Mobile APP Iteration Planning Decision Support Based On User Reviews

Mobile internet is now one of the most promising and the most attractive areas,consequently mobile Apps are gaining more and more attention.Confronted with increasing market competition,developers need to develop an iterative release plan for their Apps.The mobile App store uses a user-driven quality assessment mechanism,and this makes it easy for Apps that satisfy users to be replaced and eliminated by their competitors,so the Apps need to respond to user needs through successive releases.User reviews provided by the App store are reliable resource of valuable user needs,however,the fast iterative,continuous delivery agile development model leaves developers with insufficient time to mine user requirements from numerous and varying-quality reviews,adjust development plans based on mined user requirements,and fully test new versions.In response to the above problems,this paper proposes a series of algorithms to help the developer's with their App development process.We manage to mine App features as well as its user reaction trend from user reviews and release log,and to identify the interaction between user reviews and developer update behavior,and on this basis,we are able to recommend release time and update content for the next version and to predict user reaction on a planned version.The main research content and contributions are as follows:(1)Propose an App feature extraction method.We modify RAKE(Rapid Automatic Keyword Extraction)and combine it with Word2 Vec and K-Means to mine App features from user reviews and release log,and generate user reaction trend for every App feature.(2)Introduce a Naive-Bayes-classifier based approach for recommending release time.We set three category for release interval base on the cumulative distribution of release intervals of all Apps,and solve the recommendation problem by means of Gaussian Naive Bayesian classification algorithm,and the average accuracy can reach 72.8%.The empirical study verifies the stability of thealgorithm from two aspects: the average prediction accuracy of each category is relatively close,and the consecutive prediction results between two consecutive version are consistent,besides,the change of prediction results are consistent with general intuition.(3)Introduce a method for recommending update content based on horizontal comparison and vertical history.As for horizontal comparison,we solve the recommendation problem via Support Vector Machine classification algorithm,and the average accuracy can reach 93.7%.The experiment result proves that our method is stable because the consecutive prediction results between two consecutive version are stable.For vertical history,we calculate the “user reaction lag” of a new version by Pearson correlation coefficient,and based on which we propose an algorithm for dynamically identifying similar high-rating App,and eventually recommend update content according to the update time relevance of common App features.(4)Propose a Logistic-Regression-classifier based approach for predicting user reaction on a new version to be released.We apply Logistic Regression algorithm to predict the range of median user satisfaction during the “observation period”,and the “observation period” is figured out in the light of “user reaction lag” and average release interval.The average prediction accuracy of the three dimensions(review intensity,positive sentiment score and negative sentiment score)of user reaction is 89.8%,89.0%,and 96.4%,respectively.Although the prediction accuracy is quite good on the whole,the average prediction accuracy of different category varies a lot,therefore the algorithm still has room for improvement in stability.