在情感分析領域之中,如何將電子本文準確分類至正確的情感傾向,一直是熱門的研究議題之一。若能準確將這些網路上的產品評論文章、網誌內容作情感喜好或厭惡的分類時,將能創造出更多的應用發展,像是提供企業在產品或服務上的競爭分析、營銷分析…等。或給予使用者在查詢一事件或議題上,能快速提供大眾意見,讓使用者能迅速了解公共議題上的情感傾向。SentiWordNet為一在情感分析內重要的情感詞彙資源,以WordNet為基礎將詞彙給予情感數值來表示該詞彙的情感程度,分別給予每個詞彙在不同情況下正面、負面與中立三個傾向的情感數值,透過詞彙的情感數值便能瞭解該詞彙具有何種情感。然而,在過去情感分析的研究中,著重於如何建立與找尋出情感詞彙資源,或使用已建立好的情感詞彙進行情感分析。較少有學者探討SentiWordNet內中立詞彙對情感文章分類的影響,與依據不同領域重新修正中立詞彙情感數值,使其貼近該領域上的情感使用。因此,本研究提出以詞彙相關性方法來重新修正SentiWordNet的情感字典,改善SentiWordNet的情感字典內中立詞彙的情感數值,從中萃取出SentiWordNet情感數值錯誤或未能符合領域的詞彙,藉此來改善情感分類上的準確。
In the academic field of sentiment analysis, knowing how to correctly classify the articles to the corrective sentiment-orientation is a hot research topic. If we can correctly classify the critiques, we can create more applied development. Sentiment detection mainly detects useful information in the articles, from the articles concerning mining or extract useful information, including viewpoint, fancy, and attitude. Although, these articles are valuable, but a lot of articles are dispersal, if reading all articles will take more time and human. If we only read a few articles, this may cause prejudice. Therefore, automatic sentiment detection and classification has become one of hot topics of the day. SentiWordNet is important vocabulary resource in the sentiment analysis. SentiWordNet is based on WordNet. SentiWordNet gives three sentiment scores to each synset. Sentiment scores are representative sentiment intensities on the synset. Through the three sentiment scores one can know the sentiment-orientation of word. However, in the past, sentiment analysis of the study focuses on how to build vocabulary resources and find out the articles that are sentiment-oriented, or use the sentiment vocabulary resources classifying the articles. Few scholars in the academic field of sentiment analysis have explored sentiment objective words influential sentiment classification. Therefore, this study uses vocabulary relevance to revise SentiWordNet objective sentiment score and improve the text classification.