MultiConIR_v2

This repository hosts the MultiConIR dataset in four domains/configs:

• Books
• Legal Document
• Medical Case
• Movies
• People

Each config contains three splits:

• corpus: candidate documents (JSONL)
• queries: queries (JSONL)
• qrels: relevance judgements (JSONL)

Data format

• corpus.jsonl: one document per line (e.g., docid, text, optional metadata)
• queries.jsonl: one query per line (e.g., qid, text, optional metadata)
• qrels.jsonl: one relevance entry per line (e.g., qid, docid, rel)

(Exact fields follow the original exported format in this repo.)

Usage

from datasets import load_dataset

# Load the "queries" split of Books
ds = load_dataset("Lux1997/MultiConIR_v2", "Books", split="queries")

# Load corpus / qrels
corpus = load_dataset("Lux1997/MultiConIR_v2", "Books", split="corpus")
qrels  = load_dataset("Lux1997/MultiConIR_v2", "Books", split="qrels")