Abstract:Open Information Extraction (OpenIE) aims to construct expansive open knowledge bases (OKBs) by extracting triples (noun phrase, relation phrase, noun phrase) from unstructured text. One critical problem in OKBs is the lack of canonicalization for noun phrases and relation phrases, leading to the storage of redundant and ambiguous facts. Consequently, open knowledge base canonicalization, which clusters synonymous phrases into the same group, has emerged as an active research area. Existing approaches either leverage fact triples or source context in isolation, or at best, interact them at the clustering level. However, these approaches lack explicit interaction or only loosely couple the two types of knowledge, resulting in the potential loss of valuable intermediate information. In this paper, we propose MuFIC, a novel unsupervised framework that interacts the fact triples and source context at the feature level to address these limitations. In order to capture and integrate fine-grained fact and context knowledge, we design three levels of feature interaction: low-level context-guided feature interaction, mid-level fact-guided feature interaction, and high-level gated fusion feature interaction. Furthermore, we introduce an additional objective function via contrastive learning to improve the quality of extracted features and reduce knowledge-specific noise. Finally, we design a bidirectional feedback mechanism to better guide the learning process of joint features by harnessing side information prototype learning, and to dynamically optimize side information based on the clustering results formed by joint features. Extensive experiments on three public benchmark datasets demonstrate the superiority of our proposed framework.

Joint Open Knowledge Base Canonicalization and Linking

Jointly Canonicalizing and Linking Open Knowledge Base Via Unified Embedding Learning

Multi-level Feature Interaction for Open Knowledge Base Canonicalization

Multi-View Clustering for Open Knowledge Base Canonicalization

Open knowledge base canonicalization with multi-task learning

mOKB6: A Multilingual Open Knowledge Base Completion Benchmark

A Canonicalization-Enhanced Known Fact-Aware Framework for Open Knowledge Graph Link Prediction.

Canonicalizing Open Knowledge Bases with Multi-Layered Meta-Graph Neural Network

Joint Inference for Knowledge Base Population

COMBO: A Complete Benchmark for Open KG Canonicalization

KICE: A Knowledge Consolidation and Expansion Framework for Relation Extraction.

Interaction is Worth More Explanations: Improving Human-Object Interaction Representation with Propositional Knowledge

JCBIE: a joint continual learning neural network for biomedical information extraction

Building Large-Scale Knowledge Base for Relations from Text

Jointly Modeling Fact Triples and Text Information for Knowledge Base Completion.

Joint Biomedical Entity and Relation Extraction with Knowledge-Enhanced Collective Inference

Chinese Open Relation Extraction and Knowledge Base Establishment

CESI: Canonicalizing Open Knowledge Bases using Embeddings and Side Information

Mastering the Task of Open Information Extraction with Large Language Models and Consistent Reasoning Environment

Joint Entity Recognition and Disambiguation.