Mining user requirements from application store reviews using frame semantics

Abstract

Context and motivation: Research on mining user reviews in mobile application (app) stores has noticeably advanced in the past few years. The majority of the proposed techniques rely on classifying the textual description of user reviews into different categories of technically informative user requirements and uninformative feedback. Question/Problem: Relying on the textual attributes of reviews often produces high dimensional models. This increases the complexity of the classifier and can lead to overfitting problems. Principal ideas/results: We propose a novel semantic approach for app review classification. The proposed approach is based on the notion of semantic role labeling, or characterizing the lexical meaning of text in terms of semantic frames. Semantic frames help to generalize from text (individual words) to more abstract scenarios (contexts). This reduces the dimensionality of the data and enhances the predictive capabilities of the classifier. Three datasets of user reviews are used to conduct our experimental analysis. Results show that semantic frames can be used to generate lower dimensional and more accurate models in comparison to text classification methods. Contribution: A novel semantic approach for extracting user requirements from app reviews. The proposed approach enables a more efficient classification process and reduces the chance of overfitting.