MINE Eval expanded

uses kggen with gpt-5-nano, and gpt-oss-20b with multiple levels of reasoning + openie and graphrag options

MINE Eval (deduplication) Scikit Learn vs Faiss

============================================================
SUMMARY STATISTICS
============================================================
faiss                                    scikit-learn
Mean:                                    36.93 70.47
Median:                                  40.00 73.33
Std Dev:                                 17.72 20.36
Min:                                      0.00 13.33
Max:                                     73.33 100.00
============================================================

scikit-learn vs faiss - Average difference: 33.53%
Better cases in scikit-learn: 90
Better cases in faiss: 6
Equal cases: 4
============================================================

#69 #62

WikiQA KGGen Dataset

WikiQA Dataset may be found here: https://huggingface.co/datasets/belindamo/wiki_qa_kggen

MINE dataset and evaluations

This files were organized into a single dataset that allows to rerun evaluations, keeping this just as a backup and history.

Should use https://huggingface.co/datasets/josancamon/kg-gen-MINE-evaluation-dataset.

answers.json = generated_queries instead, this was a bad name