基于情感分析的影评数据挖掘

Data Min-ing in Film Review with Sentiment Analysis

摘要: 用户评论已经成为人们选择商品的重要参考目标，影评亦是如此。为了帮助浏览者有效的解读影评文本，了解影评中的情感因素，寻找他们想要了解的内容，并通过评分的方式让用户了解电影整体评价；本文基于中文文本情感分析，获取影评中的评论者情感，基于《战狼2》和《红海行动》两部电影的豆瓣网的影评数据，分析影评文本中的情感倾向，并构建影评特征词表，进而绘制“影评–影评特征”二分网络，分析每条评论中的情感特征，结合用户的评价评分信息，最终构建包含影评情感信息的电影评分模型。通过与原始豆瓣评分比较，发现在考虑到用户影评情感信息后的电影评分与原始豆瓣评分有一定差异，而这些差异大部分是由于用户情感导致的，据此给出电影评分评价模型。

Abstract: The purpose of this paper is to improve the resolvability of film review, the understanding of the review for network users and become comprehensive approach in evaluation of film. Based on the information of Wolf Warriors 2 and OPERATION RED SEA on Douban, we construct “review feature” form and draw the bipartite network of “review-feature”. Second, we put forward a new score model by complex network projection and sentiment analysis. Then, utilizing complex network projection compute the sentiment score of two films and combining with the “star” score from users’ scoring information get the integral score of two films. Finally utilizing the model in other film website computes the score to check it effectiveness. It shows the co-occurrence and relevance between review and the feature under the bipartite network and the sentiment difference among different feature. Comparing with the films’ score on Douban, the score with sentiment information shows better performance in explaining the film evaluation. Besides, it gets the consistent score in other film website by utilizing the model.