Mitigating group polarization through positive and neutral comment bots

Abstract

The adverse effects of group polarization on social networks are becoming increasingly apparent in today's society, undermining constructive public discourse and threatening political and social stability. To mitigate group polarization, this paper proposes the MGP-PNCB framework, consisting of three modules: polarization data collection, comment generation, and bot embedding. By inputting manually configured prompts into the GPT model, positive and neutral comments are generated and disseminated with the aid of social bots. Additionally, it introduces a polarization alleviation index designed to measure the depolarization impact of specific comments. In the experiment, 60 social bots divided into three categories of 20 each were deployed across four topics, and received 2488 comments from 2183 users over 28 days. Results show that the average sentiment polarity of comments received by bots is more positive than that of regular users. Importantly, neutral bots are more effective in mitigating group polarization than positive ones under the same topic data training.