User generated social annotations provide extra information for describing document contents. In this talk, we propose an effective method to model the categorization property of social annotations and explore the potential of combining it with classical language models for improving retrieval performance. Specifically, a new model based on LDA is presented to take annotations as an additional source for generating document contents apart from the document itself.

We describe strategies for representing and weighting the categorization property and develop an efficient inference algorithm. Experiments are carried out on a TREC collection using annotations collected from Delicious.com. Our results demonstrate the usefulness of social annotations for IR.