LLMs Underperform Graph-Based Parsers on Supervised Relation Extraction for Complex Graphs

Create a hybrid pipeline that uses LLMs for entity identification but delegates complex relation extraction to traditional graph-based parsers. This maximizes the strengths of both symbolic and neural paradigms.

Suggested repo: graph-llm-hybrid

"Stop force-fitting LLMs into relation extraction: use the right tool for the graph."

Estimated effort: 60h