Research On Emotion Prediction For Online News From Readers’ Perspective

With the rapid development of the Internet and Web 2.0, an enormous of online news services provide users with interactive platforms where users can share their subjective opinions and emotions, such as sadness, surprise and anger, towards the news articles. Such emotions can not only help understand the preferences and perspectives of individual users, but also help to establish an effective and fast public emotion monitoring. Therefore, how to predict these emotions evoked by online news articles is the main goal of this paper.This paper focuses on predicting multiple emotions of the readers evoked by online news articles based on LDA topic model and multi-label classification techniques. The main contributions include:Firstly, this paper tackles the task of predicting multiple emotions of the readers as a multi-label classification task. Conventional emotion predicting research primarily focuses on determining the emotions of the authors who created the articles, based on single-label classification methods. However, it is conflict to the observation that many news articles could evoke more than one major emotion. To the best of our knowledge, this is the first research work for addressing the task.Secondly, this paper proposed an automatic data annotation method with data of readers’ emotion votes, which greatly reduced the time and money consumption of manual annotation. This method could avoid the disadvantages of traditional manual annotation methods.Thirdly, this paper proposed a novel multi-label supervised emotion-topic model based LDA topic model, named ML-sETM, which introduces an additional emotion layer to associate latent topics with readers’emotions.To test and verify our proposed ML-sETM model, series of experiments are designed on real dataset from Sina News. Experiments results demonstrate the effectiveness of the proposed approach in emotion predicting for online news.