Finding topic-level experts in scholarly networks
Scientometrics
Hi-index | 0.00 |
As a retrieval task, expert finding has recently attracted much attention. And various methods have been proposed to rank expert candidates against topical query. The most efficient approach is document-based method that treats supporting documents as a “bridge” and ranks the candidates based on the co-occurrences of topic and candidate mentions in the supporting documents.However, such kind of methods models relevance between query and candidates on the much lower and hence less ambiguous level. It lacks of the capability to capture the hidden semantic association between queries and candidates. In this paper, we propose a hidden topic analysis based approach to estimate the relevance between query and candidates. It models query and supporting document as a word-topic-document association instead of the word-document association in language model. In addition, the prior knowledge of supporting document is considered to favor expert ranking. The empirical results on metadata corpus have demonstrated the model can effectively catch the semantic association between queries and candidates, thus improves the performance of expert finding.