Research On Users' Emotional Information Diffusion Based On Online Social Networks

With the rapid development of Internet technology and the popularity of online social networks,there is a lot of user-generated information with emotional opinions on social media.Users can influence the people around them by publishing,retweeting,commenting,thumbing up,sharing messages and so on.Thus,influential users can drive interpersonal information diffusion on social networks.Analyzing whether the information among users can be transmitted on different sentiments is a key issue for opinion formation and viral marketing.Previous works directly defined interpersonal information diffusion on each pair of users.They fail to depict the unobserved relationships between user pairs when they have the same influence originator or receiver.As a result,these methods suffer from the overfitting problem of learning propagation probability between users.In addition,there are still no effective solutions to integrate users' sentiments to estimate the propagation probability between users currently.Therefore,aiming at the above problems,the details of the three research works are as followed:First of all,for the existing related works,the propagation probability is defined on the edges of the network,and it may lead to the overfitting problem of learning parameters for the unobserved user pairs.Aim at this problem,we propose a users'information diffusion model with emotional factors.The model assumes that the main factors of users' information diffusion are the influences of propagators and the susceptibilities of recipients,defining two parameter matrices to represent them on different sentiments separately.And we utilize survival analysis model and the cascades formed by emotional posts during forwarded process to model propagation probability between users.Experimental results demonstrate that our model not only consistently outperforms existing pair-wise methods on the evaluation metrics of each task,but also reduces the model complexity effectively.Secondly,the imbalance between positive and negative cases on observed cascades may mislead the optimization direction of the users' information diffusion model with emotional factors and restrict the scalability of the model on a large dataset.To solve this issue,we propose an emotional information diffusion model with negative sampling algorithm.In this model,probabilistic sampling is repeated in each iteration according to the frequencies of negative cases in the dataset to keep the balance of positive and negative cases in the optimization process.Compared with the previous model,experimental results show that the method not only has a big improvement on the MRR metrics on "Predicting Cascade Dynamics" and "Who will Be Retweeted" tasks,but also can effectively depict different influences and susceptibilities on different sentimental polarities.At last,the topics of users' emotional information are also a non-ignorable factor of influencing the propagation probability.In this paper,we propose a users'emotional information diffusion model with topic factors based on the previous method.The model uses LDA model to learn documents'topic distributions,and the topic distributions are integrated into the users' influence and susceptibility matrices to modulate interpersonal propagation probability.And then predict the users'forwarding behavior of different topic and emotional messages on hidden networks.Experimental results show that the proposed method not only achieves good performance on different evaluation tasks,but also can describe the trend of information dissemination better,which verifies the validity of our model.