Analyze Trolling Behavior Trends in Chinese Mobile Game Forums Using Sentiment Analysis

In Chinese online social media, there is a prevailing opinion that the style of speech on the Internet is becoming increasingly rough, thoughtless, and more aggressive compared to the past, often tending towards attacking others. This style of speech, often characterized by provocative and offensive comments, is also referred to as “trolling” in the context of Chinese online discourse. Some even point out that individuals on the Internet deliberately provoke disputes for their own benefit or simply for fun. This behavior, known as “setting the tone”, involves the intentional shaping of the conversation's direction through provocative or inflammatory statements. To investigate the reasons for these changes in online speech styles and to explore possible solutions, this paper utilizes sentiment analysis to study whether there is a tendency for“trolling” in online speech within specific social media environments and to speculate on the presence of the influence of “setting the tone” behavior. In this work, we study the mobile game discussion section of the well-known Chinese Internet forum NGABBS. Most people believe that on the Chinese Internet, forums related to subcultures - especially those related to mobile games - are the most likely to have arguments and frequent occurrences of “trolls” and “setting the tone” behaviors. We collect and label forum posts from a specific discussion forum (i.e., The Resonance Solstice forum). We examined the performance of the traditional machine learning model Naive Bayes and recurrent neural networks for the sentiment analysis of our data. Then, we used our trained sentiment analysis model to predict and analyze the sentiment dynamics of the forum posts from two other game sections (i.e., the Blue Achieve CN forum and the Mobile Game General Forum) in NGABBS. Our experiment results on evaluating sentiment analysis models show that deep learning based sentiment analysis models can achieve accurate prediction. Based on the analysis of sentiment dynamics on the forums, we observe that the Chinese sub-cultural mobile game online community has clearly shifted towards a predominantly negative tone. Based on the sentiment trends analysis, we also observe an immediate shift in sentiment distribution when the community encounters negative events.