Guiding supervised topic modeling for content based tag recommendation

Automatically recommending suitable tags for online content is a necessary task for better information organization and retrieval. In this article, we propose a generative model SimWord for the tag recommendation problem on textual content. The key observation of our model is that the tags and their relevant/similar words may have appeared in the corresponding content. In particular, we first empirically verify this observation in real data sets, and then design a supervised topic model which is guided by the above observation for tag recommendation. Experimental evaluations demonstrate that the proposed method outperforms several existing methods in terms of recommendation accuracy.