Publication - Patent

A Relation Inference System Based on Knowledge Graph Embedding

The adaptive margin in PTransA

Application No.



    Xueqi Cheng, Yantao Jia, Manling Li, etc. (1st student author)

Sponsored by

    Collaborative Research Program of Chinese Academy of Sciences and Huawei Inc. (one of the world's top 500 companies).


  • Objective: Relation Inference aims to predict the relations according to the given knowledge graph.
  • Problem: Existing embedding methods learn the representations of entities, relations, and multi-step relation paths by minimizing a general margin-based loss function shared by all relation paths, and the general margin is a hyper parameter affects the predictive performance greatly.
  • Proposed to adaptively determines the margin-based loss function for each path, by encoding the interaction between relations and multi-step relation paths for any given pair of entities.

My Work

  • Generated multi-paths and their weights through path-constraint resource allocation (Yankai Lin, et al., 2015).
  • Implemented the relation inference model to determine the optimal path-based margin of a relation to seperate postivie from negative multi-paths. Thus, the predictive performance is better, e.g., Mean Rank is decreased by 50~110 compared to a top traditional model, TransR.
  • Applied to a collaborative program of Chinese Academy of Science and Huawei Inc.