Relation extraction from texts
summary
Walks through building a relation-extraction pipeline that uses OpenNRE and fine-tunes BERT-base-uncased on a curated slice of a unified distant‑supervision DB (originally ~47 million examples and ~1,000 relation types). Filters to five relation classes (no_relation, unknown, father, mother, sibling), cleans noisy distant‑supervision labels, tokenizes with entity markers, writes JSON inputs, and trains (≈70k train / 4.4k val / 13k test). Produces a usable model for simple family relations, visualizes triples with networkx, and documents common preprocessing pitfalls.
AI-written summary · about
chapters (10)
- 0:00Relation extraction intro
- 4:24What is information extraction
- 5:50Relation extraction scope
- 14:39Relation extraction basics
- 27:50Building family relation dataset
- 43:51Data preprocessing
- 50:18Training the model
- 1:07:48Model inference
- 1:22:52Visualizing knowledge graphs
- 1:29:11Summary
related talks
58mRelation Extraction from Online PH News Source
05mDistant Annotation for Training Data Preparation
19mAddressing Treatment-Relevance in Biomedical Relation Extraction
32mWhat is X up to? - NER and Relationship Extraction for Information Extraction
26mDeep Learning Architectures for Semantic Relation Detection Tasks
21mRelationships Matter: Mining relationships using Deep Learning
36mEnhanced and Efficient Reasoning in Large Language Models
38mUnderstanding and Applying Self-Attention for NLP