The heavily fragmented Android ecosystem has induced various compatibility issues in Android apps. The search space for such fragmentation-induced compatibility issues (FIC issues) is huge, comprising three dimensions: device models, Android OS versions, and Android APIs. FIC issues, especially those arising from devices, evolve quickly with the frequent release of new device models to the market. As a result, an automated technique is desired to maintain timely knowledge of such FIC issues, which are mostly undocumented. In this paper, we propose such a technique, PIVOT, that automatically learns API-device correlations of FIC issues from existing Android apps. PIVOT extracts and prioritizes API-device correlations from a given corpus of Android apps. We evaluated PIVOT with popular Android apps on Google Play. Evaluation results show that PIVOT can effectively prioritize valid API-device correlations for app corpora collected at different time. Leveraging the knowledge in the learned API-device correlations, we further conducted a case study and successfully uncovered ten previously-undetected FIC issues in open-source Android apps.