Analysis And Design Of The Sentiment Analysis Model Of Movie Review Based On Specific Theme

Nowadays,the global Internet coverage rate is rapidly increasing,and people’s various activities on the Internet are becoming more and more colorful.The number of users interested in expressing their opinions and attitudes on various online platforms has gradually increased.As an important branch of the entertainment industry,movies have Become an indispensable entertainment activity in people’s lives.Since the beginning of the era of word-of-mouth,it’s effect has become an important basis for influencing audiences to buy movie tickets,and a series of movie review websites,such as Douban,have also flourished.Therefore,sentiment analysis of movie reviews on movie review websites can not only reflect users’ overall sentiment preference for movies,but also play a strong guiding role for other users who have not watched the movie,and has good commercial use and social value.The main works of this paper are as follows:First of all,use crawler technology to obtain data sets.This article uses Python as a crawler tool,uses Python’s built-in request module,the BeautifulSoup library that can parse web pages,and the Numpy package and Pandas library that can process data to crawl and preprocess the short commentary of the movie "Nezha:Birth of the Demon Child".Next,a topic classification model is proposed to classify the data.This article uses the TF-IDF algorithm to summarize the four thematic categories of "drama script","role actors","production team" and "visual effects".By comparing feature selection methods and multiple machine learning models,word2vec+LightGBM is selected as the topic classification model of this article.Use manual annotation data to train the model,and then use the trained model to subject the unlabeled data.Afterwards,a sentiment classification model is proposed and classify the data.In this paper,a model integrating sentiment dictionary and deep neural network is used as the sentiment classification model.TF-IDF algorithm is used to extract keywords from the four types of topic data,and adds the sentiment words in the keywords to the basic dictionary.The dictionary is used to generate sentiment vectors,which is fused with Word2vec as a text representation.Finally sent it to the Bi-LSTM+Attention model for emotion classification.Finally,integrating text sentiment categories and star ratings to calculate movie theme scores.Taking the star rating as the weight,and the weighted average of the text sentiment category and the star rating is calculated as the theme rating of the movie.Compared with the original scoring method of Douban Movie,users can more intuitively understand the scoring situation of different categories of the movie.