Analysis of the Correlation of Topic Feature Changes Based on the LDA Model

Abstract

Abstract. As is well known, with the continuous development of artificial intelligence technology and the increasing accessibility of data, various social platforms are committed to using intelligent recommendation algorithms to cater to user preferences. Some platforms even exaggerate facts and push sensational, valueless information to users, leading to the "Screaming Effect" and the "Echo Chamber Effect." The prolonged existence of these effects may result in "Information Cocoon," which is detrimental to the healthy development of individuals and society. To address this issue, different topics can have varying trajectories as online comments ferment, potentially reaching a neutral consensus or resulting in polarization. As a mainstream social platform in China, Weibo users serve as nodes for the dissemination of public opinion information, and the characteristics of information release, reception, forwarding, and commenting all influence the effectiveness of communication. First, we selected the topics "COVID-19" and "IG Electronic Sports Club" as our research subjects, identifying a range of topic popularity through Baidu Index, followed by data collection. Second, we used perplexity and coherence to determine the optimal number of topics for LDA, analyzing the changes in topic feature characteristics over different time periods. Through correlation analysis, we reached the following conclusions: Higher Weibo levels correlate with better information dissemination effectiveness; the quantity of information published on Weibo negatively impacts the forwarding and commenting behaviors; and as events progress, the enthusiasm for information dissemination on Weibo declines, with daily dissemination gradually decreasing.