Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Few-shot Knowledge Graph Relational Reasoning via Subgraph Adaptation

Haochen Liu, Song Wang, Chen Chen, Jundong Li

TL;DR

Experimental results on three prevalent datasets demonstrate the superiority of the proposed framework SAFER, which enables the extraction of more comprehensive information from support triplets while minimizing the impact of spurious information when predicting query triplets.

Abstract

Few-shot Knowledge Graph (KG) Relational Reasoning aims to predict unseen triplets (i.e., query triplets) for rare relations in KGs, given only several triplets of these relations as references (i.e., support triplets). This task has gained significant traction due to the widespread use of knowledge graphs in various natural language processing applications. Previous approaches have utilized meta-training methods and manually constructed meta-relation sets to tackle this task. Recent efforts have focused on edge-mask-based methods, which exploit the structure of the contextualized graphs of target triplets (i.e., a subgraph containing relevant triplets in the KG). However, existing edge-mask-based methods have limitations in extracting insufficient information from KG and are highly influenced by spurious information in KG. To overcome these challenges, we propose SAFER (Subgraph Adaptation for Few-shot Relational Reasoning), a novel approach that effectively adapts the information in contextualized graphs to various subgraphs generated from support and query triplets to perform the prediction. Specifically, SAFER enables the extraction of more comprehensive information from support triplets while minimizing the impact of spurious information when predicting query triplets. Experimental results on three prevalent datasets demonstrate the superiority of our proposed framework SAFER.

Few-shot Knowledge Graph Relational Reasoning via Subgraph Adaptation

TL;DR

Experimental results on three prevalent datasets demonstrate the superiority of the proposed framework SAFER, which enables the extraction of more comprehensive information from support triplets while minimizing the impact of spurious information when predicting query triplets.

Abstract

Few-shot Knowledge Graph (KG) Relational Reasoning aims to predict unseen triplets (i.e., query triplets) for rare relations in KGs, given only several triplets of these relations as references (i.e., support triplets). This task has gained significant traction due to the widespread use of knowledge graphs in various natural language processing applications. Previous approaches have utilized meta-training methods and manually constructed meta-relation sets to tackle this task. Recent efforts have focused on edge-mask-based methods, which exploit the structure of the contextualized graphs of target triplets (i.e., a subgraph containing relevant triplets in the KG). However, existing edge-mask-based methods have limitations in extracting insufficient information from KG and are highly influenced by spurious information in KG. To overcome these challenges, we propose SAFER (Subgraph Adaptation for Few-shot Relational Reasoning), a novel approach that effectively adapts the information in contextualized graphs to various subgraphs generated from support and query triplets to perform the prediction. Specifically, SAFER enables the extraction of more comprehensive information from support triplets while minimizing the impact of spurious information when predicting query triplets. Experimental results on three prevalent datasets demonstrate the superiority of our proposed framework SAFER.
Paper Structure (28 sections, 16 equations, 4 figures, 3 tables)

This paper contains 28 sections, 16 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Meta-learning-based Few-shot KGR
  4. Edge-mask-based Few-shot KGR
  5. Problem Formulation
  6. Methodology
  7. Retrieving Contextualized Graphs
  8. Edge Weight Assignment
  9. Subgraph Adaptation
  10. Support Adaptation
  11. Query Adaptation
  12. Training Objective
  13. Experiments
  14. Experimental Settings
  15. Datasets
  16. ...and 13 more sections

Figures (4)

  • Figure 1: We provide an instance for the two limitations of edge-mask-based methods. In this example, there are two support triplets (music, created_by, musican) and (news article, created_by, reporter). When extracting support information by finding the common subgraph, the extraction of edges with similar meanings but in different graphs will fail, and some spurious information will be extracted, which cannot correctly represent the logical pattern of the relation created_by.
  • Figure 2: The framework of SAFER, which shows the scoring pipeline for a query tail candidate $c$ of target relation $r'$. We represent the same relations in colors, while the gray relations are all different. We first extract the contextualized graph of each support and query triplet and assign weights to all edges using an aggregation process $P_w$ (the width of edges represents weights). Then we apply another aggregation process $P_a$ and two adaptation operations to perform support information extraction and query candidate scoring.
  • Figure 3: The performance of our proposed method SAFER with different $\lambda$.
  • Figure 4: An instance on dataset ConceptNet using the edge-mask-based method CSR and our method SAFER. The figure shows part of support and query graphs and the scores of the 3-top candidates of the two methods. The shown edges prove the limitation of the extraction of common subgraphs in edge-mask-based methods.