Box Office Prediction Based On Consumer Intention And Sentiment Analysis

As a new social media,weibo has accumulated a lot of users and influence.The traditional marketing decisions are greatly influenced by the increasing online reviews,making text mining a hot topic in both business and academic circles.Each comment of different users gathers together into collective wisdom.They not only include consumer intentions but also include sentiment tendency of particular goods.These data containing consumer intentions and sentiment tendency have high value for scientific research and commercial applications.This paper choose creative staffs of film as research objectives,focus on consumer intentions and sentiment information mining and explore these information for box office prediction.Specifically the work of this paper mainly includes the following five aspects:(1)A new definition of consumer intentions classification: Even if a user expresses his/her consumer intention in microblog comment,it is not necessarily representative of the actual consumer behavior.Because consumer intentions include implicit intentions and explicit intentions,we define explicit intentions as positive consumer intentions,which is clear that the consumer is going to purchase.(2)Redefining sentiment classification standards: Due to there is no direct relationship between the number of positive comments and box office,the paper assumes that if there are more comments relative with box office growth and then movie box office will be better.Based on the assumption,the comments are divided into three categories: positive comments are explicit consumer intentions,neural comments are these comments which express positive attitude towards film and negative comments are these comments which express negative or bad attitude towards film.(3)Consumer intentions mining based on SVM: this paper proposed the consumer intention classification algorithm based SVM.Firstly,defining two types of features of positive consumer intentions;Secondly,using manual annotation obtains positive consumer comments and then expressing comments text by using the two type of features.Thirdly,Training consumer classification model and use the model to classify comments text.At last,by comparing the classification result with manual annotation,precision is up to 73%.(4)Sentiment classification based on SVM: In order to mining the information from comments,the paper firstly summarizes the existing emotional resources;Secondly,using SO-PMI to judge the candidate words sentiment attribution and automatically obtain the domain dictionary.And then constructing a sentiment dictionary for microblog comments.Besides,due to the diversity of express means in microblog,we should pre-process review texts and use sentiment words to express these comments.At last,using Libsvm to classify comments into different categories.(5)Predicting box office by using consumer intentions an sentiments tendency: The paper discusses how to use consumer intentions and sentiment tendency for box office prediction and the experiment is carried out using linear regression model and SVM.Results shows that opening weekend box office prediction has better effect when introducing consumer intentions and the explanatory power is up to 87%.Besides,when compared with baseline experiment,the relative error absolute value decreases by 24 percentage points in opening week,and the relative absolute value decreases by 14 percentage points.