Chinese Micro-blog Emotional Analysis

[[abstract]]本論文以台大楊昌樺所提出的分類方法為基礎，以噗浪微網誌中使用者所發 表帶有表情符號的意見做為語料庫，進行正面與反面的情緒分析。在建立情緒辭 典時，本論文使用DMR(Difference Multiply difference Ratio)值取代楊昌樺使用 CO(Collocation Strength)值來建立情緒辭典。 本論文以DMR 值做為判斷情緒詞彙對情緒的相關性以及重要性的標準，並且 經過DMR 值排列後，把在訓練資料兩個情緒類別中出現次數相同的詞彙給排除掉 來產生辭典。將建立出來的辭典透過楊昌樺的分類方法進行分類後，在支援向量 機部分的平均辨識率為71 77%，而使用CO 值所建立的辭典的辨識率為63 87%， 顯示出本研究所使用DMR 值的特徵選取標準，在支援向量機的部分有明顯的進 步。[[abstract]]This research is based on Yang's classification method using Plurk micro-blog users post comments with emoticons as a corpus for analysis of positive and negative emotions When creating emotional dictionary we replace the CO value that Yang used in his paper by DMR value to create emotional dictionary This research uses DMR value as the criteria of judging the relevance and importance of the emotional vocabulary After the arrangement of the DMR value exclude the vocabulary which has the same occurrences in two emotional categories in the training data to generate dictionary We use the dictionary with Yang's classification method At the part of support vector machine we get the average recognition rate 71 77% and we get the average recognition rate 63 87% by using the dictionary created by the CO value Showing that in this study the DMR feature selection criteria makes a significant improvement at the part of support vector machin