- 學位論文
線上社群發文之複合情緒分析
Complex emotion analysis for online community postings
摘要
網路社群蓬勃發展有目共睹，已成為現代人生活的一部分。大多數人除了在其中閱讀他人文章外，也會透過發文抒發己見，甚至發洩情緒。因此，社群成員情緒的監測便成為管理者的重要課題之一。管理者雖可觀察討論區中的發文獲得相關資訊，但由於資料量過於龐大，上述方法不但曠日廢時，且準確性堪慮。因此，學者們便提出各種情緒偵測方法，分析討論區發文造成之情緒反應，以協助管理者早期發現並發展因應對策。前人研究提出之方法固然有效，但面對日益複雜的網路社群，仍有待改進之處。首先，相關研究經常使用正、負、中立等情緒做為分類標記，無法提供合理的分析結果。此外，前人研究已歸結出目前機器學習方式的三大問題: 分別是依賴大量訓練資料、不同訓練導致結果不一致與推理過程的不透明。在運用機器學習於情緒分析時，需有更精細的設計與考量。有鑑於此，本研究以發展更有效的情緒分析方法為目標，設計一套多維度情緒偵測方法-Dimensional Emotion Identification with Multi-BERT (DEIMB)。首先，我們採用考量人類感受之情緒沙漏模型(Hourglass of Emotions)來表示量化的複合情緒，讓文章標籤有更一致的定義，以提升情緒分析結果的解析度。為顧及更高解析的情緒維度，我們提出一套以Google BERT語意分析模型為基礎的情緒分析方法DEIMB。配合情緒沙漏模型，針對不同情緒維度分別建立極性偵測模型與程度值偵測模型，最後再加以組合，以提供更具參考價值之複合情緒判別結果。為驗證提出方法之有效性，本研究以網路社群實際發文資料進行分析。經實驗結果發現，本研究提出方法能在情緒極性方面取得合理的準確率。此外，與傳統正負極性判別結果不同，本研究之結果能提供更高解析複合情緒描述，有助於社群管理者深入了解發文者的心情。對於多維度情緒偵測而言，準確性雖不如預期，但具有較高的情緒解釋性，顯示此做法具有發展之潛力。
並列摘要
The prosperous development of online communities is obvious to all and has become a part of modern people's life. In addition to reading other people's articles, most people also express their opinions and even vent their emotions through articles. Therefore, monitoring the emotions of community members has become one of the important topics for managers. Although managers can obtain relevant information by observing the postings in the discussion forum, due to the huge amount of data, the above method is not only time-consuming, but also inaccurate. Therefore, scholars have proposed various emotion detection methods to analyze the emotional responses caused by posting in discussion forums, so as to assist managers in early detection and development of countermeasures. Although the methods proposed by previous studies are effective, there is still room for improvement in the face of increasingly complex online communities. First, related research often uses positive, negative, neutral and other emotions as classification labels, which cannot provide reasonable analysis results. In addition, previous studies have concluded three major problems in current machine learning methods: reliance on a large amount of training data, inconsistent results caused by different training, and opaque reasoning process. When using machine learning for sentiment analysis, more careful design and considerations are required. In view of this, this study aims to develop a more effective emotion analysis method, and designs a multi-dimensional emotion detection method-Dimensional Emotion Identification with Multi-BERT (DEIMB). First, we use the Hourglass of Emotions model that considers human feelings to represent quantified composite emotions, so that article labels have more consistent definitions, so as to improve the resolution of sentiment analysis results. In order to take into account the emotional dimension of higher parsing, we propose a set of sentiment analysis method DEIMB based on the Google BERT semantic analysis model. With the hourglass of emotions, a polarity detection model and a degree value detection model are established for different emotional dimensions, and finally combined to provide a more reference value for compound emotion discrimination results. In order to verify the effectiveness of the proposed method, this study analyzes the data of actual articles published by the online community. The experimental results show that the method proposed in this study can achieve reasonable accuracy in emotional polarity. In addition, different from the traditional positive and negative discrimination results, the results of this study can provide a more analytic composite emotional description, which is helpful for community managers to deeply understand the mood of the sender. For multi-dimensional emotion detection, although the accuracy is not as good as expected, it has high emotion interpretability, indicating that this approach has potential for development.
參考文獻
延伸閱讀
- Chuang, Y. Y. (2018). 社群文章之情緒分析與重排以避免社會過載 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU201800702
- 吳佩芬（2016）。特色街區導覽系統之使用者情緒與沉浸經驗研究。設計學報，21(2)，25-48。https://www.airitilibrary.com/Article/Detail?DocID=16068327-201606-201606300014-201606300014-25-48
- Yeh, J. H. (2010). 連續中文語音情緒辨識之研究 [doctoral dissertation, Tatung University]. Airiti Library. https://www.airitilibrary.com/Article/Detail?DocID=U0081-3001201315105774
- Chiu, T. H. (2020). 透過主題和情緒分析探討社群關係 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU202002163
- Huang, W. Y. (2012). A Study on the Combination of Emotion Keyword to Improve the Speech Emotion Recognition Accuracy [master's thesis, Tatung University]. Airiti Library. https://www.airitilibrary.com/Article/Detail?DocID=U0081-3001201315113175