Public Opinion Prediction by Combining Link Analysis and Sentiment Detection

• Main Authors: Ting-Wei Liu, 劉庭瑋
• Other Authors: Jenq-Haur Wang
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/7jm328

碩士 === 國立臺北科技大學 === 資訊工程系研究所 === 105 === With the popularity of social network and mobile network, people can easily share opinion with others in social networking site. More and more research tends to analyze user data on the website rather than use traditional methods. By researching user data on the website, we can improve the accuracy of prediction because of the large number of data. In order to improve the accuracy of opinion prediction. We combined sentiment detection and link analysis in this paper. In terms of sentiment detection, we use Long Short-Term Memory (LSTM) networks to train the sentiment classifier, and predict user sentiments in the posts by hierarchical method. In terms of link analysis, we calculate influence score of users and posts by TURank. We also change the link in TUrank, solve the memory problem of computing by observe social networking site. At the end, we predict user opinion by sentiment detection result, and then use influence score as user weight to calculate opinion polarity score. We use the result of 2016 United States presidential election to evaluate our method. Results shows our classier have a high F-measure of 72.3%, and can effectively predict the election result by combining influence score.