Tag recommendation for social bookmarking: Probabilistic approaches

Abstract

Tagging has become increasingly popular with the explosion of user-created content on the web. A ‘tag’ can be defined as a group of keywords that makes organizing, browsing and searching for content more efficient. Users apply tags to a variety of web-based, shareable content including photos, videos, news articles, bookmarks, friends, etc. Tag suggestions for blog posts or web-pages have changed the focus of the tagging process from generation to recognition, thus making it less time and effort intensive. In this paper an intelligent tag recommendation agent is proposed, that recommends tags for bookmarks stored in one of the popular social bookmarking websites, Del.ici.ous.1 ¹

http://www.delicious.com.

We develop various probabilistic approaches to recommend tags to be used by users while adding new bookmarks. We have developed content-based and collaborative filtering mechanism that are used by these recommendation agents. Additionally, these tag recommender agents learn to classify the tags according to their semantic similarity based on collaborative tagging by the users. This approach can therefore be used to facilitate folksonomy formation for social networks. We also empirically verify the hypothesis that similar web pages are tagged with similar tags. We also present a comparison between the proposed recommendation approaches for the agents.