隨著Web 2.0的概念被提出，加上近年來社群媒體興起，情感分析（sentimentanalysis）逐漸成為新興研究的趨勢，其相關研究與應用的價值也越來越重要。意見單元（或稱情感評價單元）是評價語句中的評價對象及其對應的意見詞的組合，由於意見單元決定了此評價的意見傾向，因此意見單元的抽取是為情感分析領域的重要任務之一。本研究採用系統發展研究法建置一套基於語句層級中文語法規則的意見單元抽取方法之雛型系統，並使用資料探勘技術歸納出意見單元的抽取規則，以建立意見單元抽取模式。研究以「智慧型手機」產品的評論文章驗證方法架構，實驗結果發現，同時使用語句結構與句法路徑結構作特徵屬性，有助於本系統意見單元抽取模式品質的提升，且語句結構在意見單元抽取較句法路徑結構具影響性。研究結果顯示，本研究所建立的意見單元抽取模式，與相關研究的意見單元抽取方法比較，具有較佳的F-Measure值。

Purpose- Through Web 2.0 concepts being advocated to bring about internet opinion groups growing in recent years, the field of Chinese Sentiment Analysis related research has expected more attention and value. Opinion Unit (or Appraisal Entity) is to define the association of opinion words and their corresponding subjects. Because of the opinion unit regulates the polarity of comments, extracting and analyzing opinion units is significant task for the field study of Chinese Sentiment Analysis. Design/methodology/approach- This paper used the systems development process in information systems research to build a prototype system of a method of opinion unit recognition based on the syntactic rules in Chinese we proposed, and used the techniques of data mining to summarize opinion unit recognition rules to establish an opinion unit extracting mode. Findings- The study subjected to smartphone discussing comments was used to test our method of opinion unit recognition and the experiment indicated using the sentence structure and syntactic path structures as feature attributes would contribute to opinion unit extracting mode, and the statement structure was more influential in the opinion unit recognition rules. Results showed that our opinion unit extraction mode is better than correlation studies in F-Measure. Research limitations/implications- This paper focuses on the discussing comments related to smartphone appraisal group. Hence, it is suggested that future research may apply our method of opinion unit extracting mode to other areas, such as computer, car or food. Also, future research is recommended to compare with using other data mining classification, such as SVM, Decision Tree, K-NN or Bayesian Statistics. Practical implications- This paper proposes the extraction principle of opinion unit. In commerce, it may apply to the sentiment analysis of products usage discussing comments. Also, future research may use the proposed attribute of statement structure and attribute of syntactic path structures to make an extensive study. Originality/value- This paper proposes a method of opinion unit extraction based on statement level that can be applied to the discussing comments about smartphone appraisal group. Also, it implement the method and use data mining classification with attribute of statement structure and attribute of syntactic path structures to fulfill a rule of opinion unit extraction.