Different tasks of GraphML

  • Node classification: Predict a property of a node.(e.g. Categorize online users)
  • Link prediction: Predict whether there are missing links between two nodes.(e.g. Kownledge graph completion)
  • Graph classification: Categorize different graphs(e.g. Molecule property prediction)
  • Clustering(Community): Detect if nodes form a community(e.g. Social circle detection)
  • Graph generation(e.g. Drug discovery)
  • Graph evolution(e.g. Physical simulation)

Applications

Node level:

Protein Folding: 憑藉氨基酸序列預測蛋白質的3D結構。

Edge level:

Recommender system(推薦系統):

讓推薦關聯性更高的事物在embedding後的距離更接近。

Drug Side Effects(藥物副作用預測):

以藥物和蛋白質聯繫建構如圖graph後對兩種藥物之間的未知link做預測。

Subgraph level:

Traffic Prediction(交通通行時間預測)

以路段作為節點，以路段之間的連通情況作為edge構建graph，將graph輸入GNN以預測通行時間。

Graph level:

Drug Discovery

以原子為node，化學鍵為edge構建graph輸入深度學習模型中預測有效的抗生素分子式。

參考

  • StandFord CS224W http://web.stanford.edu/class/cs224w/