Recommendations Based On LDA Topic Model In Android Applications

With the popularity of smart phones, mobile applications have become an essential element in people’s lives. Wherever you are, the Android market can allow people to download these applications. When you open Android mobile phone market in a mobile application program interface, we can see not only the application content presentation but also other information. The user can also see the scoring for this application, comments and so on. The text describes the application content information and user reviews, no more than 200 words, the contents of which are closely related to mobile applications. How to use these text data to obtain valuable information for the users has been an important research field of data mining, it is also the focus of this study.The topics of different mobile applications are the summary of their contents. This generalization to some extent reflects the different mobile application content core idea. Therefore mobile applications relating to mining user interest analysis has important significance, the results relating to mining can provide data support for topic-based personalized recommendation applications. This article extracts the contents and users’ descriptions from a real Android Market dataset, and builds a new topic model called combine LDA to analyze different topics of each mobile application. By combine LDA model we can analyze the topic probability distribution of each mobile application, and then we can calculate the similarity and recommend to users with high similarity applications.In addition, we use perplexity as a criterion. We compare the combine LDA topic model with the topic model only uses the users’ descriptions. Experimental results show that, the perplexity of combine LDA model is lower than the ordinary LDA. combine LDA model also has fewer iterations. These results show the superiority of combine LDA. The main contributions are as follows:1) According to the information characteristics in the Android phone market, use LDA to process these text data;2) We combine the contents of mobile applications and users’ descriptions information to construct new combine LDA model, makes the model structure more reasonable;3) Data acquisition from the real Android market, combine LDA model can achieve good results on these real data sets. The model can analyze the topic distribution of different mobile applications, and the results are applied to calculate the similarity and personalized recommendations.