Automating App Review Classification based on Extended Semantic

Abstract

Automatic classification of app reviews can help developers quickly read reviews to identify and fix software bugs or add new software functions to meet user requirements. Text mining technologies have been widely used in reviews classification in recent years. However, the accuracy of app reviews classification is limited because of the generally short length of reviews and limited information, and classification models are prone to overfitting due to the diversity and unstructured characteristics of app reviews. In this paper, we propose an automatic classification approach for app reviews based on extended semantic. Specifically, we first reduce noisy data in app reviews by preprocessing, and annotate app reviews using frame semantics and splice the annotation results with reviews to extend the semantic information and text length of reviews. Then, to reduce the probability of overfitting, we integrate the pre-trained models to learn the semantic information of extended app reviews and classify reviews. We evaluate the effectiveness of proposed approach in multiple popular apps, and the experimental results show that it outperforms the state-of-art baselines.