由於電子商城蓬勃發展，人們的消費習慣也隨著改變。在過去，人們在購買商品前，通常是先透過親友之間的口耳相傳的方式得知該商品的產品評價後，再進行消費；而在現今消費者在購買商品前，通常會透過網路上的討論區或論壇，取得他人對於該商品有關的意見評論，在進行消費，為了能有效且快速取得文章內有關產品或是商家資訊，因此中文意見探勘系統就為之重要。本論文主要是修正(陳子龍 2012)所提出的句型文法，歸納出四種可以修正的錯誤並且加以修正，其中三種是不符合文法規則的錯誤，分別是對等句中的連接句錯誤、比較句中的不符合文法規則錯誤以及一般句的Topic繼承錯誤，而剩下的一種是一般句中的意見矛盾的問題。實驗結果顯示各月份和不同討論區之準確率、回收率和F1均有提升，雖然提升不多，但是從實驗結果之中，顯示我們所修正的方法是正確以及有成效的。 With the maturation of the e-commerce, people''s consumption habit has changed. In the past, before consumer purchased product, they know the product’s evaluation from their friends or relatives. Today, they can know the product’s evaluation form other people''s comments on the website such as Internet forums and discussion groups. In order to obtain the Chinese articles about products or manufacturer’s information correctly and effectively, the Chinese opinion mining system is important. In our study, we improved the sentence analysis grammar defects of the Chinese opinion mining system [1] which are : (1) Coordination sentence error problem(2) Comparative sentences does not comply with the rules of grammar errors (3) The general sentence topic inherited error (4) The contradictions error of the general sentence.We proposed the statistical algorithm to improve the contradictions error. The experimental results shows the method can enhance the precison and recall.
