A survey of training on graphs: taxonomy, methods, and Applications

author ：Jun Xia, Yanqiao Zhu, Yuanqi Du, Stan Z. Li

brief introduction ： image BERT Such a pre training language model （PLM） Has revolutionized natural language processing （NLP） Pattern . Inspired by its diffusion , People think about the pre training graph model （PGM） A lot of research has been done . because PGMs Powerful model architecture , Rich knowledge can be obtained from a large number of marked and unmarked graphic data . The knowledge of implicit coding in model parameters can benefit various downstream tasks , And help to alleviate several basic problems of graph learning . In this paper , The author provides the first comprehensive PGMs An overview of . The author first introduces the limitations of graph representation learning , Then it introduces the motivation of graph pre training . then , According to the classification, the existing PGM Systematic classification . Next , The authors introduce PGMs Applications in social recommendation and drug discovery . Last , The author summarizes several promising research directions 、 As a guiding reference for future research .

Paper download ：https://arxiv.org/pdf/2202.07893