Aiming at the efficiency of user behavior big data mining under Hadoop framework, an improved Apriori mining algorithm for association rules is proposed in this paper. Firstly, the modeling of item sets classification under Hadoop framework is realized. Then through the analysis of the mining steps of the traditional association rule Apriori algorithm, the generation method of the candidate item set is improved, and the useless information is removed with the flag information, which could effectively reduce the number of transactions and projects, thereby shorten the task processing time. In the specific implementation process, Map Reduce processing is performed on the improved Apriori algorithm flow. Simulation experiments show that the improved Apriori mining algorithm has higher execution efficiency than that of the traditional Apriori algorithm.