Tag's Depth-Based Expert Profiling Using a Topic Modeling Technique

Abstract

Expert finding and expert profiling are two important tasks for organizations, researchers, and work seekers. This importance can also be seen in online communities especially with the explosion of social networks. Expert finding on one hand addresses the task of finding the right person with the appropriate knowledge or skills. Expert profiling on the other hand gives a concise and meaningful description of a candidate expert. This paper focuses on what social tagging can bring to improve expert finding and profiling. A novel expertise indicator that models and assesses an expert based on the expert's tagging activities is proposed. First, tags are used as interest indicator to build candidate's profiles; then, Latent Dirichlet Allocation algorithm (LDA) is used to construct the tags distribution over topics by exploiting the tag's semantic characteristics. Topics of interest are then filtered using tag's depth. The latter is finally used as the expertise indicator. Experiments performed on the stack overflow dataset show the accuracy of the proposed approach.