A Data-driven Approach for Mining Software Features based on Similar App Descriptions and User Reviews Analysis

Abstract

Mobile app development necessitates extracting domain-specific, essential, and innovative features that align with user needs and market trends. Determining which features provide a competitive advantage is a complex task, often managed manually by product managers. This study addresses the challenge of automating feature mining and recommendation by identifying similar apps based on user-provided descriptions. The proposed approach integrates Named Entity Recognition (NER) for feature extraction from mined Google Play app data with BERT (Bidirectional Encoder Representations from Transformers) and Topic Modeling to find comparable apps. Our top-performing model, which uses Non-negative Matrix Factorization (NMF) for Topic Modeling with Sentence-BERT (SBERT) embeddings, achieves an F1 score of 87.38%.