運用文字探勘與情感分析探討台灣社群平台上的憂鬱訊息風險與情緒狀態

Using Text Mining and Sentiment Analysis to Analyze the Risk of Depressive Messages and Emotional States on Social Media in Taiwan

社群媒體平台的蓬勃發展為研究使用者心理健康狀態提供嶄新視角。本研究應用文字探勘與情感分析技術，探討台灣PTT社群論壇「Prozac板」上的憂鬱相關貼文及其情緒特徵。研究分析2017至2021年間的文章內容，結果顯示使用者自行標示的「陰天」（Cloudy）（表達負面情緒）及「晴天」（Sunny）（表達正面情緒）類別文章在情感分析分數上存在顯著差異。「陰天」類別文章的情感分析平均分數明顯低於「晴天」類別。詞頻統計分析發現，「陰天」類別文章頻繁出現「焦慮」、「厭惡」等負面情緒詞彙，而「晴天」類別則傾向使用正向情感詞彙。當將「陰天」文章依風險程度進一步分類（正常、輕度、中度、重度）後，雖發現中重度風險類別的比例相對較低，但其潛在危險性不容忽視。本研究凸顯社群媒體作為憂鬱症早期預警工具的應用價值，同時指出在社群媒體平台建立心理健康支持機制的重要性。

The rise of social media platforms has created new opportunities for examining mental health discussions and emotional states among users. This study employs text mining and sentiment analysis to investigate depressive messages and emotional expressions on the Prozac Board, a prominent mental health forum in Taiwan. Analyzing posts from 2017 to 2021, we identified significant differences between posts categorized as“Cloudy”（陰天）(indicative of negative sentiment) and“Sunny”（晴天）(indicative of positive sentiment). Sentiment scores for“Cloudy”is significantly lower than the average for“Sunny”posts. Term frequency analysis revealed the prevalent use of negative emotion-related terms such as“Anxious”and“Hate”in“Cloudy”posts, whereas“Sunny”posts featured terms associated with positive emotions. Further classification of“Cloudy”posts into risk levels (Normal, Mild, Moderate, Severe) highlighted a concerning number of posts in higher-risk categories, with a small but significant portion indicating severe depressive symptoms. This study underscores the potential of social media as a tool for the early identification of depressive risk and the urgent need for targeted mental health interventions on social media platforms, motivating the audience for change.