Predicting Box Office Based On Fine-grained Sentiment Analysis

Compared with traditional word-of-mouth,social media and online reviews have become an important reference for consumers because they break through the time and geographical limitations,spread faster and have a broader influence.Therefore,how to use online reviews to predict sales performance is currently a major focus of the sales industry.Previous studies on online reviews mostly used coarse-grained sentiment analysis method,but usually some features were generally affirmed while others were denied in the reviews.It is precisely because these positive and negative comments on features cancel each other that coarse-grained analysis fails to accurately quantify the emotional polarity of the reviews.For a more complete analysis,we need to extract the features or attributes of consumer reviews and determine consumers'emotional attitudes toward each feature,that is,a more granular analysis of online reviews can provide new insights.Based on this,this paper takes the actual reviews left by consumers on social platforms as the research basis,and explores the relationship between fine-grained online word-of-mouth and box office income in the context of China's film market.Firstly,a structured process method was proposed to calculate fine-grained sentiment index,specifically:1.Carry out the Jieba word segmentation,de-stopping words and part-of-speech tagging and other data cleaning operations;for the scraped Douban reviews;2.Use SO-PMI method and Word2vec technology to expand the sentiment dictionary and feature lexicon;3.Extract the subjective comment sentences with feature-opinion pairs by combining semantic and syntactic analysis,and introduce the deep learning model LSTM to realize the judgment of sentiment tendency;(4)Combine the frequency and sentiment polarity of the features to determine the weight,and use the bullish index formula to construct a daily fine-grained sentiment index.Then,based on 540 panel data,the fixed effect model of fine-grained emotion index and box office revenue was established.In addition,in order to investigate the potential complementary effect of online search data on online word of mouth data sources,this paper introduced online search data variable on the basis of the original model,and tested the relationship between online word of mouth,Internet search and box office through panel-vector autoregressive model.The research conclusions of this paper are as follows:1.The empirical research results show that the structured method proposed in this paper extracts the feature-viewpoint and calculates the corresponding fine-grained sentiment index method is scientific and effective.The fixed effect prediction model based on fine-grained affective index can not only provide more explanations for boxoffice changes,but also have better actual prediction accuracy(R~2=0.70,MAPE=0.352,RMSE=1.321).2.This paper examines the interaction between online search data and publicly available comment data on social media sites.The results show that the online search data does not overlap with the information in online word-of-mouth.The online search index contains prediction information other than online word-of-mouth,which is a potential supplement to the prediction effect of the fine-grained sentiment index model.This paper enriches the theoretical research on online word-of-mouth sentiment mining,online search data,movie marketing and other related aspects.The fine-grained sentiment index mining process proposed can provide a reference for the research in other fields.The box office prediction model constructed in this paper can provide a basis for cinema line companies to make reasonable propaganda strategies and arrange movie screenings.