Research on user behavior clustering algorithm based on mobile application

Abstract

In mobile applications, user behavior clustering analysis divides users into different groups so that operators can provide personalized services, optimize page organization and functional design. On the basis of clickstream and custom event, this paper first proposed two collection schemes and formats of user behavior data, then a two-layer clustering algorithm and a fuzzy clustering algorithm are proposed respectively. The former first-layer clustering adopts an improved DBSCAN algorithm, which replacesthe neighborhood radius with a newly-designed user session similarity and has optimized the merge condition of clusters. While, the latter directly extracts feature vectors of user sessions and use an improved FCM algorithm, where an effective method is used to initialize membership matrix to accelerate the convergence speed and giving a weighted value to membership matrix solves the problem of local optimum. Experimental results show that both algorithms are effective and matches the advance assigned user test distributions, which proves that the two clustering methods are applied to different analysis scenarios, depending on the business requirement of developers.