中文寫作多面向評分系統

Abstract

自動寫作評閱的研究，在自然語言中佔了重要的一環，尤其是在中文研究上甚是艱難，雖然陸陸續續已有評閱系統之研究產生，但目前的系統皆只針對文章單一方面給分，無法有效提供使用者在寫作技巧上哪方面較微弱之資訊。因此本文提出一個非監督系統，針對中文寫作評分不同面向，分別給予分數以及分數的統整，除了給予使用者在立意取材以及結構組織上的分數外，也根據使用者所寫作的文章給予錯別字回饋的資訊。實驗結果在不同面向上能有相當程度的正確率，在分數統整上，正確率可達到94%。此外錯別字判斷的正確率能達到72%，可作為老師批閱或是學生寫作上的輔助工具。

The research of the automated essay scoring is important in the natural language and it is more difficult when applying it on the Chinese language. Although some scoring systems have been proposed, they only score the article in one way. They can not provide the information that which aspect in the article the users should strengthen to improve their writing skill efficiently. Thus, this paper proposed an unsupervised learning system that could grade essays from multi-dimension and give misspell information to the user as well. The experiment shows that the adjacent rate in the overall experiment is about 94% and the misspell judgment rate is about 72%.