Using Kullback-Leibler Divergence Language Models to Find Experts in Enterprise Corpora

The issue of expert finding within an organization has received increased attention in past few years due to its significant importance in knowledge management. Till now, various solutions have been proposed to solve this problem. Among these solutions,generative probabilistic language modeling techniques are most frequently adopted. In this work, we propose a novel model to find experts in enterprise corpora based on Kullback-Leibler Divergence Language Model which has been shown to have better retrieval performance than basic language model in the ad hoc retrieval task.Besides, our methods set a document cutoff to restrict the number of documents that used as evidence of expertise when estimating the probability of a candidate being an expert. Finally, we take out experiments on the benchmark provided by TREC. Experimental results show that the approaches based on Kullback-Leibler Divergence outperform methods based on basic language model and the incorporation of document cutoff also brings substantial gains to the final results.