Publication - Patent

A Fine-Grained Entity Typing Method for Knowledge Base Population

Dependency Graph

Application No.

    CN201510033050.4

Authors

    Xueqi Cheng, Yantao Jia, Hailun Lin, Manling Li, etc. (2nd student author).

Sponsored by

    National Natural Science Foundation of China No. 61402442

Abstract

  • Objective: Fine-grained entity typing aims to assgin entities with fine-grained types, which can help to populate the knowledge bases.
  • Problem: Existing fine-grained entity typing based on rules or weak supervision. Rule based methods require much human effort, and weak supervision based methods suffer from context deficiency.
  • Proposed to construct a dependency graph of entities and types by entity linking, and then get types by random walk on the graph.

My Work

  • Implemented the entity linking module by using context similarity.
  • Selected features to define context similarities, including word frequency, abbreviation, alias, and so on.
  • Implemented the random walk module to generate types. The random walk starts from every node in the graph, and walk to another node according to a transfer possibility.