The Building Method Of Auto-adapt Context Based Topic Model

In Internet era, through Internet people can quickly convey information. With the advent of the mobile Internet era, people can publish information through microblogging and other social media platforms anytime. Learning topics in microblogging text is not only very important to master the Internet public opinion, Internet users' life, but also in recommendation system, business advertising and other Internet services industry applications. In Internet era, a great feature of these social media of user-generated content is the text short and the content can only express little information, compared to the traditional media industry such as newspapers, magazines, etc., the information expressive of social media in Internet era, relatively with inadequate content.Based on the traditional topic model, the arbitrary contexts introduce to build the Auto-adapt Context Based Topic model(aLDA). aLDA model can auto adapt the context types of documents, making aLDA topic model is applied to different document sets need not to change the structure of the topic model, and enhanced the expansibility a LDA model.ALDA in parameter estimation using iterative solution method, use matrix of demensioin reduction for the initial text-topic matrix, selected most relevant topic for the documents,then use the selected topics substitution model parameter estimation again and model parameters estimation is required to be solved many times, so as to make the parameter estimation with higher precision, and also reduce the complexity of topic model to improve the estimation accuracy. Data used in this paper, the experiment is part of Sina Weibo microblogging text data and parts papers from CNKI. After screening with no income topic model context class results comparison shows that the proposed method on topic model to extract the topics of more efficient, especially when dealing with less expressive content of the text made the text more satisfactory results.