Research On Key Technologies Of Content Recommendation Services In Mobile Networks

With the rapid development of mobile networks,people have the opportunity to access more and more multimedia applications every day.Large-scale,explosive information usually makes users difficult to choose.As an advantageous tool for screening data,content recommendation services have been widely used in mobile networks.In the relatively complex network environment of mobile networks,how to recommend the most appropriate relevant information for users based on heterogeneous content and sequential user-content interaction behavior has become a research hotspot in the recommendation field.Information heterogeneity refers to the richer sources and types of information.Since heterogeneous information networks can effectively contain comprehensive structural information and rich semantic information,some works have used heterogeneous information networks to model the relationship between different types of nodes in the recommendation system.However,most of the existing recommendation methods based on heterogeneous information networks only focus on the representation learning process in heterogeneous information networks,but ignore the importance of heterogeneous edge learning.Moreover,large-scale network representation learning algorithms are currently difficult to apply to mobile networks with huge data scale and requiring timely response to user needs.In addition to the feature of information heterogeneity,content recommendation services under mobile networks also have a large number of user-item interactions that occur sequentially over time,that is,sequential interaction data.Among them,the session is a typical sequential data,which is widely used in mobile network applications.Session recommendation services often need to identify the user's intentions through these sequential user-item interactions,so as to recommend relevant content for them.However,the choices of users in mobile networks are flexible and changeable.It is very difficult to predict users' personalized preferences through limited user information and uncertain user behaviors.Therefore,in view of the two important feature of information heterogeneity and sequential user-item interactions in content recommendation services under mobile networks,this paper studies the recommendation algorithm for using heterogeneous information and session.The specific work is as follows:(1)Aiming at the heterogeneity of content in mobile networks,this paper proposes a recommendation algorithm based on heterogeneous information networks,called HIN-MRS.This algorithm takes into account the user's preferences and the heterogeneous relationship of related content,and can recommend more appropriate and interpretable content for users.First,this paper uses the obtained content-related text to mine the user's content preferences.Then,a small-scale heterogeneous information network is constructed according to the user's preference topic.And on this basis,a graph algorithm that can automatically learn the heterogeneous relationship in the network is designed to generate recommendation results.Finally,the experimental results based on real data sets verify the interpretability,accuracy and effectiveness of the proposed algorithm in solving cold start problems.(2)Aiming at the sequential feature of user-content interaction in mobile networks,this paper proposes a novel multi-aspect aware session-based recommendation algorithm called MASR.This model comprehensively considers many factors such as users' sequential interaction behaviors,item feature,users' current interests,and users' long-term preferences.It also uses a self-attention mechanism to capture the changing laws of the sequence information of the current session,which makes it more accurate to predict the user's next behavior in an anonymous session.Compared with previous models that only focus on sparse sequential data,the MASR considers more finegrained behaviors of users other than clicking.Finally,the experimental results based on real data sets show that the proposed MASR recommendation results are more accurate and the training speed is faster.