Mouse Movement-based User Sentiment Analysis

In recent years,with the rapid development of artificial intelligence,sentiment analysis in the field of human-computer interaction has become a hot topic,and the sentiment analysis based on mouse movements has been a new research direction.In previous studies,website administrators usually measure the quality of website services and users' satisfaction by analyzing the website traffic or investigating users' feelings,which is not convenient and may disturb users.In addition,there is no way to find which area in the webpage causes user's negative emotion.Therefore,it is necessary to find an efficient way to analyze the sentiment of users in webpages and find the webpage areas causing negative emotions.Through investigating the current efforts of the sentiment analysis methods based on mouse movements,this paper not only improves the method of analyzing the users' sentiment in webpages,but also proposes a novel method of finding the webpage areas causing negative emotions.Therefore,it can not only recognize users' emotion by the mouse,but also respond to website administrators.It generates feedback information from users to websites,which can improve the quality of website design and service.In this paper,the main contributions are the following three aspects:(1)Data acquisition and preprocessing.In this paper,a site is designed to collect users' mouse movements as the original sample data with Java Script.Furthermore,the original sample data should be preprocessed to get the experimental sample data.This paper extracts 108 features from users' mouse movements to construct the unique feature set.(2)An improved webpage sentiment analysis method based on mouse movements.Firstly,four dimensionality reduction methods are used to reduce the dimension of the experimental sample data.Besides,five classification methods in machine learning are used to determine users' emotion(negative / non negative)in webpages.In this way,the webpage sentiment analysis method based on mouse movements can be improved and the accuracy can be enhanced with the best dimensionality reduction method and classification method.The experimental results show that the average accuracy in webpages reaches 90.94%.The chisquare test method is robust and has a better dimensionality reduction effect than other methods.The support vector machine is relatively robust and has the better classification results than others.(3)A novel sentiment analysis method based on mouse movements in the webpage areas.Firstly,We divide a webpage into several areas.Besides,We calculate the complexity of each area in the webpage,which is considered as a feature and added into the feature set.Finally,a novel sentiment analysis method based on mouse movements in the webpage areas is proposed in this paper,by which the heat map of users' sentiment in the webpage area can be drawn.In many experiments,the average highest accuracy of the three page areas reaches 82.81%,which can demonstrate the feasibility and effectiveness of the proposed method.The recursive feature elimination method and the chi-square test method are robust and perform better than other methods.In addition,the support vector machine is superior to others.