The task-dependent effect of tags and ratings on social media access

Abstract

Recently, online social networks have emerged that allow people to share their multimedia files, retrieve interesting content, and discover like-minded people. These systems often provide the possibility to annotate the content with tags and ratings.Using a random walk through the social annotation graph, we have combined these annotations into a retrieval model that effectively balances the personal preferences and opinions of like-minded users into a single relevance ranking for either content, tags, or people. We use this model to identify the influence of different annotation methods and system design aspects on common ranking tasks in social content systems.Our results show that a combination of rating and tagging information can improve tasks like search and recommendation. The optimal influence of both sources on the ranking is highly dependent on the retrieval task and system design. Results on content search and tag suggestion indicate that the profile created by a user's annotations can be used effectively to adapt the ranking to personal preferences. The random walk reduces sparsity problems by smoothly integrating indirectly related concepts in the relevance ranking, which is especially valuable for cold-start users or individual tagging systems like YouTube and Flickr.

Assessing the quality of textual features in social media

“…Indeed, many authors have proposed the use of tags to enhance services such as searching, recommendation, clustering and indexing (Hotho, Jaschke, Schmitz, & Stumme, 2006;Byde et al, 2007;Li et al, 2008;Schenkel et al, 2008;Song et al, 2008;Sigurbjornsson & van Zwol, 2008;Clements et al, 2010;Guy et al, 2010). Others have addressed the characterization of tagging systems (Paul et al, 2010;Golder & Huberman, 2006;Santos-Neto et al, 2010).…”

“…In Belém et al (2011) we extended traditional tag co-occurrence based methods to include not only tags that had been previously assigned to the objects but also terms extracted from other textual features, applying several heuristic metrics to capture the relevance of each candidate term as a recommendation for the target object. Following different approaches, Clements et al (2010) exploited random walks on a social annotation graph combining content, tags and users, whereas Rendle and Schmidt-Thie (2010) proposed a personalized tag recommendation method that exploits the factorization of tag assignment events into matrices modeling the interactions among users, objects and tags. The relationships among people, tags and objects were also investigated by Guy et al (2010): the authors proposed to recommend objects that are strongly related to people in a user's social network as well as objects related to the user's tags.…”

“…The goal of this task is to support users during the process of tagging Web 2.0 objects, by recommending ''good'' (i.e., high quality) tags for describing an object (Guan et al, 2009;Song et al, 2008;Sigurbjornsson & van Zwol, 2008;Byde et al, 2007;Lipczak et al, 2009;Rendle & Schmidt-Thie, 2010;Belém et al, 2011;Menezes et al, 2010;Clements et al, 2010). We here focus on recommending tags because: (1) it is the most studied textual feature on the Web 2.0, being of more interest to the community; (2) our characterization and classification results indicate that this is a feature of quality (with respect to several aspects) when used in isolation; (3) recommending new terms to other features does not make much sense as they are typically composed of full sentences with semantic meanings, while the TAGS feature contains set of terms; and (4) the recommended tags can be, at least partially, automatically evaluated, as we shall see.…”

A comparative study of heterogeneous item recommendations in social systems

“…Finally, we removed those users with less than 20 user contacts in the user set, updating the social network by removing relations with contacts that no longer belonged to the final user set, and users who had no relations. The previous threshold (20) was established analysing the user contact histogram, so as to avoid the long tail users. As shown in Table 1, the final dataset contains 1.9K users, 69.2K bookmarked Web pages, 437.6K tag assignments, and 15.3K friend relations.…”

New grand challenge for multimedia information retrieval: bridging the utility gap

“…Another critical bottleneck results from still insufficient insights on how to handle multi-modal information encoded in a typical social graph model. Addressing this bottleneck would typically involve research on cross-modal analysis, normalization and fusion of information [8] [57]. Finally, MIR should rely more strongly on recent works in the domain of recommender systems, and in particular on the works on trust-aware recommendation (e.g.…”