An emotion-based independent cascade model for sentiment spreading
Introduction
Users express opinions of events or products on OSNs. The opinions contain subjects and emotions, which could be spread by the retweeting behavior. The spreading process of emotions is referred to as sentiment spreading. The evolution of sentiment spreading is studied in this paper based on the process of information spreading. We focus on whether the sentiments of the retweets are changed and what the sentiments of the retweets transform to during the process of sentiment spreading.
The independent cascade model (IC) and the linear threshold model (LT) are two basic models for information diffusion. The basic process of the IC model is as follows. Some active users are chosen to spread the information in the network. They try to activate their neighbors according to the edge weights. The basic process of the LT model is as follows. Each user has a threshold. When the sum of its edge weights is larger than the threshold, the user retweets messages. These two basic models mainly handle the problem that whether users retweet messages. We pay attention to the sentiment during the process of information diffusion. Thus, an emotion-based IC model (EIC) for sentiment spreading is proposed to study the process of sentiment spreading in three steps. First, propagation probabilities of edges are proposed to predict whether users retweet messages. Second, a learning model is applied to predict whether the sentiments of retweets are changed. User features, structural features and tweet features are taken into account in the learning model. Third, transforming weights are introduced to determine what the sentiments of retweets are.
Experiments were conducted on the dataset of Sina Weibo.1 Sina Weibo, a Twitter-like microblogging service, provide a platform for users to express emotions. First, the basic sentiment characters of retweets were analyzed, such as the sentiment proportions of retweets and the distributions of retweets whose sentiments are changed per tweet. Second, the process of sentiment spreading is studied. The experimental results demonstrate that the proposed model achieves better performance than two baseline methods.
The following parts of this paper are organized as follows. Section 2 summarizes the related works on sentiment analysis and information spreading. Section 3 gives the problem statement and Section 4 introduces the emotion-based independent cascade model for sentiment spreading. Section 5 presents the experimental results and Section 6 makes a conclusion.
Section snippets
Related works
This part introduces the related works on sentiment analysis and information diffusion.
Problem statement
This part introduces some definitions. A directed graph denotes the social network, where denotes a set of users and denotes a set of directed edges.
Definition 1 Sentiment s denotes the attitude toward events or products. Fine-grained sentiment classification categorizes sentiments into seven classes: objective, happy, angry, sad, fearful, disgusted, and surprised. Some numbers are used to denote sentiments:
An emotion-based independent cascade model
In this section, an emotion-based independent cascade model for sentiment spreading is proposed. The proposed model would handle the three aforementioned problems. Fig. 2 shows the actions defined for the three problems. The explicit definitions are given later.
Definition 5 Neighbors N(vi) denote the set of users who have connections with user vi. Nin(vi) and Nout(vi) denote the sets of in-neighbors and out-neighbors of user vi, respectively. Example 1 As shown in Fig. 2, user v2 is an out-neighbor of user v1, v2 ∈ N
Dataset
The dataset of Sina Weibo is applied to verify the proposed model. With the APIs of Sina Weibo, some users were selected as the seed users and their neighbor information was crawled. The final dataset contains 219,837 users, 2,739,814 tweets, and 3,293,709 undirected relations.
Sentiment analysis
A lexicon-based algorithm is used for sentiment analysis. For a given sentiment, if the artificially identified class i is the same as predicted class j, then the sentiment prediction is correct.
The fine-grained sentiment
Conclusions
An emotion-based independent cascade model was proposed to study the process of sentiment spreading. The proposed model categorizes the process of sentiment spreading into three steps. First, the propagation probability is introduced to predict whether users retweet messages. Second, a learning model is used to predict whether the sentiments of retweets are different from the sentiments of the original tweets, if users retweet messages. User features, structural features and tweet features are
Acknowledgments
This work was supported in part by National Basic Research Program of China (973 Program) and Natural Science Foundation of China (No. 61402045).
References (26)
- et al.
Sentiment contagion in complex networks
Physica A
(2014) - et al.
Esis: emotion-based spreader-ignorant-stifler model for information diffusion
Knowl. Based Syst.
(2015) - et al.
Influence maximization in social networks under an independent cascade model
Physica A
(2016) - et al.
Survey on aspect-level sentiment analysis
IEEE Trans. Knowl. Data Eng.
(2015) - et al.
From tweets to polls: linking text sentiment to public opinion time series
Proceedings of the fourth International AAAI Conference on Weblogs and Social Media
(2010) - et al.
Do users rate or review? boost phrase-level sentiment labeling with review-level sentiment classification
Proceedings of the 37th International ACM SIGIR Conference on Research and Development in Information Retrieval
(2014) - et al.
Automatic construction of domain and aspect specific sentiment lexicons for customer review mining
Proceedings of the 22nd ACM International Conference on Information and Knowledge Management
(2013) - et al.
Thumbs up?: sentiment classification using machine lsearning techniques
Proceedings of the ACL-02 Conference on Empirical Methods in Natural Language Processing
(2002) - et al.
Suit: a supervised user-item based topic model for sentiment analysis
Proceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence
(2014) - et al.
Unsupervised sentiment analysis with emotional signals
Proceedings of the 22nd International Conference on World Wide Web
(2013)
Coarse-to-fine review selection via supervised joint aspect and sentiment model
Proceedings of the 37th International ACM SIGIR Conference on Research and Development in Information Retrieval
Jointly discovering fine-grained and coarse-grained sentiments via topic modeling
Proceedings of the 22nd ACM international conference on Multimedia
Twitter sentiment analysis with deep convolutional neural networks
Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval
Cited by (17)
How emotional communication happens in social media: Predicting “Arousal-Homophily-Echo” emotional communication with multi-dimensional features
2022, Telematics and Informatics ReportsA Survey on Information Diffusion over Social Network with the Application on Stock Market and its Future Prospects
2023, Wireless Personal CommunicationsApplication of Promotion Process Based on Epidemic Models Considering Bidirectionality
2022, Wuhan University Journal of Natural Sciences