社群媒體上分手文章的性別差異：文本分析取徑

A Text Analysis Approach to Analyzing Gender Differences in Breakup Posts on Social Media

本研究以台灣社群媒體上的分手文章為分析對象，比較兩種使用於文本分析的詞彙分析方式用於區辨作者性別時的優劣。使用封閉式詞彙分析時，本研究根據過去研究，以中文版LIWC詞典中的人稱代名詞與非代名詞類為預測變項，分別建立邏輯迴歸模型預測作者性別。結果顯示，人稱代名詞有較佳的預測力；唯其無法反映分手的特性。使用開放式詞彙分析時，本研究以寫作風格分析的演算法，直接從分手文章中找出最能代表不同性別的關鍵詞。結果發現即使只選取前1%的關鍵詞建立模型，模型的表現也優於以封閉式詞彙分析建立之模型。接著，針對各性別關鍵詞所對應的中文LIWC詞類，分別進行網路分析。根據網路節點的中間度指標，本研究發現動詞、副詞、關係詞、社交歷程詞、生理歷程詞和認知歷程詞為兩性分手文章中共有的LIWC心理語文特徵；但唯獨女性才有情感歷程詞、性詞和否定詞。由此推知，台灣社群媒體使用者中，女性比男性在文章中更有情感方面的表現。

Two approaches to text analysis have been applied in the current work to investigate gender differences in breakup posts on social media in Taiwan. First, we calculated the probabilities of the types of words, such as personal pronoun and other word categories based on the Chinese Linguistic Inquiry and Word Count(LIWC), occurring in the posts to predict author's gender. The results showed that personal pronoun outperformed other word types at predicting gender for social media break-up posts. Second, we conducted stylometric analysis on these posts to extract keywords for different gender. The occurring probabilities of these keywords were then used to predict the author's gender of the post. The results showed that including keywords in the top one percent as predictors enabled a model to perform better than the first approach. A network analysis was carried out, respectively, for each gender to examine the psychological and linguistic features of these keywords and their relationship with reference to the Chinese LIWC. The typical features, defined in terms of the centrality indices, such as word types of verb, adverb, relative, social process, biological process and cognitive mechanism were found to be common for both gender. However, features of affection words, sexual words, and negate words showed up only for breakup posts authored by females. We conclude that among Taiwanese users of social media females were more likely than males to make affective statements.