探勘主題式情感性社群媒體

近年來,社群網路已成為人們創造、分享或交換資訊與想法等內容的重要管道。而使用者所產生的內容包含大量使用者的想法與情感,儼然成為分析產品偏好、商業策略、行銷活動、社會事件、政治活動以及健康醫療等的重要資料來源。因此,本論文提出三個方法來分析社群網路中具情感的使用者內容,分別為MPM (Mining Perceptual Maps)、MAE (Mining Arousing Events)與MSS (Mining Social Support)。 MPM方法從大量的使用者評論中,利用使用者對於不同產品特徵的正負情感表達,自動建立知覺圖與雷達圖。探勘知覺圖與雷達圖可提供企業以直觀的方法描述企業與競爭對手產品的相對關係。MAE方法使用讀者評論情緒與情緒強度,自動發掘新聞文章中的情緒激發事件,其中每個新聞文章包含一篇新聞本文與數篇讀者評論。探勘情緒激發事件可提供政治家施政參考或推薦熱門新聞給讀者。MSS方法分析健康醫療媒體中,社會支持與發問者情緒轉移過程的關係。發問者態度常隨著情緒轉移而轉變,所以藉由情緒轉移分析社會支持,可更了解病人需求、態度與想法,進而提供較適當的醫療照顧。實驗結果顯示MPM方法能夠從消費者評論中,辨識不同廠牌或價格下各個智慧手機的競爭優勢;MAE方法能更準確地了解讀者的情緒與強度反應，並發掘更細緻的事件;MSS方法則顯示擁有不同疾病的人可能表現不同的負面情緒轉移,並需要不同類型的社會支持。

關鍵字

資料探勘；隱含狄利克雷分布；情感分析；情緒轉移；社會支持；消費者評論；新聞文章；健康醫療

並列摘要

Online social networks expedite social interactions where people create, share or exchange information and ideas. The contents generated by users in a social network usually contain a large volume of user opinions and feelings. They can be used as an effective vehicle to analyze product preferences, business strategies, marketing campaigns, social events, political movements, and healthcare experiences. Therefore, in this dissertation, we propose three LDA-based methods, called MPM (Mining Perceptual Maps), MAE (Mining Arousing Events) and MSS (Mining Social Support), to mine affective social media in social networks. The MPM method automatically builds perceptual maps and radar charts from consumer reviews based on users’ sentiment polarities toward different product features. Mining perceptual maps and radar charts can help companies gain knowledge of their and competitors’ products. The MAE method extracts emotionally arousing events from a collection of news documents based on readers’ emotions and intensities, where every news document contains a news article and some readers’ comments. Mining emotionally arousing events may provide a quick reference for politicians and a new aspect for hot news recommendation for readers. The MSS method mines social support and users’ emotion transitions from online healthcare social media. Mining social support by considering emotion transitions may help us better understand patients’ needs, attitudes and opinions, and provide more appropriate assistance since the changes of emotions often coincide with the changes of attitudes. Experimental results show that the MPM method can find the strengths and weaknesses of various mobile phones of different brands and different levels of prices from consumer reviews. The MAE method can better predict the readers’ emotions and intensities for unseen news articles, and discover better-quality and more subtle events using intensity. The MSS method shows that people with different diseases may express very different negative emotion transitions, and need various types of social support.

並列關鍵字

Data Mining ； LDA ； Sentiment Analysis ； Emotion Transition ； Social Support ； Consumer Reviews ； News Documents ； Healthcare

參考文獻

[2] M.H. Alam and S.K. Lee, “Semantic aspect discovery for online reviews,” Proceedings of IEEE International Conference on Data Mining, pp. 816-821, 2012.

[7] R. Bledow, K. Rosing and M. Frese, “A dynamic perspective on affect and creativity,” Academy of Management, vol. 56, no.2, pp. 432-450, 2013.

[9] J. W. Brehm, “The intensity of emotions,” Personality and Social Psychology Review, vol. 3, no. 1, pp. 2-22, 1999.

[13] A.T. Chen, “Exploring online support spaces: Using cluster analysis to examine breast cancer, diabetes and fibromyalgia support groups,” Journal of Patient Education and Counseling, vol. 87, no. 2, pp. 250-257, 2012.

[14] S.M. Chen and M.H. Huang, “Automatically generating the weather news summary based on fuzzy reasoning and ontology techniques,” Information Sciences, vol. 279, pp. 746-763, 2014.

國際替代計量

探勘主題式情感性社群媒體

全文下載

主題瀏覽