Multi-source Machine Learning for Natural Language Relation Extraction
Vision
We develop multi-source information (relation and attribute) extraction techniques that are robust to limited labeled data and noise. We leverage semantic information and prompts for zero-shot relation extraction, and visual semantic information for multimodal multi-feature few-shot relation extraction. We also develop n-ary cross-sentence relation extraction methods for both supervised and unsupervised settings.
Funded Projects
- Natural Language Processing for Teaching and Research in Engineering Education – National Science Foundation (NSF) 2022
- Content Analytics for Smart Search – Xerox Workplace Innovation Research 2018
- Insider Threat Detection for Document Content Security – Xerox Workplace Innovation Research Program 2017
Publications
- Knowledge-Enhanced Multi-Label Few-Shot Product Attribute-Value Extraction – ACM CIMK 2023
- Clustering-based Unsupervised Generative Relation Extraction – IEEE Big Data 2022
- Unsupervised Relation Extraction: A Variational Autoencoder Approach – EMNLP 2021
- Zero-shot Relation Classification from Side Information – CIKM 2021 – code, data, presentation
- Prompt-based Zero-shot Relation Classification with Semantic Knowledge Augmentation Preprint 2021
- Reinforcement Learning-based N-ary Cross-Sentence Relation Extraction – Preprint 2020