init

README.md

@@ -1,3 +1,121 @@
 ---
-license: mit
+license: cc-by-sa-3.0
+pretty_name: JaQuAD QG
+languages: ja
+multilinguality: monolingual
+size_categories: 10K<n<100K
+source_datasets: extended|wikipedia
+task_categories: question-generation
+task_ids: question-generation
 ---
+
+# Dataset Card for "qg_squad"
+
+## Table of Contents
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-fields)
+  - [Data Splits](#data-splits)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+
+## Dataset Description
+- **Repository:** [https://github.com/asahi417/lm-question-generation](https://github.com/asahi417/lm-question-generation)
+- **Paper:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+- **Point of Contact:** [Asahi Ushio](http://asahiushio.com/)
+- **Size of downloaded dataset files:** 284.1 MB
+- **Size of the generated dataset:** 269 MB
+
+### Dataset Summary
+[JaQuAD](https://github.com/SkelterLabsInc/JaQuAD) dataset for question generation (QG) task. The test set of the original 
+data is not publicly released, so we randomly sampled test questions from the training set.
+
+### Supported Tasks and Leaderboards
+* `question-generation`: The dataset can be used to train a model for question generation.
+  Success on this task is typically measured by achieving a high BLEU4/METEOR/ROUGE-L score.
+
+### Languages
+Japanese (ja)
+
+## Dataset Structure
+### Data Instances
+#### plain_text
+- **Size of downloaded dataset files:** 284.1 MB
+- **Size of the generated dataset:** 269 MB
+An example of 'train' looks as follows.
+```
+{
+  "question": "What is heresy mainly at odds with?",
+  "passage": "Heresy is any provocative belief or theory that is strongly at variance with established beliefs or customs. A heretic is a proponent of such claims or beliefs. Heresy is distinct from both apostasy, which is the explicit renunciation of one's religion, principles or cause, and blasphemy, which is an impious utterance or action concerning God or sacred things.",
+  "answer": "established beliefs or customs",
+  "sentence": "Heresy is any provocative belief or theory that is strongly at variance with established beliefs or customs .",
+  "passage_sentence": "<hl> Heresy is any provocative belief or theory that is strongly at variance with established beliefs or customs . <hl> A heretic is a proponent of such claims or beliefs. Heresy is distinct from both apostasy, which is the explicit renunciation of one's religion, principles or cause, and blasphemy, which is an impious utterance or action concerning God or sacred things.",
+  "passage_answer": "Heresy is any provocative belief or theory that is strongly at variance with <hl> established beliefs or customs <hl>. A heretic is a proponent of such claims or beliefs. Heresy is distinct from both apostasy, which is the explicit renunciation of one's religion, principles or cause, and blasphemy, which is an impious utterance or action concerning God or sacred things.",
+  "sentence_answer": "Heresy is any provocative belief or theory that is strongly at variance with <hl> established beliefs or customs <hl> ."
+}
+```
+### Data Fields
+The data fields are the same among all splits.
+#### plain_text
+- `question`: a `string` feature. 
+- `passage`: a `string` feature.
+- `answer`: a `string` feature.
+- `sentence`: a `string` feature.
+- `passage_answer`: a `string` feature, which is same as the passage but the answer is highlighted by a special token `<hl>`.
+- `passage_sentence`: a `string` feature, which is same as the passage but a sentence containing the answer is highlighted by a special token `<hl>`.
+- `sentence_answer`: a `string` feature, which is same as the sentence but the answer is highlighted by a special token `<hl>`.
+
+Each of `passage_answer`, `passage_sentence`, and `sentence_answer` feature is assumed to be used to train a question generation model,
+but with different information. The `passage_answer` and `sentence_answer` features are for answer-aware question generation and 
+`passage_sentence` feature is for sentence-aware question generation.
+
+### Data Splits
+
+|   name   |train|validation|test |
+|----------|----:|---------:|----:|
+|plain_text|75722|     10570|11877|
+
+## Dataset Creation
+### Curation Rationale
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+### Source Data
+#### Initial Data Collection and Normalization
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+#### Who are the source language producers?
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+### Annotations
+#### Annotation process
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+#### Who are the annotators?
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+### Personal and Sensitive Information
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+## Considerations for Using the Data
+### Social Impact of Dataset
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+### Discussion of Biases
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+### Other Known Limitations
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+## Additional Information
+### Dataset Curators
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+### Licensing Information
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+### Citation Information
+[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)

ja_sentence_split.py

@@ -0,0 +1,41 @@
+import re
+from typing import List
+import spacy
+
+__all__ = 'SentSplit'
+
+
+class JASplitter:
+    """ JA sentence splitter from https://github.com/himkt/konoha/blob/master/konoha/sentence_tokenizer.py """
+
+    PERIOD = "。"
+    PERIOD_SPECIAL = "__PERIOD__"
+    PATTERNS = [re.compile(r"（.*?）"), re.compile(r"「.*?」")]
+
+    @staticmethod
+    def conv_period(item) -> str:
+        return item.group(0).replace(JASplitter.PERIOD, JASplitter.PERIOD_SPECIAL)
+
+    def __call__(self, document) -> List[str]:
+        for pattern in JASplitter.PATTERNS:
+            document = re.sub(pattern, self.conv_period, document)
+
+        result = []
+        for line in document.split("\n"):
+            line = line.rstrip()
+            line = line.replace("\n", "")
+            line = line.replace("\r", "")
+            line = line.replace("。", "。\n")
+            sentences = line.split("\n")
+
+            for sentence in sentences:
+                if not sentence:
+                    continue
+
+                period_special = JASplitter.PERIOD_SPECIAL
+                period = JASplitter.PERIOD
+                sentence = sentence.replace(period_special, period)
+                result.append(sentence)
+
+        return result
+

process.py

@@ -0,0 +1,100 @@
+""" Script to process raw SQuAD file for Question Generation format
+gsplit -l 1000 -d --additional-suffix=.jsonl train.jsonl train
+gsplit -l 1000 -d --additional-suffix=.jsonl test.jsonl test
+gsplit -l 1000 -d --additional-suffix=.jsonl validation.jsonl validation
+"""
+import json
+import os
+import re
+from tqdm import tqdm
+from typing import List, Dict
+from datasets import load_dataset
+from ja_sentence_split import JASplitter
+
+HIGHLIGHT_TOKEN = '<hl>'
+SPLITTER = JASplitter()
+
+
+def get_sentence(document: str):
+    return [str(s) for s in SPLITTER(document)]
+
+
+def process_single_data(data: Dict):
+    """ Convert single raw json data into QG format """
+    example = {'question': data["question"], 'passage': data["context"]}
+
+    # check answer
+    answer_text = data['answers']['text'][0]
+    answer_start = data['answers']['answer_start'][0]
+    answer_end = answer_start + len(answer_text)
+    assert example['passage'][answer_start: answer_end] == answer_text
+    example['answer'] = answer_text
+
+    # get sentence
+    position = example['passage'].find(example['answer'])
+    assert position != -1
+    before_tmp = get_sentence(example['passage'][:position])
+    if len(before_tmp) == 0:
+        before = ''
+        before_sentence = ''
+    else:
+        if before_tmp[-1].endswith('。'):
+            before = ' '.join(before_tmp)
+            before_sentence = ''
+        else:
+            before = ' '.join(before_tmp[:-1])
+            before_sentence = before_tmp[-1]
+    after_tmp = get_sentence(example['passage'][position + len(example['answer']):])
+    if len(after_tmp) == 0:
+        after = ''
+        after_sentence = ''
+    else:
+        after = ' '.join(after_tmp[1:])
+        after_sentence = after_tmp[0]
+    example['sentence'] = '{}{}{}'.format(before_sentence, example['answer'], after_sentence)
+
+    # get passage_sentence
+    source_text = '{0}{1}{2}{1}{3}'.format(before, HIGHLIGHT_TOKEN, example['sentence'], after)
+    example['passage_sentence'] = re.sub(r'\s+', ' ', source_text)
+
+    # get passage_answer
+    source_text = '{0}{1}{2}{1}{3}'.format(
+        example['passage'][:position], HIGHLIGHT_TOKEN, example['answer'],
+        example['passage'][position + len(example['answer']):])
+    example['passage_answer'] = re.sub(r'\s+', ' ', source_text)
+
+    # get sentence_answer
+    before = get_sentence(example['passage'][:position])
+    if len(before) == 0 or before[-1].endswith('。'):
+        before = ''
+    else:
+        before = before[-1]
+    after = get_sentence(example['passage'][position + len(example['answer']):])
+    if len(after) == 0:
+        after = ''
+    else:
+        after = after[0]
+    source_text = '{0}{1}{2}{1}{3}'.format(before, HIGHLIGHT_TOKEN, example['answer'], after)
+    example['sentence_answer'] = re.sub(r'\s+', ' ', source_text)
+    for _k in example.keys():
+        example[_k] = example[_k].replace('。\n\n', '。').replace('。\n', '。')
+    return example
+
+
+if __name__ == '__main__':
+    jaquad_data = load_dataset("SkelterLabsInc/JaQuAD")
+    data_dev = jaquad_data['validation']
+    data_train = jaquad_data['train']
+    data_train = data_train.shuffle(seed=1)
+    data_test = [data_train[i] for i in range(len(data_dev))]
+    data_train = [data_train[i] for i in range(len(data_dev), len(data_train))]
+
+    data_all = {'train': data_train, 'validation': data_dev, 'test': data_test}
+
+    output = './data/processed'
+    os.makedirs(output, exist_ok=True)
+    for k, _data in data_all.items():
+        with open('{}/{}.jsonl'.format(output, k), 'w') as f:
+            for single_data in tqdm(_data):
+                single_data = process_single_data(single_data)
+                f.write(json.dumps(single_data) + '\n')

qg_jaquad.py

@@ -0,0 +1,72 @@
+import json
+import datasets
+from datasets import load_dataset
+from datasets.tasks import Summarization
+
+logger = datasets.logging.get_logger(__name__)
+_DESCRIPTION = """
+[JaQuAD](https://github.com/SkelterLabsInc/JaQuAD) dataset for question generation (QG) task. The test set of the original 
+data is not publicly released, so we randomly sampled test questions from the training set.
+"""
+_URL = 'https://huggingface.co/datasets/asahi417/qg_jaquad/raw/main/data/processed'
+_URLS = {
+    'train': ['{}/train{:02d}.jsonl'.format(_URL, i) for i in range(28)],
+    'test': ['{}/test{:02d}.jsonl'.format(_URL, i) for i in range(4)],
+    'validation': ['{}/validation{:02d}.jsonl'.format(_URL, i) for i in range(4)]
+}
+
+
+class QGJaquadConfig(datasets.BuilderConfig):
+    """BuilderConfig for SquadQG"""
+
+    def __init__(self, **kwargs):
+        """BuilderConfig for SquadQG.
+        Args:
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(SquadQGConfig, self).__init__(**kwargs)
+
+
+class QGJaquad(datasets.GeneratorBasedBuilder):
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "answer": datasets.Value("string"),
+                    "question": datasets.Value("string"),
+                    "sentence": datasets.Value("string"),
+                    "passage": datasets.Value("string"),
+                    "sentence_answer": datasets.Value("string"),
+                    "passage_answer": datasets.Value("string"),
+                    "passage_sentence": datasets.Value("string")
+                }
+            ),
+            supervised_keys=None,
+            task_templates=[
+                Summarization(task='question generation', text_column="passage_answer", summary_column='question')
+            ],
+            homepage="https://github.com/asahi417/lm-question-generation"
+        )
+
+    def _split_generators(self, dl_manager):
+        downloaded_file = dl_manager.download_and_extract(_URLS)
+        return [
+            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": downloaded_file["train"]}),
+            datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepaths": downloaded_file["validation"]}),
+            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepaths": downloaded_file["test"]}),
+        ]
+
+    def _generate_examples(self, filepaths):
+        _key = 0
+        for filepath in filepaths:
+            logger.info("generating examples from = %s", filepath)
+            with open(filepath, encoding="utf-8") as f:
+                _list = f.read().split('\n')
+                if _list[-1] == '':
+                    _list = _list[:-1]
+                for i in _list:
+                    data = json.loads(i)
+                    yield _key, data
+                    _key += 1