Detecting User Significant Intention via Sentiment-Preference Correlation Analysis for Continuous App Improvement

Jianmao Xiao, Shizhan Chen, Qiang He, Hongyue Wu, Zhiyong Feng, Xiao Xue

Abstract: Detecting users’ significant intentions (e.g., new features wanted) timely and precisely is crucial for developers to update and maintain their apps in the competitive mobile app market. Sentiment and preference mining from crowd reviews provide an opportunity to proactively collect app users’ intentions, e.g., bug fixing and feature refinement. However, users’ sentiment and preferences often change over time due to either internal factors (e.g., new bugs) or external factors (e.g., new competitors). This makes it difficult for app developers to grasp users’ sentiment and preferences in time. In this paper, we propose a novel and automated framework named DSISP for detecting users’ significant intentions effectively via sentiment-preference correlation analysis. DSISP first employs sentiment analysis and NLP (Natural Language Processing) techniques to obtain sentence-level sentiment scores and fine-grained user preference features from app reviews in different time slices. Then, the temporal correlation between user sentiment and preferences is analyzed, which can be used to monitor users’ sentiment tendency and preferences in time. Finally, DSISP identifies users’ dramatically-changing sentiment (e.g., sentiment valley) to detect users’ significant intentions. We evaluate the feasibility and performance of DSISP by using real-world app reviews and app official changelogs. The experimental results show that DSISP can detect users’ significant intentions effectively and efficiently, with a precision of 0.962 on average. It can help app developers keep track of how their users’ intentions evolve over time so that they can improve their apps correspondingly and continuously.