A Review Analysis Method Guided By App Descriptions

App store provides a massive and fast-growing data resource for the development of mobile application.How to mine the information effectively based on the demand of developers has become a hotspot research question.As an important kind of data,reviews is a main way for developers to understand the needs of users.Among many App stores,Google Play Store is the most representative App store.In Play Store,a variety of mobile phone applications are classified,and both the description of the developer for the mobile phone application,and the comments of the users on the mobile phone application are collected.In addition,the downloads of mobile apps is over ten million.With the downloading of mobile phone applications,the quantity of user reviews grow rapidly,which makes the traditional way by using manual analysis can't handle these review analysis problems.At the same time,the developers of mobile applications also want to obtain the useful information in the comments related to App features quickly,in order to support the rapid iterative development of App.In past methods,data analysts only focus on review analysis or description analysis and there is no method to combine them for analysis.In order to utilize the reviews reasonably to support the development and evolution of App products,this paper introduces App descriptions into the process of review analysis.That is,we get the features of a certain product domain by mining the description of App products and use text classification technology to associate App reviews and App features.Then the result is quantified to provide more effective help information to App developers.And the author proposes an description guided review analysis and recommend method:Firstly,topic analysis technology is used to obtain the high-level App features from descriptions,and the results are formalized by Topic-Based Domain Model(TBDM);Secondly,review classifiers are trained based on the model and the sentiment of users in the reviews are mined to transform the reviews to a group of quantified attributes,in this way,the relationship between objective features and subjective user's feeling is established;Finally,quantitative analysis of model is proceeded based on developer preference,so that the reviews are sorted and summarized for information recommendation.Combined with App description and user reviews,we provide information for developers to support their product development and evolution.Meanwhile,to evaluate our approach,a serial of experiments are conducted using the data in Google Play Store.The average F-measure of the classifier we trained exceeds by 80%.And through artificial evaluation,the results show that our approach can obtain the sentiment information from reviews effectively and recommend them according to the demand of developers.