Tags recommending based on social graph

Abstract

As fast development of social network, UGC (user generated content) has played a very important role in "Web 2.0". However most of UGC is non-structured data, which is hard to be used by search engine or user recommending system. Social mining is the way to make UGC accessible. But UGC are trivial, noisy, sparse, causing social mining methods inefficient. In this paper, we propose a tag recommending approach based on social graph. Social graphic recommending can reduce mining depending on UGC, thus be able to generate high quality tags. Our most contribution is to combine social graph with LDA algorithm to find users' latent common interest, thus extract tags. We did experiment on real data crawled from Sina Weibo. The evaluation showed that our approach archived much better precision and recall than baseline methods. © 2013 Springer-Verlag.