mFollowIR.py

import datasets
import json

_CITATION = '''@article{weller2024mfollowir,
  title={{mFollowIR: a Multilingual Benchmark for Instruction Following in Information Retrieval}},
  author={Weller, Orion and Chang, Benjamin and Yang, Eugene and Yarmohammadi, Mahsa and Barham, Sam and MacAvaney, Sean and Cohan, Arman and Soldaini, Luca and Van Durme, Benjamin and Lawrie, Dawn},
  journal={arXiv preprint TODO},
  year={2024}
}'''

_DESCRIPTION = 'Dataset load script for mFollowIR combining multiple languages'

_LANGUAGES = ["fas", "rus", "zho"]

class mFollowIRCrossLingual(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name=f"{config_name}-{lang}",
            version=datasets.Version("1.0.0"),
            description=f"{config_name} configuration for {lang} language in mFollowIR dataset."
        )
        for config_name in ["qrels_og", "qrels_changed", "corpus", "queries", "top_ranked"]
        for lang in _LANGUAGES
    ]

    def _info(self):
        name = self.config.name
        if name.startswith("qrels"):
            features = datasets.Features({
                "query-id": datasets.Value("string"),
                "corpus-id": datasets.Value("string"),
                "score": datasets.Value("float64"),
            })
        elif name.startswith("corpus"):
            features = datasets.Features({
                "_id": datasets.Value("string"),
                "title": datasets.Value("string"),
                "text": datasets.Value("string"),
            })
        elif name.startswith("queries"):
            features = datasets.Features({
                "_id": datasets.Value("string"),
                "text": datasets.Value("string"),
                "instruction_og": datasets.Value("string"),
                "instruction_changed": datasets.Value("string"),
            })
        elif name.startswith("top_ranked"):
            features = datasets.Features({
                "qid": datasets.Value("string"),
                "pid": datasets.Value("string"),
            })

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            supervised_keys=None,
            homepage="https://arxiv.org/abs/2304.12367",
            license=None,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        name, lang = self.config.name.split("-")
        
        if name in ["qrels_og", "qrels_changed"]:
            filepath = dl_manager.download_and_extract(f"https://huggingface.co/datasets/jhu-clsp/mFollowIR-{lang}-cl/resolve/main/{name}/test.jsonl")
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TEST,
                    gen_kwargs={"filepath": filepath},
                ),
            ]
        elif name in ["corpus", "top_ranked"]:
            filepath = dl_manager.download_and_extract(f"https://huggingface.co/datasets/jhu-clsp/mFollowIR-{lang}-cl/resolve/main/{name}.jsonl")
            return [
                datasets.SplitGenerator(
                    name=name,
                    gen_kwargs={"filepath": filepath},
                ),
            ]
        elif name in ["queries"]:
            filepath = dl_manager.download_and_extract(f"https://huggingface.co/datasets/jhu-clsp/mFollowIR/resolve/main/{lang}_map_final.jsonl")
            return [
                datasets.SplitGenerator(
                    name=name,
                    gen_kwargs={"filepath": filepath},
                ),
            ]
            
    def _generate_examples(self, filepath):
        name = self.config.name.split("-")[0]
        with open(filepath, "r", encoding="utf-8") as f:
            for idx, line in enumerate(f):
                data = json.loads(line)
                if name.startswith("qrels"):
                    yield idx, {
                        "query-id": data["query-id"],
                        "corpus-id": data["corpus-id"],
                        "score": data["score"],
                    }
                elif name == "corpus":
                    yield idx, {
                        "_id": data["_id"],
                        "title": data["title"],
                        "text": data["text"],
                    }
                elif name == "queries":
                    yield idx, {
                        "_id": data["query_id"],
                        "text": data["ht_text"],
                        "instruction_og": data["instruction_og"],
                        "instruction_changed": data["instruction_changed"],
                    }
                elif name == "top_ranked":
                    yield idx, {
                        "qid": data["qid"],
                        "pid": data["pid"],
                    }