Latent Hot Topic Mining With Propagation Models On Social Network

With the rise of mobile internet, social networking platforms are increasingly being well known and used. People communicate each other on social platforms, which would produce a variety of topics. The paper mainly studies latent hot topics prediction on social networks based on propagation models. We first designed a novel topic prediction model to predict latent hot topics. Then we provided two point sampling methods on social networks to save space and time of the above model.Topic mining based on social network has been an important research field. Ignoring that the structure of social network plays an important role in topic mining, a lot of studies focus on topic mining from textual documents. Few studies are paying attention to topic propagation in the network, but they don’t make full use of abundant attribute information in social networks. Moreover, none of them studied the prediction of latent hot topic on social network. Here we design a novel algorithm for modeling propagation. To the best of our knowledge, the model is the first one to study Latent Hot Topic Prediction(LHTP) with propagation models in the social network. It provided confident constraints of over fitting for hot topic evaluation criteria, and based on it we provided the extended model LHTPEX which could reduce the runtime with ensuring the same effect. Different from the ones based on heterogeneous information network, our model concentrates on rich labels of heterogeneous network, which is proved to be more effective. We apply the model on real data sets. Comparing with TMBP, the result suggests that our proposed approach is more accurate and effective.At the same time, topic models would spent more time and more storage space with the increase of the size of data when topics are spread in the social network. Howerver, most topics focus on some key nodes and parts of nodes have no significant effect on topic propagation in the real process of topic propagation. If we could reasonably cut some nodes in the social network during the spread of topics, the runtime of the program and the storage space both would be reduced. To solve the above problem, we designed two novel point sampling methods to reduce the number of nodes in the social network. The two methods presented in this paper introduced the thought of recommend system into the research on sampling methods of topic propagation models and have a certain novelty.With the experimental analysis, we analysis the impact of different sampling methods of propagation model on the effectiveness, the space, running time and the robustness of the graph.