Evaluation of LLM-geneated boolean search queries

prompts.csv contains the prompts and respective keyword for embedding.
The first row is the name of the corresponding file in ordered_items_per_prompt/

ordered_items_per_prompt/ contains one file per prompt.
Each of those files has the IDs of all items in the observatory, ordered by relevance.
The second column is the distance of the embedding of the keyword and the embedding of the item.
The third column indicates whether the item is relevant ('y') or irrelevant ('n' or blank).

For each (prompt,model) combination, a file is created in results_rewrites/
It contains the generated query for that (prompt,model).

For each (prompt,model) combination, a file is created in results_items/
It contains the items that match the generated query in the observatory.

results/metrics contains the macro-metrics for each prompt.
The columns are: model, precision, recall, latency.