/dl 2 http://citeseer.ist.psu.edu