The problem
Companies have thousands of reports and contracts in free text. Labelling enough data to train an extractor for each new relation is too expensive; few-shot relation extraction learns from 5–10 examples per relation.
Who needs it
Maintenance and manufacturing, legal and procurement departments, insurance.
Idea of the approach
- Start from FewRel to learn the method, then a small domain set (public maintenance logs or contracts) labelled by the students.
- Compare a prompt-based few-shot extractor (in the spirit of GAPrompt/KHSPrompt) with an LLM given examples.
- Load the extracted triples into a graph database (Neo4j) with a search interface.
What you will build and deliver
- Domain dataset, extractor and knowledge-graph demo.
- Paper.
Skills needed
Python, NLP basics, Hugging Face; databases.
Research contribution & where to publish
Contribution: domain transfer of prompt-based few-shot RE with a real KG application. Venues: Findings of ACL/EMNLP, COLING, Knowledge-Based Systems, Neurocomputing.
Key references — start reading here
- X. Han, H. Zhu, P. Yu, Z. Wang, Y. Yao, Z. Liu, M. Sun. FewRel: A Large-Scale Supervised Few-Shot Relation Classification Dataset with State-of-the-Art Evaluation. EMNLP, 2018.
- H. Djigal et al.. Few-Shot Relation Extraction Based on Prompt Learning: A Taxonomy, Survey, Challenges and Future Directions. ACM Computing Surveys, 2025.
- GAPrompt authors (H. Djigal et al.). GAPrompt: A soft prompt compression model for few-shot relation extraction. Knowledge-Based Systems, 2025.
- A. Hogan, E. Blomqvist, M. Cochez, C. d'Amato, G. de Melo, C. Gutierrez, et al.. Knowledge Graphs. ACM Computing Surveys, 2021.
Interested?
Send the topic title and the courses you have taken.
Contact Dr. Hamza Djigal