Team AxIS

Overall Objectives
Application Domains
New Results
Contracts and Grants with Industry
Other Grants and Activities

Section: New Results

Latent Variable Models for Expert Finding

Participants : Elena Smirnova, Brigitte Trousse.

This work takes place in the context of Elena Smirnova's Ph.D thesis. supervised by B. Trousse (AxIS) and K. Avratchenkov (Maestro).

The majority of existing expertise retrieval algorithms use document collection as a main source of evidence and analyze content with respect to persons associated with the documents [70] . In this setting, latent variable models can be directly applied to expert finding task because it implies existence of “hidden” experts who relate to “observed” documents. We propose an extension of latent Dirichlet allocation model [73] and apply it to expert finding problem. Our approach infers topics of expertise for each person and combines that with social relationships among persons. We assume that if a person has some expertise in the topic and so do persons linked to him, one have more confidence in that person's level of expertise. This also reflects an idea that person's expertise is related to other's through social relationships. Our model enables persons, topics and social relationships to be suitably coupled. Moreover, sparsity in topic representation of a document and person's expertise profile is easily achieved setting appropriate prior parameters.


Logo Inria