DEFECTS PREDICTION IN APPS USING USER REVIEWS AND RATINGS

Abstract

In the current digital era approximately 2 million applications (a.k.a. apps) are present on app store which allow users to give ratings and reviews. App developers face serious challenges in getting user feedback. Every app developer is in constant dilemma of DIPMAP: Did I program a poor mobile app? The app developer constantly strives for eliminating the defects to increase the user base and app rating. The app developer wants to exploit the expressive power of raw user reviews regarding issues faced by app users while using the app. But with the sheer volume of these raw reviews a lot of knowledge goes untapped which is useful for app developers. We propose an unsupervised novel model for defect prediction using app reviews by (i) review preprocessing (ii) Making Vector Representations of reviews (iii) classifying review into broad classes (iv) Making prioritized defect phrases.