NeSi: Netizen Simulator for Evaluating Internet Public Opinion Analysis System

Clarifying potential public opinion information by analyzing comments is an important feature of modern Internet public opinion analysis systems. However, evaluating such systems by analyzing real netizens’ comments requires significant human effort, which can be time-consuming and expensive. Netizen simulation is a cost-effective technique for evaluating Internet public opinion analysis systems. In this paper, we propose a conversational netizen simulator, called NeSi, for the automatic evaluation of such Internet public opinion analysis systems. Describing public opinion incidents, NeSi can automatically generate a sequence of comments based on the Opinion Leader-Follower model. The comments were varied and sentimental. Through a set of experiments that include both automatic and expert human evaluation, we show that NeSi generates comments that are not only related to public opinion but also comparable to human-generated utterances.