关键词: 社交网络, 情感倾向性, 熵值法, 选情预测

Abstract:

In order to reflect the real public political opinion and improve the accuracy of election prediction, this paper proposes an election forecast model that combines the outcomes of polls and sentiment analysis of netizen. For the poll data, a time-series-based model is used to decrease the bias of polling agencies, and a reverse normalization model is built to infer the political opinion of uncommitted people. For the social network data, a sentiment classification-quantitative model for social network users is used to predict the election result. To improve the prediction accuracy, a fusion-prediction model based on the entropy method is proposed to combine both poll information and sentiment analysis information of network users. Experiments with the ground truth of real historical elections in a certain area show that the proposed model is superior to polls or netizen sentiment analysis according to the accuracy and Mean Relative Error（MRE）.

Key words: social network, sentiment orientation, entropy method, election prediction