Latent Dirichlet Allocation for Tag Recommendation Ralf Krestel Peter Fankhauser wolfgang Nejd 3S Research Center 3S Research Center L3S Research Center Leibniz Universitat Hannover Leibniz Universitat Hannover Leibniz Universitat Hannover Germany Germany kristel@L3S. de fankhauser L3S.de nejdI@L3S.de ABSTRACT INTRODUCTION Tagging systems have become major infrastructures on the Tagging systems (23 like Flickr, Last. fn Web. They allow users to create tags that annotate and cat- have become major infrastructures on the Web. These sys- egorize content and share them with other users, very helpful tems allow users to create and manage tags to annotate and in particular for searching multimedia content. However, as categorize content. In social tagging sy tagging is not constrained by a controlled vocabulary and he user can not only annotate his own content but also annotation guidelines, tags tend to be noisy and sparse. Es- content of others. The service offered by these systems is pecially new resources annotated by only a few users have twofold: They allow users to publish content and to search often rather idiosyncratic tags that do not reflect a common for content. Thus tagging also serves two purposes for the perspective useful for search. In this paper we introduce an pproach based on Latent Dirichlet Allocation(LDA)for recommending tags of resources in order to improve search. 1. Tags help to organize and manage own content, and Resources annotated by many users and thus equipped with a fairly stable and complete tag set are used to elicit la- tent topics to which new resources with only a few tags are Tag recommendation can focus on one of the two aspects. mapped. Based on this, other tags belonging to a topi Personalized tag recommendation helps individual users to an be recommended for the new resource. Our evaluation annotate their content in order to manage and retrieve their shows that the approach achieves significantly better preci own resources. Collective tag recommendation aims at mak sion and recall than the use of association rules, suggested ing resources more visible to other users by recommending previous work, and also recommends more specific tags tags that facilitate browsing and search Moreover, extending resources with these recommended tags However, since tags are not restricted to a certain vocabu- ignificantly improves search for new resources lary, users can pick any tags they like to describe resources Thus, these tags can be inconsistent and idiosyncratic, both Categories and subject Descriptors due to users' personal terminology as well as due to the dif- ferent purposes tags fulfill [15. This reduces the usefulness E1 [Data]: Data Structures--Graphs and networks; H 3.3 of tags in particular for resources annotated by only a few Information Storage and Retrieval: Information Search users(aka cold start problem in tagging), whereas for pop- and Retrieval-Clustering, Information filtering; 1. 2.7 Ar- ular resources collaborative tagging typically saturates at tificial Intelligence]: Natural Language Processing-Lan- some point, i.e., the rate of new descriptive tags quickly de- guage models creases with the number of users annotating a resource [18 The goal of the approach presented in this paper is to over- General terms come the cold start problem for tagging new resources. To this end, we use Latent Dirichlet Allocation(LDA) to elicit Algorithms, Experimentation, Measurement latent topics from resources with a fairly stable and complete ag set to recommend topics for new resources with only a Keywords few tags. Based on this, other tags belonging to the recom- nended topics can be recommended. Compared to an ap- social bookmarking system, delicious, tag recommendat proach using association rules, suggested previously for tag tag search recommendation, our approach achieves significantly better precision and recall. Moreover, the recommended tags are more specific for a particular resource, and thus more useful Permission to make digital or hard copies of all or part of this work for for searching and recommending resources to other users 9 personal or classroom use is granted without fee provided that copies are The remainder of this paper is organized as follows. In not made or distributed for profit or commercial advantage and that copie Section 2, we define the problem of tag recommendation bear this notice and the full citation on the first page. To copy otherwise, to more formally, and introduce the two approaches based on permission and/or a fee. RecSys www.lastfm.com Copyright2009ACM978-1-60558-435-5/09/10.510.00 ttp://delicious.comLatent Dirichlet Allocation for Tag Recommendation Ralf Krestel L3S Research Center Leibniz Universität Hannover Germany krestel@L3S.de Peter Fankhauser L3S Research Center Leibniz Universität Hannover Germany fankhauser@L3S.de Wolfgang Nejdl L3S Research Center Leibniz Universität Hannover Germany nejdl@L3S.de ABSTRACT Tagging systems have become major infrastructures on the Web. They allow users to create tags that annotate and cat￾egorize content and share them with other users, very helpful in particular for searching multimedia content. However, as tagging is not constrained by a controlled vocabulary and annotation guidelines, tags tend to be noisy and sparse. Es￾pecially new resources annotated by only a few users have often rather idiosyncratic tags that do not reflect a common perspective useful for search. In this paper we introduce an approach based on Latent Dirichlet Allocation (LDA) for recommending tags of resources in order to improve search. Resources annotated by many users and thus equipped with a fairly stable and complete tag set are used to elicit la￾tent topics to which new resources with only a few tags are mapped. Based on this, other tags belonging to a topic can be recommended for the new resource. Our evaluation shows that the approach achieves significantly better preci￾sion and recall than the use of association rules, suggested in previous work, and also recommends more specific tags. Moreover, extending resources with these recommended tags significantly improves search for new resources. Categories and Subject Descriptors E.1 [Data]: Data Structures—Graphs and networks; H.3.3 [Information Storage and Retrieval]: Information Search and Retrieval—Clustering, Information filtering; I.2.7 [Ar￾tificial Intelligence]: Natural Language Processing—Lan￾guage models General Terms Algorithms, Experimentation, Measurement Keywords social bookmarking system, delicious, tag recommendation, tag search 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. 1. INTRODUCTION Tagging systems [23] like Flickr1 , Last.fm2 or Delicious3 have become major infrastructures on the Web. These sys￾tems allow users to create and manage tags to annotate and categorize content. In social tagging systems like Delicious the user can not only annotate his own content but also content of others. The service offered by these systems is twofold: They allow users to publish content and to search for content. Thus tagging also serves two purposes for the user: 1. Tags help to organize and manage own content, and 2. Find relevant content shared by other users. Tag recommendation can focus on one of the two aspects. Personalized tag recommendation helps individual users to annotate their content in order to manage and retrieve their own resources. Collective tag recommendation aims at mak￾ing resources more visible to other users by recommending tags that facilitate browsing and search. However, since tags are not restricted to a certain vocabu￾lary, users can pick any tags they like to describe resources. Thus, these tags can be inconsistent and idiosyncratic, both due to users’ personal terminology as well as due to the dif￾ferent purposes tags fulfill [15]. This reduces the usefulness of tags in particular for resources annotated by only a few users (aka cold start problem in tagging), whereas for pop￾ular resources collaborative tagging typically saturates at some point, i.e., the rate of new descriptive tags quickly de￾creases with the number of users annotating a resource [18]. The goal of the approach presented in this paper is to over￾come the cold start problem for tagging new resources. To this end, we use Latent Dirichlet Allocation (LDA) to elicit latent topics from resources with a fairly stable and complete tag set to recommend topics for new resources with only a few tags. Based on this, other tags belonging to the recom￾mended topics can be recommended. Compared to an ap￾proach using association rules, suggested previously for tag recommendation, our approach achieves significantly better precision and recall. Moreover, the recommended tags are more specific for a particular resource, and thus more useful for searching and recommending resources to other users [9]. The remainder of this paper is organized as follows. In Section 2, we define the problem of tag recommendation more formally, and introduce the two approaches based on 1 http://www.flickr.com 2 http://www.lastfm.com 3 http://delicious.com