Productivity-based Features from Article Metadata for Fuzzy Rules to Classify Academic Expert

Abstract

Since modeling expertise is necessary in an expert recommendation system, this paper addressed the issue to obtain researcher expertise in the academic field on certain topic interest. The profile considers productivity and dynamicity of an expert. The productivity of research activities through published articles as research output determine expertise that changes over time to indicate the dynamicity aspect. Here, the resulted expertise status on certain topic interest augments the expert profile. However, the expertise status is unavailable in the expert finder dataset. This paper discussed on approaches to classify the status from features of productivity and dynamicity in the form of fuzzy rules, which can be applied later in the expert recommendation system. Then, the approaches include of determining topics, mapping expertise candidates, extracting features, and labeling expertise status for training to generate fuzzy rules. Because of unavailable expertise status, to get better labels, the results of linear model and clustering were compared. Based on the empirical experiments, rules trained from scaled data with expertise labels from fuzzy clustering gave better results. After simplifying the rules, if-then forms with two features were representable enough for identifying the status of specialist or thriving experts on a topic interest.