A Multi-Layer Framework for Detecting Fake Reviews and Improving App Rating Integrity

User ratings and reviews are essential for app recommendations but are increasingly undermined by fake and manipulated feedback. This paper presents RCAR (Rating-Corrected App Recommendation), a multi-layer framework designed to detect review fraud and improve rating accuracy. RCAR combines sentiment–rating alignment, comment similarity, graph-based tie strength, and community detection to assess reviewer credibility and correct app scores. Evaluated on over 100,000 reviews, RCAR achieves an AUC-PR of 0.94 in identifying genuine reviewers and reduces fake review influence by 40%. It also improves rating accuracy, notably increasing LINE’s score from 2.15 to 3.35 and decreasing VK Video’s from 2.64 to 1.44. These results demonstrate RCAR’s effectiveness in enhancing app store transparency and recommendation quality.