Aiming at the accuracy of network resource recommendation, the complex behavior of network users is studied. Through front-end embedded JS code and Java Web Framework custom filter, multi-data source acquisition is realized, and data acquisition is preprocessed. Complex behavior is normalized by entropy weight method. Flume+Kafka+Hbase are used to provide reliable high-speed storage and query, and complex behavior of network users is obtained, which is precisely recommended for network resources. Provide reliable and fast service. The complex behavior analysis method is applied to the network resource recommendation system. After testing, the system supports multi-channel fast and stable acquisition, reliable storage of millions of data, millisecond query speed, high quality of analysis results, and effectively improves the network resource recommendation in complex behavior without affecting the existing business logic. Real time and accuracy.