Research On APP Recommendation Algorithms Based On Clustering Of Heterogeneous Information Network

Mobile Applications (APPs) markets use personalized recommendation systems to help users find content of interest from the mass of the APPs. However, due to the characteristics of the APP field, APP recommendation algorithm exists a range of issues. For example, the use of the distribution of APP is prone to heavy head and long tail phenomenon, data sparseness problem, cold start problems and so on. Heavy head and long tail phenomenon will hinder the development of APP. Data sparseness problem is limiting the efficiency of the recommendation algorithm, and with the growing number of APP, these issues will become increasingly prominent.The emergence of the Netflix competition has greatly promoted the development of the field of recommendation, but the research of APP recommendation algorithm is still at the beginning stage. At present, the APP recommendation algorithm of mobile application market is mainly concentrated in the relevance recommendation, the popular recommendation and the new product recommendation, etc. These traditional methods do not solve the problem of the APP recommendation. With the increasing number of APP market, the APP market will have more APP information and user’s information, how to use the information to help solve a series of problems faced by APP is very important.According to the characteristics of the APP data set, this paper proposes a method combining sorting, clustering technology and recommendation algorithm. Firstly, constructed APP heterogeneous information network, and then using two sorting algorithm to obtain the Sort distribution of subsidiary type of object. Second, create a mixed probability generation model for center type on the basis of sort distribution and using the EM algorithm to estimate the optimal value of the parameter. And using bayes1 theorem to obtain the posteriori probability of objects. Thirdly, based on the the clustering distribution of object to reclassify these objects. Sorting and clustering is an iterative process of calculation, until clustering results converge. Finally carry out two different collaborative filtering algorithm based on the APP and user clustering results, which is based on a pseudo-score IBCF (Iterm-Based Collaborative Filtering) algorithm and time-based recession UBCF (User-Based Collaborative Filtering) algorithm. In this paper, experimental analysis of data sets are from 360 Mobile Assistant Application market. The experimental results show that the proposed algorithm is able to improve the actual effect of APP recommendation.