Finding experts in community question answering services: A theme based query likelihood language approach

Abstract

Community question answering services provide an open platform for users to acquire and share their knowledge. In the last decade, popularity of such services has increased noticeably. Large number of unanswered questions is a major problem for the growth of such services. A common way to address this issue is to route a new question to some selected users who have the potentiality in answering the question. Expert finding is the process of selecting such potential answerers. In this article, we have introduced an efficient method for expert finding using the theme in query likelihood language (QLL) model. Theme of a query is nothing but its subject matter and we have decided it based on the parts of speech (POS) of the words in the query. Depending on the theme of the given question, its similarity to a question in the archive is determined using the QLL model. Aggregating the similarity values of the questions a user answered previously (i.e., in the archive), his/her expertise for the given question is obtained. The performance of the proposed method is verified on a real world dataset (obtained from Yahoo! Answers) and it is found to be quite encouraging.

Expert recommendation in community question answering: a review and future direction

“…(Yi and Godavarthy, 2014) proposed a predictive language model to solve the future expert finding problem. Their method probabilistically estimates the association between a candidate e and a topic m in a future time t 2 ; while the method described in (Momtazi and Naumann, 2013) estimates the mentioned probability according to their association in current time t 1 . (Srba et al, 2015) employ a probability model based on latent topics identified by LDA for the expertise estimation employed in QAbased approaches with non-QA sources of data to estimate users' knowledge early and more accurately for users with low levels of QA activity.…”

Convolutional neural networks for expert recommendation in community question answering

“…The results showed that statistical topic models can be regarded as a suitable alternative for expert recommendation. Mandal et al [6] introduced a method using the theme in query likelihood language model for expert finding. In this case, the theme of the query was based on the parts of speech (POS) of the words in the query.…”