Design And Implementation Of GNN For Recommendation Algorithm

With the development of the Internet,information overload has become a challenge for people to find the information that they are interested in.The recommendation system as an information filtering system that can learn the user's interests and hobbies based on the user's profile or historical behavior records,greatly reduces the time for users to filter information,which is of great help to improve user experience and alleviate the problem of information overload.Traditional recommendation methods usually rely on direct user-item interactions while neglect the complex graph structure of user-user interactions,item-item interactions etc.,and also always ignore the sequential information of interactions.Graph neural network performs convolution directly on graph structure by aggregating information of neighboring nodes,which can be applied to non-Euclidean fields such as graphs to effectively capture the dependency relationship and higher-order representation between nodes.Therefore,it has been widely used in recommendation scenarios in recent years.Hence,this paper proposes a series of recommendation algorithms based on graph neural network,applying graph neural network to learn more informative user and item representations for recommendation,consisting of two research work and one practical application.To be more specific,this paper first studies the heterogeneous graph neural network recommendation algorithm,and proposes graph neural news recommendation with long-term and short-term interest modeling(GNewsRec)algorithm.Secondly,this paper studies the sequential recommendation on heterogeneous information network(HIN)settings,and proposes the sequence-aware heterogeneous graph neural collaborative filtering(SHCF)algorithm.Finally,this paper studies the package recommendation problem under the "TopStories" recommendation scenario of WeChat with the real industrial data,and proposes the intra-and inter-package attention networks(IPRec)model to verify the effectiveness of the graph neural network recommendation algorithm in the real industrial recommendation scenario.The graph in real application scenario always contains multiple types of node and relationships,i.e.heterogeneous information network(also called heterogeneous graph).Traditional news recommendation algorithm such as collaborative filtering methods that rely on direct user-item interactions or content based recommendation methods that based on the content of user history interactions are difficult to capture the high order information.While heterogeneous graph neural network model the complex topology structure with the attribute of heterogeneous information network,can capture more informative semantic and structural information,which promotes the recommendation performance.In this paper,we first model the news recommendation problem to a user-news-topic heterogeneous graph and propose a novel Graph neural News Recommendation with long-term and short-term interest modeling(GNewsRec)algorithm.Experimental results on real world datasets show that our proposed model significantly outperforms state-of-the-art methods on news recommendation.Moreover,traditional heterogeneous information network based methods or sequential recommendation methods either consider only heterogeneous collaborative signals in the interactions or model user embedding based on only their own item interaction sequence,which either can hardly capture a user's dynamic preferences or face a common data sparsity problem.In this paper,we propose a novel Sequence-aware Heterogeneous graph neural Collaborative Filtering model,called SHCF,which can address the above problems by considering both the high-order heterogeneous collaborative signals and sequential information.Specifically,we first construct a heterogeneous information network(HIN)by enriching the user-item bipartite graph with additional attribute information,and then design novel message passing layers for learning user and item embedding.For user embedding,we consider the sequential information to capture user's dynamic interests over time with a position-aware self-attention mechanism,and capture user's fine-grained static preferences on different aspects of an item with an element-wise attention mechanism.For item embedding,we carefully incorporate the heterogeneous attribute information with dual-level attention,which alleviates the data sparsity problem.Extensive experiments on three real-world datasets illustrate that our model can improve the recommendation performance compared with the state-of-the-art methods.Finally,to verify the effectiveness of the graph neural network recommendation algorithm in the real industrial recommendation scenario,this paper studies the package recommendation problem under the "TopStories" recommendation scenario of WeChat and propose an Intra-and inter-package attention network for Package Recommendation(IPRec).Specifically,for package modeling,an intra-package attention network is put forward to capture the object-level intention of user interacting with the package,while an inter-package attention network acts as a package-level information encoder that captures collaborative features of neighboring packages.In addition,to capture user's preference representation,we present a user preference learner equipped with a fine-grained feature aggregation network and coarse-grained package aggregation network.Extensive experiments on three real-world datasets demonstrate that IPRec significantly outperforms the state of the arts.Moreover,the model analysis demonstrates the interpretability of our IPRec and the characteristics of user behaviors.