情緒詞權重計算與分類演算法對於情緒分析結果之影響－－以臉書粉絲團議題分析為例

On The Effects of Emotional Term-Weight Calculation and Document Classifiers on Sentiment Analysis - Taking Topic Analysis of Facebook Fan Groups As an Example

現今社群網路發展迅速、使用率普及，透過情緒分析掌握民意的取向乃成為近年來熱門研究領域之一。本研究旨在探討不同分類法搭配不同權重方式對於社群網路文本情緒分類的效能，以提出具有較高效能的情緒分析組合方法。然而，情緒分析的方法有很多種，透過相關文獻探討，最後本研究以臉書粉絲團議題分析為例，採用六種權重計算方式對網路文本的詞彙進行情緒詞權重計算，並且針對SVM與KNN兩種分類演算法對文本做分類預測。接著比較分類結果之準確率，再透過T檢定評估這些組合方法之精準率與召回率。結果發現本研究所提出TF-RF-IDF搭配SVM分類演算法為最佳的情緒分析組合方法，最後本研究推薦採用SVM分類法搭配TF-RF-IDF之權重計算方式最適用於了解網友的負面情緒反應。預期本研究成果可以提供未來輿情分析相關系統開發者做為參考之經驗法則和基礎，以利開發出對於網路文本情緒分析具有高精準率的輿情分析系統。

Due to the rapid development and widespread usage of social networking sites, it has been recently gaining much research attention to gauge the orientation of public opinions on specific subjects through sentiment analysis of the social comments in these websites. This research is aimed to investigate the effect of different combinations of classification methods and weighting schemes of sentimental terms on the classification performance of social sentiment comments. Among other combinations, this researchfocuses on combinations of six weighting schemes of sentimental terms and two classifiers, namely, SVM and the KNN. Read data from a Facebook fan group is analyzed by these combinations of methods using R. The research results are discussed in terms of accuracy, precision and recall rates. Furthermore, T tests of the precision and recall rates are conducted to find out which combination is the best. Finally, suggestions about the use of the analysis methods in different situations are given according to the confusion matrix of the test results. It is expected that this research might provide an empirical basis and guiding principles for the implementation of a practical system of social sentiment analysis.