Mining User Opinions in Mobile App Reviews: A Keyword-based Approach

To improve user satisfaction, mobile app developers are interested in relevant user opinions such as complaints or suggestions. An important source for such opinions is user reviews on online app markets. However, manual review analysis for useful opinions is often challenging due to the large amount and the noisy-nature of user reviews. To address this problem, we propose M.A.R.K, a keyword-based framework for semiautomated review analysis. The key task of M.A.R.K is to analyze reviews for keywords of potential interest which developers can use to search for useful opinions. We have developed several techniques for that task including: 1) keyword extracting with customized regularization algorithms; 2) keyword grouping with distributed representation; and 3) keyword ranking with ratings and frequencies analysis. Our empirical evaluation and case studies show that M.A.R.K can identify keywords of high interest and provide developers with useful user opinions.