TaxoExpan: Self-supervised Taxonomy Expansion with Position-Enhanced Graph Neural Network (2020-01-26T00:00:00.000000Z)

TL;DR

A novel self-supervised framework, named TaxoExpan, which automatically generates a set of ⟨query concept, anchor concept⟩ pairs from the existing taxonomy as training data, and develops two innovative techniques, including a position-enhanced graph neural network that encodes the local structure of an anchor concept in theexisting taxonomy and a noise-robust training objective that enables the learned model to be insensitive to the label noise in the self- supervision data.

Abstract

Taxonomies consist of machine-interpretable semantics and provide valuable knowledge for many web applications. For example, online retailers (e.g., Amazon and eBay) use taxonomies for product recommendation, and web search engines (e.g., Google and Bing) leverage taxonomies to enhance query understanding. Enormous efforts have been made on constructing taxonomies either manually or semi-automatically. However, with the fast-growing volume of web content, existing taxonomies will become outdated and fail to capture emerging knowledge. Therefore, in many applications, dynamic expansions of an existing taxonomy are in great demand. In this paper, we study how to expand an existing taxonomy by adding a set of new concepts. We propose a novel self-supervised framework, named TaxoExpan, which automatically generates a set of ⟨query concept, anchor concept⟩ pairs from the existing taxonomy as training data. Using such self-supervision data, TaxoExpan learns a model to predict whether a query concept is the direct hyponym of an anchor concept. We develop two innovative techniques in TaxoExpan: (1) a position-enhanced graph neural network that encodes the local structure of an anchor concept in the existing taxonomy, and (2) a noise-robust training objective that enables the learned model to be insensitive to the label noise in the self-supervision data. Extensive experiments on three large-scale datasets from different domains demonstrate both the effectiveness and the efficiency of TaxoExpan for taxonomy expansion.

Authors

Jiawei Han

12 papers

Zhihong Shen

3 papers

Kuansan Wang

6 papers

TL;DR

Abstract

Authors

References71 items

Discovering Hypernymy in Text-Rich Heterogeneous Information Network by Exploiting Context Granularity

Every Child Should Have Parents: A Taxonomy Refinement Algorithm Based on Hyperbolic Term Embeddings

Position-aware Graph Neural Networks

A User-Centered Concept Mining System for Query and Document Understanding at Tencent

Taxonomy-Aware Multi-Hop Reasoning Networks for Sequential Recommendation

TaxoGen: Constructing Topical Concept Taxonomy by Adaptive Term Embedding and Clustering

Graph Classification using Structural Attention

HiExpan: Task-Guided Taxonomy Construction by Hierarchical Tree Expansion

TaxoGen: Unsupervised Topic Taxonomy Construction by Adaptive Term Embedding and Clustering

Representation Learning with Contrastive Predictive Coding

Enriching Taxonomies With Functional Domain Knowledge

Hierarchical Graph Representation Learning with Differentiable Pooling

Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generation

Graph Convolutional Neural Networks for Web-Scale Recommender Systems

Comparing Constraints for Taxonomic Organization

A Comparison of Two Paraphrase Models for Taxonomy Augmentation

Hearst Patterns Revisited: Automatic Hypernym Detection from Large Text Corpora

A Web-scale system for scientific knowledge exploration

End-to-End Reinforcement Learning for Automatic Taxonomy Induction

An End-to-End Deep Learning Architecture for Graph Classification

Learning Deep Generative Models of Graphs

Junction Tree Variational Autoencoder for Molecular Graph Generation

FastGCN: Fast Learning with Graph Convolutional Networks via Importance Sampling

Graph Attention Networks

SetExpan: Corpus-Based Set Expansion via Context Feature Selection and Rank Ensemble

Entity Set Expansion via Knowledge Graphs

Inductive Representation Learning on Large Graphs

Taxonomy Induction Using Hypernym Subsequences

MetaPAD: Meta Pattern Discovery from Massive Text Corpora

Deep Sets

Understand Short Texts by Harvesting and Analyzing Semantic Knowledge

Extending WordNet with Fine-Grained Collocational Information via Supervised Distributional Learning

Learning Term Embeddings for Taxonomic Relation Identification Using Dynamic Weighting Neural Network

Supervised Distributional Hypernym Discovery via Domain Adaptation

Semi-Supervised Classification with Graph Convolutional Networks

Enriching Word Vectors with Subword Information

SemEval-2016 Task 14: Semantic Taxonomy Enrichment

MSejrKu at SemEval-2016 Task 14: Taxonomy Enrichment by Evidence Ranking

XGBoost: A Scalable Tree Boosting System

Improving Hypernymy Detection with an Integrated Path-based and Distributional Method

CrowdTC: Crowdsourced Taxonomy Construction

An Overview of Microsoft Academic Service (MAS) and Applications

Inclusive yet Selective: Supervised Distributional Hypernymy Detection

Structured Learning for Taxonomy Induction with Belief Propagation

A hierarchical Dirichlet model for taxonomy expansion for search engines

Taxonomy discovery for personalized recommendation

Distributed Representations of Words and Phrases and their Compositionality

OntoLearn Reloaded: A Graph-Based Algorithm for Taxonomy Induction

PATTY: A Taxonomy of Relational Patterns with Semantic Types

Constructing Task-Specific Taxonomies for Document Collection Browsing

Probase: a probabilistic taxonomy for text understanding

A Semi-Supervised Method to Learn and Construct Taxonomies Using the Web

Building Enterprise Taxonomies

Named Entity WordNet

Towards new information resources for public health - From WordNet to MedicalWordNet

Characterising Measures of Lexical Distributional Similarity

Medical Subject Headings (MeSH).

Snowball: extracting relations from large plain-text collections

An Information-Theoretic Definition of Similarity

Automatic Acquisition of Hyponyms from Large Text Corpora

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Reserating the awesometastic: An automatic extension of the WordNet taxonomy for novel terms

ArchiWordNet: Integrating WordNet with Domain-Specific Knowledge

Proceedings of the Twenty-Second International Joint Conference on Artificial Intelligence A Graph-Based Algorithm for Inducing Lexical Taxonomies from Scratch

, to predict the matching score between a query concept and a candidate position

DeepSetMLP : Another self-supervised method that extends ParentMLP by adding information of candidate position’s children nodes

F1 is the harmonic mean of Wu&P accuracy and recall

The same tree boosting model described in the previous subsection 5.1.3

Recall is the percentage of concepts for which an attached parent is predicted

The ensemble model of FWFS and ETF, which adds the FWFS property as a binary feature into the Lamb-daMART model in ETF

5.2.3 Baseline Methods

Field of Study

Journal Information

Name

Venue Information

Name