指導教授：鄭卜壬臺灣大學：資訊工程學研究所葉育昇Yeh, Yu-ShengYu-ShengYeh2014-11-262018-07-052014-11-262018-07-052014http://ntur.lib.ntu.edu.tw//handle/246246/261438網路上的評論文章一直影響著我們的購買決策，但隨著影響程度的日益增強，針對評論文章進行情緒分析(Sentiment)的相關研究也如雨後春筍般地快速地增加。之前以情緒字典為基礎進行文章意見分析的方法大多都忽略了一個事實，那就是每一個字的正反面意見極性其實會隨著上下文的意思而改變。因此，本論文的研究目的係希望由已知意見極性的訓練文章當中，能夠自動建立一個與上下文相關的情緒字典，而當中就算是同一個字也隨著上下文的不同，而表現出相異的意見極性。實驗進行在以三個不同領域的評論文章，實驗結果顯示本篇論文所提出來的方法確實能提升意見分析的效能。As the influence of online reviews on buying decisions has grown, research interests on mining opinions for determining the polarity of a document have increased. Previous lexicon-based methods of sentiment analysis pay less attention to the fact that the context may change word polarity. Given the sentiment information of documents in the training data, the goal of this study is to learn different polarity scores for the same word with various contexts. The experiments have been conducted on the real data sets in three domains. The results of the experiments demonstrate the feasibility of the proposed method.Acknowledgements I 摘要 III Abstract IV Table of Contents V List of Figures VI List of Tables VII Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Thesis Overview 4 Chapter 2 RELATED WORK 6 2.1 Thesaurus-based Approach 7 2.2 Corpus-based Approach 7 2.3 Context-aware Approach 8 Chapter 3 Methodology 10 3.1 Lexicon Definition 10 3.1.1 Context Entry 10 3.1.2 Context-aware Sentiment Lexicon 10 3.2 Context-aware Sentiment Lexicon Construction 11 3.3 Context Vector Generation 12 3.4 Clustering of Context Vectors 14 3.5 Estimate Polarity in Context Entry 17 3.6 Comparison Method 20 Chapter 4 EXPERIMENTS 22 4.1 Data Description 22 4.2 Performance Comparison 23 4.3 Analysis of Gaussian Parameter 24 4.4 Analysis of Sentiment Polarity Stability 25 4.5 Experiment on Machine Learning Method 27 4.5.1 TF-IDF weighted Method 27 4.5.2 Weight-Threshold Method 28 4.6 Example of Context-Aware Lexicon 33 Chapter 5 Conclusion and Future Work 39 References 41754072 bytesapplication/pdf論文公開時間：2019/08/21論文使用權限：同意有償授權(權利金給回饋學校)情緒分析情緒字典自然語言處理前後文相關機器學習梯度下降法支援向量機thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/261438/1/ntu-103-P99922001-1.pdf