Topic Detection In Social Media And Future Influence Analysis In Scientific Articles Based On Topic Model

Web2.0 and Internet technology maturity to promote user-generated content has gradually become a new way for users to use the Internet.Users and creator of Internet resources,people and Internet interactive mode has been sublimated.People tend to share original personalized suggestions on the web platform,and opinion leaders,experts,and so on.They are willing to share professional content and contribute wisdom to related areas.For example,ordinary users usually in Twitter and other social media platform to share their own lives.Experts will publish scientific research literature in the academic platform for learning and reading.Both of them is the text,but in the text mining methods and application exploration is very different.The challenge of research is how to find the information needed by different users from the massive data efficiently and accurately.The main work of this paper is to use the topic model to improve the method of social media short text topic mining and the prediction of literature in the future.The main idea of the topic model is to explore the relationship between the topics and words,and the relationship between the document and the topic,or to guide the results of the model by means of the potential topic of the textual content.Based on different scenarios to construct a suitable theme model to a certain extent,we achieve different purpose in data mining.The length of Twitter text is short,sparse,non-standard language and other characteristics lead to the traditional LDA,PLSA on this text environment cannot be effectively excavated.It is worth mentioning that the semantic analysis method introduced into the literature mining in this paper is a novel and challenging idea,compared with the traditional method of using the statistical method of reference.In order to study the topic mining in the short text environment of social media,this paper proposes a new topic model HTTM,which uses the time and tag information in Twitter message(tweets)to add a new "label-subject-Time " level to improve the expression of the topic,push the effect of clustering and theme in the time series under the evolution of the effect.The final experimental results show that the HTTM model has some validity in the above aspects.In order to predict the influence of literature,this paper proposes a TTRM model to predict the future influence of the literature.The model is based on the article feature word / pairs of links,respectively,the time of publication of the literature and the content of the article will be innovative and important modeling.Topic model is used in the process of modeling the importance of the articles among the current literature on the importance degree.In the experiment,the validity of the TTRM model in the literature sorting and influence prediction was confirmed by using the literature data set.In contrast,the use of the reference PageRank model,and TF-IDF as the article importance modeling method of the MRR-ranking model for comparison.Results show that TTRM in the literature rankings and literature influence prediction have a certain advantage.And it is a good proof of our hypothesis,some of the contents of the literature for the contribution of innovation in the article,while the content of the article on the basis of the literature analysis of the conclusions will be more accurate.