Abstract： In recent years, with the popularization of mobile phones and mobile Internet, the APP market has developed vigorously. However, due to its widespread nature, it has gradually become a “severe disaster area” for illegal crimes. A large number of counterfeit APPs, gambling and fraud APPs constitute a great impact on social security threat. In order to identify counterfeit APPs, gambling frauds and other illegal APPs more quickly and effectively, this paper collects a large number of APPs from Internet, and proposes a similar recommendation model for illegal APPs based on image deep learning algorithms, text analysis algorithms, and index evaluation algorithms. This model conducts similar comprehensive evaluation on the 6 dimensions of APP icon, screenshot, name, server IP, framework structure and developer SHA1 signature. From more than 300 000 APPs, 5 197 APPs are found to be similar, and 55 965 similar APPs are associated. Based on the abnormal APP library, 8 233 abnormal APPs are recommended.
Key words : APP similar；yolov4 algorithm；entropy method；deep learning
