Improving Graph-based Approaches for Personalized Tag Recommendation

Abstract

Social tagging applications allow users to annotate online resources, resulting in a complex network of interrelated users, resources and tags often called a Folksonomy. A folk- sonomy is often represented as a hyper-graph in which each hyper-edge connects a user, resource and tag. This tripartite hyper-graph is often used by data mining applications to provide services for the user such as tag recommenders. This paper provides an overview on the state of the art of graph-based tag recommendation from a critical perspective. In addition, we suggest improving the existing graph-based tag recommendation techniques by introducing a new model of the folksonomy as a directed graph. I. INTRODUCTION Collaborative tagging systems such as Delicious 1 , lastFm 2 , and Bibsonomy 3 have emerged as powerful applications for Internet users. Tagging systems support users with several benefits. First they allow users to organize their own data with a level of freedom not possible in traditional taxonomic filing systems. Secondly they provide users with the means to openly share this information among friends and colleagues. Thirdly they also allow anyone to utilize the collective knowledge of others for discovering new topics, resources or perhaps even new friends. While social tagging systems have many benefits, they also present several challenges. Most tagging applications permit unsupervised tagging; users are free to use any tag they wish to describe a resource. This unsupervised tagging can result in tag redundancy - in which several tags have the same meaning - or tag ambiguity - in which a single tag has many different meanings. Such inconsistencies can confound users as they attempt to utilize the folksonomy. It can be difficult for users to traverse the sheer volume of data. Moreover, noise in the data can impede the users experience. Data mining applications such as tag recommenders make it easier for the user to navigate the system. Tag recommendation, the suggestion of tags during the annotation process reduces the user effort. By reducing the effort users are encouraged to tag more frequently, apply more tags to an individual resource, reuse common tags, and perhaps use tags the user had not previously considered. Moreover, user error is reduced by eliminating redundant tags caused by capitalization inconsistencies, punctuation errors, 1http://delicious.com/ 2 http://www.last.fm/ 3 www.bibsonomy.org/ misspellings and other discrepancies. The tag recommender can further promote a core tag vocabulary steering the user toward adopting certain tags while not imposing any strict rules. The tag recommender may even avoid ambiguous tags in favor of tags that offer greater information value. This may aid other users when navigating through the folksonomy to find interesting resources related to a tag which is more often used by other users. In order to develop a recommender applications in a social tagging system the first step is to create a model of the folksonomy that takes into account the information flow between users, resources and tags. In this paper we present a critical view on the existing graph-based tag recommendation approaches specifically on one of the most popular techniques, called FolkRank. In addition, we propose a weighted directed graph which models the informational channels of a folksonomy. We then apply PageRank to this model for tag recommendation. Our claim for a directed graph to model the folksonomy is based on the observation that the user navigation from one object (user, resource, or tag) to another object is not symmetric and by considering different weights on the edges of each direction we can better model the navigating from one node to the other.