Research And Implementation Of Expert Retrieval Based On Voting Model

With the continuous development of enterprise information, expert retrieval has become increasingly noticed to by retrieval research organizations. Feldman, Sherman and others’ report highlighted the importance of enterprise information access in particular. Therefore, employees can effectively access to important information has become the key to improve competitiveness. The traditional search engine can hardly meet the workers’professional needs. Then TREC offered expert search task in2005, which is completely different from traditional information retrieval task, and whose search goals are no longer a document list, but a list of experts. So-called an expert list is a person list ranked by the expertise authority.Through analysis and comparison of existing expert retrieval methods, this thesis adopts Voting Model as the basic model. In the Voting Model, expert retrieval is viewed as a voting problem. Create a profile for each candidate, which is a describing document set describing the candidate. For query Q, the relevant documents are R (Q). For each candidate, a document in his profile and being one of R (Q) is regarded as a vote to the candidate. On the basis of the Voting Model, we take the precise relationship between candidate and the documents into account, and the correlation coefficient signifies the reliability of the vote to the candidate by the document. The relationship is mined based on the form, position of the candidate appearing in the document, and the document quality. The thesis ranks experts not only by his profile voting for him, but also by his social network. The more experts in one’s social network, and the more authoritative they are, then it is more likely that he is an authoritative expert. In the social network, two candidates’association appears as the co-occurrence in one document, and the hyperlink between their profile documents. Experiments show that the improved Model (Dev-Voting Model) put forward in this paper, has obtained a remarkable improvement in precision ratio compared to the Voting Model. Also the ranking algorithm makes the improvement more remarkable. So the Dev-Voting Model and the Expert Rank algorithm is reasonable and effective in the expert retrieval.