图嵌入算法研究进展

图嵌入算法是将高维网络信息映射至低维后用实数向量表示的一种方法，用于解决推荐系统、社区发现及节点分类等。近年来，随着科技的进步，图数据呈现海量、异构、高维、多模态等特点，机器学习等人工智能算法对高性能的图嵌入算法的需求日益增加，图嵌入已成为国内外人工智能领域的研究热点之一。对图嵌入算法的研究进展、技术原理及基础理论进行了综述，系统概述了已有的主流图嵌入算法，包括基于降维方法的图嵌入、基于矩阵分解的图嵌入、基于网络拓扑结构的图嵌入、基于神经网络的图嵌入、基于生成式对抗网络的图嵌入和基于超图网络的图嵌入，对这些算法进行了分析与比较，并给出了相应的应用场景；归纳总结了常用的测试数据集及其评价标准；最后，展望了图嵌入算法的研究趋势和方向。

Research progress of graph embedding algorithms

As an important form of expressing the relationship among entities, graph networks have been widely used in data analysis, relational reasoning, and information services. For these applications, how to reasonably represent network characteristic information is the primary task of network analysis research. Graph embedding technology solves the problem of how to efficiently and reasonably map massive, heterogeneous, and complex high-dimensional graph data to low-dimensional vector space while still retaining the original data feature information. This paper aims to survey the algorithm and research progress of graph embedding in recent years, analyze the development status of this field, and explore the direction for subsequent research. First, it reviews the principle and basic theory of graph embedding technology, then systematically investigates the current mainstream graph embedding algorithms, including graph embedding approaches based respectively on dimensionality reduction, matrix decomposition,network topology,neural network, generative adversarial network, and hypergraph. Then we show the application scenarios of graph embedding technology and introduce the commonly used test data sets and evaluation criteria. Finally, we highlight the future research trends and directions of graph embedding, such as dynamic graph embedding, graph embedding scalability and interpretability.