Competence Estimation: Classifying Expertise of Web Discussion Participants

Abstract

Web discussion bulletin boards have been attracting much attention in recent years. They enable people to hold discussions online, whereas these have traditionally been conducted face-to-face at town meetings, etc. However, in a large-scale bulletin board, not all of the participants have a deep understanding of a topic when engaging in discussion. In particular, it is important to automatically classify a person who is making useful posts if a bulletin board is intended to make agreements. This paper proposes an automated method to identify the expertise of participants by defining expertise as a requisite argumentative competence. In this paper, we propose novel features for competence estimation models: lexical features (IDF, discourse marker, and topic similarity) and a directed influence graph feature. Furthermore, in the evaluation experiments, we evaluate the precision-recall curve against the baseline. As for datasets for evaluation, the expertise of the participants in the data of discussion conducted in the actual Web discussion bulletin board is annotated in seven grades. The experimental results demonstrate that the proposed methodology is effective in many cases.