A review of recommendation system research based on bipartite graph

MATEC Web of Conferences 336, 05010 (2021) CSCNS2020 https://doi.org/10.1051/matecconf/202133605010 A review of recommendation system research based on bipartite graph Ziteng Wu. *, Chengyun Song, Yunqing Chen, and Lingxuan Li School of Computer Science and Engineering, Chongqing University of Technology, Chongqing, China Abstract. The interaction history between users and items is usually stored and displayed in the form of bipartite graphs. Neural network recommendation based on the user-item bipartite graph has a significant effect on alleviating the long-standing data sparseness and cold start of the recommendation system. The whole paper is based on the bipartite graph. An review of the recommendation system of graphs summarizes the three characteristics of graph neural network processing bipartite graph data in the recommendation field: interchangeability, Multi-hop transportability, and strong interpretability. The biggest contribution of the full paper is that it summarizes the general framework of graph neural network processing bipartite graph recommendation from the models with the best recommendation effect in the past three years: embedding layer, propagation update layer, and prediction layer. Although there are subtle differences between different models, they are all this framework can be applied, and different models can be regarded as variants of this general model, that is, other models are fine-tuned on the basis of this framework. At the end of the paper, the latest research progress is introduced, and the main challenges and research priorities that will be faced in the future are pointed out. 1 Introduction With the rapid development of information network technology, data information has exponentially increased [1]. If users want to dig out effective information from a large amount of information, they need to use recommendation tools. Recommendation technology is an effective method of information screening. The above alleviated the problem of data overload [2]. The core of the recommendation system is the recommendation algorithm, which constructs a preference model by analyzing user behavior information, user portraits, and item attributes, and pushes the items that best match the user’s interest in the network to users, so that users can get rid of being surrounded by spam. The dilemma of not being able to find the target data increases the user's dependence and experience. At present, the application of the recommendation system is reflected in all aspects of life, Taobao’s guess you like it, Douyin’s video * Corresponding author: © The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (http://creativecommons.org/licenses/by/4.0/). MATEC Web of Conferences 336, 05010 (2021) CSCNS2020 https://doi.org/10.1051/matecconf/202133605010 recommendation, QQ Music’s daily playlist, Weibo’s hot search list and WeChat look, etc. It can be said that life is more colorful due to the existence of the recommendation system. Data in the field of machine learning is generally divided into European data (text, image, audio and video, etc.) and non-European data (manifold, node graph, molecular structure graph, cell graph, etc.) as shown in Figure 1. A graph is one of the basic data structures, usually represented by G=(V, E), that is, a graph is composed of a node and its adjacent edges, and most phenomena or scenarios can be used to capture the special relationship in the graph [3].There are a large number of user-item interaction lists in the recommendation field. These interaction histories can form a huge bipartite graph of network topology (Figure 2). The bipartite graph intuitively expresses the connection between users and items, and how to use user-item 2 It has become a research hotspot to bring more benefits to businesses. (a) European space (b) Non-european space Fig. 1. Examples of European and non-European data. The strong learning ability of deep learning technology in images and text has attracted more and more researchers to apply deep learning methods to graph data, learn feature extraction and representation of graph structures, and graph neural networks (GNN) It came into being [4]. Graph neural network is a deep learning-based method running on the graph domain, which makes up for the problem that traditional deep models cannot generalize to graph data. The development of graph neural network enables node information and nodeto-node relationship information in the recommendation system to be fully mined, bringing greater commercial value. (a) List of User-item interactions (b) Bipartite graph Fig. 2. User-item bipartite graph. 2 Introduction to recommendation system development The recommendation system, as the user's preferred filter, is constantly updated with people's needs and technological development. The development of the recommendation system has experienced three generations of technical improvements. The first generation is traditional recommendation technology. Among them, the recommendation based on collaborative filtering [5] is the most widely used in traditional recommendation. The item that best matches the user’s explicit preference is recommended 2 MATEC Web of Conferences 336, 05010 (2021) CSCNS2020 https://doi.org/10.1051/matecconf/202133605010 to the user. This has a defect that the user’s explicit preference matrix has a high dimension but a very large distribution. Sparse, cold start problems exist for new users or new projects. The second generation is a recommendation technology based on a deep model. Many studies use deep learning techniques to complete recommendation tasks [6]. The recommendation system based on deep learning can already solve the problem of data sparseness, but the cold start problem of the recommendation system is still not effectively solved. In order to get more connections with new users or new projects, try to add auxiliary information to the relationship between users and projects, establish the relationship between projects and projects, etc. The topological bipartite graph structure data can naturally represent users and projects The relationship between time, project and project, user and user, even after adding auxiliary information, the graph structure can still be connected. Therefore, in the field of recommendation systems, industry and academia are increasingly inclined to use graph data. The third generation is a recommendation system based on graph neural networks with continuous achievements in recent years. Graph neural network (GNN) is a general deep learning framework defined specifically for processing bipartite graph data structures (Figure 2), through Perform end-to-end recommendation modeling on graph data, learn more deep features of nodes, edges, and subgraphs [7], provide full scoring grades for those unobserved cross-complementarity, and then predict (...truncated)