xmarket_ml.py

"""XMarket retrieval dataset for multiple languages"""


import json
import gzip

from typing import List

import datasets

_DESCRIPTION = """\
An ecommerce category to product retrieval dataset in multiple languages.
The data comes from:
https://xmrec.github.io/
"""

_HOMEPAGE_URL = 'https://xmrec.github.io/'
_LANGUAGES = {'es': 'es', 'de': 'de', 'en': 'us'}
_VERSION = '1.0.0'


URLS = [
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Arts_Crafts_and_Sewing/metadata_%s_Arts_Crafts_and_Sewing.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Automotive/metadata_%s_Automotive.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Books/metadata_%s_Books.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/CDs_and_Vinyl/metadata_%s_CDs_and_Vinyl.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Cell_Phones_and_Accessories/metadata_%s_Cell_Phones_and_Accessories.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Digital_Music/metadata_%s_Digital_Music.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Electronics/metadata_%s_Electronics.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Grocery_and_Gourmet_Food/metadata_%s_Grocery_and_Gourmet_Food.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Home_and_Kitchen/metadata_%s_Home_and_Kitchen.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Industrial_and_Scientific/metadata_%s_Industrial_and_Scientific.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Kindle_Store/metadata_%s_Kindle_Store.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Movies_and_TV/metadata_%s_Movies_and_TV.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Musical_Instruments/metadata_%s_Musical_Instruments.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Office_Products/metadata_%s_Office_Products.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Sports_and_Outdoors/metadata_%s_Sports_and_Outdoors.json.gz',
    'https://ciir.cs.umass.edu/downloads/XMarket/FULL/%s/Toys_and_Games/metadata_%s_Toys_and_Games.json.gz',
]


class XMarketMLConfig(datasets.BuilderConfig):
    """BuilderConfig for XMarket."""

    def __init__(self, languages=None, **kwargs):
        super(XMarketMLConfig, self).__init__(
            version=datasets.Version(_VERSION, ''), **kwargs
        )
        self.languages = languages


class XMarketML(datasets.GeneratorBasedBuilder):
    """The XMarket category to product retrieval dataset"""

    BUILDER_CONFIGS = [
        XMarketMLConfig(
            name=f"{name}-{lang}",
            description=f'{name.title()} of the multilingual XMarket dataset.',
            languages=None,
        )
        for name in ['corpus', 'queries', 'qrels'] for lang in _LANGUAGES
    ]

    BUILDER_CONFIG_CLASS = XMarketMLConfig

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self._data = None

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "_id": datasets.Value("string"),
                    "title": datasets.Value("string"),
                    "text": datasets.Value("string"),
                }
            ),
            supervised_keys=None,
            homepage=_HOMEPAGE_URL,
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager):
        json_files = []
        for lang in self.config.languages:
            url_lang = _LANGUAGES[lang]
            for url in URLS:
                json_files.append(dl_manager.download(url % (url_lang, url_lang)))
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    'files': json_files,
                    'split': 'test',
                },
            ),
        ]

    def _generate_examples(
        self,
        files: List[str],
        split: str = None,
    ):
        if not self._data:
            corpus = []
            queries = dict()
            qrels = dict()
            for file in files:
                with gzip.open(file, 'rt') as f:
                    for line in f:
                        product = json.loads(line)
                        corpus.append(
                            {
                                "_id": product['asin'],
                                "title": product['title'],
                                "text": product['description'],
                            }
                        )
                        for category in product['categories']:
                            if category in queries:
                                qid = queries[category]
                                qrels[qid].append(product['asin'])
                            else:
                                qid = f'q{len(queries)}'
                                queries[category] = qid
                                qrels[qid] = [product['asin']]
            self._data = {
                'corpus': corpus,
                'queries': queries,
                'qrels': qrels,
            }
        if self.config.name.startswith('corpus'):
            for line in self._data['corpus']:
                yield line['_id'], line
        elif self.config.name.startswith('queries'):
            for query, qid in queries.items():
                yield qid, {
                    "_id": qid,
                    "title": '',
                    "text": query,
                }
        elif self.config.name.startswith('qrels'):
            for qid, dids in qrels.items():
                yield qid, {
                    "_id": qid,
                    "title": '',
                    "text": ' '.join(dids),
                }
        else:
            raise ValueError(f'Unknown config name: {self.config.name}')