In view of the repeated scanning of the database and the potential massive candidate sets involved in the Apriori algorithm, an improved method, with the compressed matrix and the transaction value introduced in the process, is proposed to solve such problems as the timeliness and accuracy of the analysis of network user behaviors, with a further application of the improved algorithm to Spark, a cloud computing platform. The experimental results verify the better performance and higher efficiency of the proposed method, with evident advantages in the user behavior analysis.