Deep Learning Based Public Opinion Analysis of Game Communication

In this study, we conduct the public opinion monitoring of live broadcast is contented by gathering barrage data from gaming-oriented live streams on platforms such as TikTok. We devised a lexicon specific to the gaming domain and compiled a corpus consisting of 890,206 words within this domain. Additionally, we identified 122 previously unrecognized sensitive words pertinent to the gaming domain. Leveraging techniques such as TextRank for keyword extraction, co-occurrence semantic network analysis, the Latent Dirichlet Allocation (LDA) model for topic discovery, sensitive word recognition, and the Bidirectional Long Short-Term Memory (BiLSTM) model for sentiment analysis, we proficiently elucidate the primary themes, content structure, and emotional trends present within the barrage. Furthermore, we developed a deep learning-based visualization system tailored for monitoring gaming-related public opinion. This system serves as a comprehensive tool, encompassing four core modules: keyword extraction, topic discovery, sentiment analysis, and sensitive word matching, which are further subdivided into 11 sub-modules to facilitate the visualization of analysis outcomes.