README - Evaluation Toolkit SKG-LLM-RAG
=========================================

This supplementary package provides evaluation code and datasets for the paper:

  "Leveraging Survey Metadata for LLM Reasoning via Knowledge Graphs"

The code implements three evaluation tasks with extended functionality:
  1. Semantic Retrieval
  2. Link Prediction
  3. Topic Discovery

Files included
--------------
- eval_tasks.py
- eval_tasks_runner.py

    Python module implementing the evaluators with sentence-embedding
    support (auto-fallback to TF-IDF), per-query rankings, and bootstrap CIs.

- Example dataset files (referenced in examples below):
    * b17017_schema_kg_prov.jsonl      (knowledge graph with provenance)
    * b17017_skg_nodes.csv             (SKG node table)
    * b17017_queries.csv               (semantic retrieval query set)
    * b17017_bm25_corpus.jsonl         (BM25 flat corpus)
    * b17017_text_encoder_corpus.jsonl (text-encoder flat corpus)
    * b17017_llm_rag_flat_corpus.jsonl (LLM-RAG flat corpus)
    * qa_outputs_demo.jsonl            (demo QA outputs with citations)