Graph Convolutional Neural Networks For Network Representation Learning And Recommendation System

Graph Convolutional Neural Network is a discipline that based on irregular or non-Euclidean data to conduct forecast and analysis.Data science is developing rapidly today,graph convolutional neural network has lots of applications in all walks of life.The main content of this paper could be divided into two parts:1.We propose a new graph convolutional neural architecture based on a depth-based representation of graph structure,called the depth-based subgraph convo-lutional neural networks(DSCNN),which integrates both the global topological and local connectivity structures within a graph.Our idea is to decompose a graph into a family of K-layer expansion subgraphs rooted at each vertex,and then a set of convolution filters are designed over these subgraphs to capture local connectivity structural information.Specifically,we commence by establishing a family of K-layer expansion subgraphs for each vertex of graph,which can pro-vide global topological arrangement information contained within a graph.We then design a set of fixed-size convolution filters to capture the local structural information within the graph,and has the weight sharing property among differ-ent positions of subgraph;the pooling operation acts directly on the output of the preceding layer without any preprocessing scheme(e.g.,clustering or other techniques).Experiments on three graph-structured datasets demonstrate that our model DS-CNNs are able to outperform six state-of-the-art methods at the tasks of node classification.2.Item-based collaborative filtering(CF)has been heavily used for building recommendation systems in industrial applications.However,most existing ap-proaches attempt to enhance the representation ability of user's profile with her historically interaction,there has been relatively less work that focus on rep-resentation enhancements of the item.Here we propose Tree-based Subgraph Convolutional Neural Networks(TSCNN)for item-based CF.TSCNN com'bines efficient adamic-pooling and convolution operation,which could aggregate infor-mation from the item's local neighborhood graph and transform them to enhance the item's representation ability.Compared to prior graph convolutional net-work approaches,we develop a novel adamic-pooling operation on highly efficient Adamic\Adar measures to suppress the weight of active users while aggregating information from the item's local neighborhood graph(item-to-item)that based on user-to-item interaction.Experiments on a public dataset and an Alibaba real dataset demonstrate the effectiveness of proposed approaches,which achieve su-perior performance compared with state-of-the-art methods.This work opening up a new research possibility for the future development of the graph convolution neural networks in the recommendation system.