去除劇情干擾的動漫與輕小說評論支持度分析系統

Abstract

現在的出版作品，有許多的管道通路可以推廣，而隨著市場的擴大，許多的作家及出版業者也想加入並占有一席之地，知道讀者愛好的趨勢潮流是必須的，網路的作品評論是一個參考的重要指標，許多研究利用關鍵字抽取及句型結構分析等自然語言處理的方法來直接分析評論中支持或反對作品的程度。然而，在劇情類型作品評論判定的過程中時常遇到表示意見的關鍵詞彙並不是用於作品本身評論而是作品中情節之文字描述，為此，本論文在解決此問題上提出使用詞性標註與句型結構樹及文句間關聯性質來做處理，在本篇論文中不同於以往的研究是單純建立意見特徵字典，在本篇論文中我們提出了建立劇情字典來幫助我們將純劇情文句分開以排除作品內容對於意見判斷的干擾而達到改善意見判定正確率的目標。

In recent years, there are many author and publishing house want to join the market of cartoon, comic, and novel. So, they may want to find what type of book do readers like. In the Internet, there are so many critics and impressions in it. They may become important indicator to help finding the stream of market. Many researches use key word extraction and analysis of sentence structure. However, in the critics or impressions, we can find the sentence of story discussing or statement in them. It may make us finding the opinion sentence more difficult. In this paper, we use POS tagging and Parsing Tree and relation between nearby sentence to solve this problem. Different from the former work that only had opinion feature word dictionary, we construct a story feature word dictionary to help us taking the story sentence apart. It can improve the accuracy of opinion judgment.