Research On Key Aspects Of Sentiment Spreading In Social Media

With the development of Web2.0, users publish text, photo, audio and video messages to share their statuses or interesting contents. Different from traditional media, messages spread faster in social media. Thus, social media becomes an important medium for information spreading. Texts are basic mes-sages in social media. They contain substantial contents and emotions. Emo-tions spread by retweeting behaviors. Based on social theory, users have emo-tional influence on other users in social media. This paper focuses on sentiment spreading from four aspects.1. A conformity-based rank model is proposed to find influential users. Whether retweeters keep the same sentiment as original users is taken into consider-ation. Emotional conformity is introduced to denote how users conform to original users from the emotional perspective. Conforming weights are in-troduced to denote how two users keep the same sentiment after retweeting messages. Users are categorized into three classes. Influence is calculated for different classes. Experiments were conducted on Sina Weibo to verify the effectiveness of the proposed model.2. The emotion-based independent cascade model (IMIC-OC) is proposed for positive influence maximization. The IC-OC model for information spreading is proposed to explain how users build their opinions. The pro-cess of information spreading is as follows. In the beginning, users hold positive or negative opinions with probabilities. When more users become involved in discussions, users change their opinions to opposite opinions with probabilities. Influence is calculated when users stop changing their opinions. The proposed model has larger positive influence than baseline methods.3. A user-level model for emotion prediction is proposed to improve the per-formance of predicting target emotions. Existing methods in emotion pre-diction mainly use personal emotion and friend emotion to predict tar-get emotions. Public sentiment is introduced to denote the sentiment of the majority in the network. Public conformity is calculated to measure the degree of a user conforming to the public sentiment. According to the public conformity, users are categorized into three classes: approvers,independents, and starters. The user-level model for emotion prediction predict target emotions of different classes of users, taking into account of the public sentiment, individual sentiment, friend sentiment and pseudo-friend sentiment. Experiments conducted on Sina Weibo show that the proposed model could achieve performance improvements to some exist-ing methods in most cases.4. Two emotion-based models for information diffusion are proposed to study the process of emotion spreading. The emotion-based SIS (Spreaders-Ignorants-Stiflers) model is proposed to study the situation that emotions are not changed during the process of information spreading. Weights are introduced to represent the retweeting strength for each emotion. The emotion-based independent cascade model is proposed to study the situ-ation that emotions could be changed in information spreading. The pro-posed model divides the process of sentiment spreading into three steps.First, propagating probability is introduced to predict whether users re-publish messages. Second, a learning model is applied to predict whether emotions are changed after re-publishing. User features, structural fea-tures and tweet features are introduced in the learning model. Third, trans-forming weights are calculated to predict what sentiments of retweets trans-form to. Experimental results indicate that the emotion-based independent cascade model has the best performance.