{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/edge-attention-based-multi-relational-graph","title":"Edge Attention-based Multi-Relational Graph Convolutional Networks","arxiv_id":"1802.04944","date":"2018-02-14","proceeding":null,"authors":["Chao Shang","Qinqing Liu","Ko-Shin Chen","Jiangwen Sun","Jin Lu","Jin-Feng Yi","Jinbo Bi"],"abstract":"Graph convolutional network (GCN) is generalization of convolutional neural\nnetwork (CNN) to work with arbitrarily structured graphs. A binary adjacency\nmatrix is commonly used in training a GCN. Recently, the attention mechanism\nallows the network to learn a dynamic and adaptive aggregation of the\nneighborhood. We propose a new GCN model on the graphs where edges are\ncharacterized in multiple views or precisely in terms of multiple\nrelationships. For instance, in chemical graph theory, compound structures are\noften represented by the hydrogen-depleted molecular graph where nodes\ncorrespond to atoms and edges correspond to chemical bonds. Multiple attributes\ncan be important to characterize chemical bonds, such as atom pair (the types\nof atoms that a bond connects), aromaticity, and whether a bond is in a ring.\nThe different attributes lead to different graph representations for the same\nmolecule. There is growing interests in both chemistry and machine learning\nfields to directly learn molecular properties of compounds from the molecular\ngraph, instead of from fingerprints predefined by chemists. The proposed GCN\nmodel, which we call edge attention-based multi-relational GCN (EAGCN), jointly\nlearns attention weights and node features in graph convolution. For each bond\nattribute, a real-valued attention matrix is used to replace the binary\nadjacency matrix. By designing a dictionary for the edge attention, and forming\nthe attention matrix of each molecule by looking up the dictionary, the EAGCN\nexploits correspondence between bonds in different molecules. The prediction of\ncompound properties is based on the aggregated node features, which is\nindependent of the varying molecule (graph) size. We demonstrate the efficacy\nof the EAGCN on multiple chemical datasets: Tox21, HIV, Freesolv, and\nLipophilicity, and interpret the resultant attention weights.","url_abs":"http://arxiv.org/abs/1802.04944v2","url_pdf":"http://arxiv.org/pdf/1802.04944v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"edge-attention-based-multi-relational-graph","repo_url":"https://github.com/Luckick/EAGCN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"attribute","task_name":"Attribute"}],"methods":[{"method_slug":"gcn","method_name":"GCN"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1802.04944","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}