README

.gitignore

@@ -0,0 +1 @@
+cache

README.md

@@ -0,0 +1,82 @@
+---
+language:
+- en
+license:
+- other
+multilinguality:
+- monolingual
+size_categories:
+- 1K<n<10K
+pretty_name: ConceptNet with High Confidence
+---
+# Dataset Card for "relbert/conceptnet_high_confidence"
+## Dataset Description
+- **Repository:** [RelBERT](https://github.com/asahi417/relbert)
+- **Paper:** [https://home.ttic.edu/~kgimpel/commonsense.html](https://home.ttic.edu/~kgimpel/commonsense.html)
+- **Dataset:** High Confidence Subset of ConceptNet
+
+### Dataset Summary
+The selected subset of ConceptNet used in [this work](https://home.ttic.edu/~kgimpel/commonsense.html), which compiled 
+to fine-tune [RelBERT](https://github.com/asahi417/relbert) model.
+
+## Dataset Structure
+### Data Instances
+An example of `train` looks as follows.
+```
+{
+    "relation_type": "AtLocation",
+    "positives": [["fish", "water"], ["cloud", "sky"], ["child", "school"], ... ],
+    "negatives": [["pen", "write"], ["sex", "fun"], ["soccer", "sport"], ["fish", "school"], ... ]
+}
+```
+
+### Data Splits
+|  name   |train|validation|
+|---------|----:|---------:|
+|conceptnet_high_confidence| 25 |      24|
+
+### Number of Positive/Negative Word-pairs in each Split
+
+| relation_type    |   positive (train) |   negative (train) |   positive (validation) |   negative (validation) |
+|:-----------------|-------------------:|-------------------:|------------------------:|------------------------:|
+| AtLocation       |                383 |               1768 |                      97 |                     578 |
+| CapableOf        |                195 |               1790 |                      73 |                     600 |
+| Causes           |                 71 |               1797 |                      26 |                     595 |
+| CausesDesire     |                  9 |               1793 |                      11 |                     595 |
+| CreatedBy        |                  2 |               1796 |                       0 |                       0 |
+| DefinedAs        |                  0 |                  0 |                       2 |                     595 |
+| Desires          |                 16 |               1794 |                      12 |                     595 |
+| HasA             |                 67 |               1814 |                      17 |                     595 |
+| HasFirstSubevent |                  2 |               1796 |                       0 |                       0 |
+| HasLastSubevent  |                  2 |               1796 |                       3 |                     593 |
+| HasPrerequisite  |                168 |               1803 |                      57 |                     592 |
+| HasProperty      |                 94 |               1801 |                      39 |                     605 |
+| HasSubevent      |                125 |               1798 |                      40 |                     609 |
+| IsA              |                310 |               1764 |                      98 |                     603 |
+| MadeOf           |                 17 |               1793 |                       7 |                     593 |
+| MotivatedByGoal  |                 14 |               1796 |                      11 |                     595 |
+| NotCapableOf     |                 15 |               1793 |                       0 |                       0 |
+| NotDesires       |                  4 |               1795 |                       4 |                     592 |
+| PartOf           |                 34 |               1801 |                       7 |                     593 |
+| ReceivesAction   |                 18 |               1793 |                       8 |                     593 |
+| SymbolOf         |                  0 |                  0 |                       2 |                     596 |
+| UsedFor          |                249 |               1815 |                      81 |                     588 |
+| SUM              |               1795 |              35896 |                     595 |                   11305 |
+
+### Citation Information
+```
+@InProceedings{P16-1137,
+  author = 	"Li, Xiang
+		and Taheri, Aynaz
+		and Tu, Lifu
+		and Gimpel, Kevin",
+  title = 	"Commonsense Knowledge Base Completion",
+  booktitle = 	"Proceedings of the 54th Annual Meeting of the Association for      Computational Linguistics (Volume 1: Long Papers)    ",
+  year = 	"2016",
+  publisher = 	"Association for Computational Linguistics",
+  pages = 	"1445--1455",
+  location = 	"Berlin, Germany",
+  doi = 	"10.18653/v1/P16-1137",
+  url = 	"http://aclweb.org/anthology/P16-1137"
+}
+```

augment_negative.py

@@ -0,0 +1,12 @@
+import json
+from glob import glob
+from itertools import chain
+
+for i in glob("dataset/*.jsonl"):
+    with open(i) as f:
+        tmp = [json.loads(o) for o in f.read().split('\n') if len(o) > 0]
+    for r in tmp:
+        r['negatives'] = r['negatives'] + list(
+            chain(*[o['positives'] for o in tmp if o['relation_type'] != r['relation_type']]))
+    with open(i, 'w') as f:
+        f.write('\n'.join([json.dumps(r) for r in tmp]))

conceptnet_high_confidence_v2.py

@@ -0,0 +1,86 @@
+import json
+import datasets
+
+logger = datasets.logging.get_logger(__name__)
+_DESCRIPTION = """[ConceptNet with high confidence](https://home.ttic.edu/~kgimpel/commonsense.html)"""
+_NAME = "conceptnet_high_confidence"
+_VERSION = "2.0.0"
+_CITATION = """
+@inproceedings{li-16,
+title = {Commonsense Knowledge Base Completion},
+author = {Xiang Li and Aynaz Taheri and Lifu Tu and Kevin Gimpel},
+booktitle = {Proc. of ACL},
+year = {2016}
+}
+@InProceedings{P16-1137,
+  author = 	"Li, Xiang
+		and Taheri, Aynaz
+		and Tu, Lifu
+		and Gimpel, Kevin",
+  title = 	"Commonsense Knowledge Base Completion",
+  booktitle = 	"Proceedings of the 54th Annual Meeting of the Association for      Computational Linguistics (Volume 1: Long Papers)    ",
+  year = 	"2016",
+  publisher = 	"Association for Computational Linguistics",
+  pages = 	"1445--1455",
+  location = 	"Berlin, Germany",
+  doi = 	"10.18653/v1/P16-1137",
+  url = 	"http://aclweb.org/anthology/P16-1137"
+}
+"""
+
+_HOME_PAGE = "https://github.com/asahi417/relbert"
+_URL = f'https://huggingface.co/datasets/relbert/{_NAME}/raw/main/dataset'
+_URLS = {
+    str(datasets.Split.TRAIN): [f'{_URL}/train.jsonl'],
+    str(datasets.Split.VALIDATION): [f'{_URL}/valid.jsonl'],
+    "full": [f'{_URL}/valid.jsonl', f'{_URL}/train.jsonl'],
+}
+
+
+class ConceptNetHighConfidenceV2Config(datasets.BuilderConfig):
+    """BuilderConfig"""
+
+    def __init__(self, **kwargs):
+        """BuilderConfig.
+        Args:
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(ConceptNetHighConfidenceV2Config, self).__init__(**kwargs)
+
+
+class ConceptNetHighConfidenceV2(datasets.GeneratorBasedBuilder):
+    """Dataset."""
+
+    BUILDER_CONFIGS = [
+        ConceptNetHighConfidenceV2Config(name=_NAME, version=datasets.Version(_VERSION), description=_DESCRIPTION),
+    ]
+
+    def _split_generators(self, dl_manager):
+        downloaded_file = dl_manager.download_and_extract(_URLS)
+        return [datasets.SplitGenerator(name=i, gen_kwargs={"filepaths": downloaded_file[i]}) for i in _URLS.keys()]
+
+    def _generate_examples(self, filepaths):
+        _key = 0
+        for filepath in filepaths:
+            logger.info(f"generating examples from = {filepath}")
+            with open(filepath, encoding="utf-8") as f:
+                _list = [i for i in f.read().split('\n') if len(i) > 0]
+                for i in _list:
+                    data = json.loads(i)
+                    yield _key, data
+                    _key += 1
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "relation_type": datasets.Value("string"),
+                    "positives": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
+                    "negatives": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
+                }
+            ),
+            supervised_keys=None,
+            homepage=_HOME_PAGE,
+            citation=_CITATION,
+        )

get_stats.py

@@ -0,0 +1,35 @@
+import pandas as pd
+from datasets import load_dataset
+
+data = load_dataset('relbert/conceptnet_high_confidence')
+stats = []
+for k in data.keys():
+    for i in data[k]:
+        stats.append({'relation_type': i['relation_type'], 'split': k, 'positives': len(i['positives']), 'negatives': len(i['negatives'])})
+df = pd.DataFrame(stats)
+df_train = df[df['split'] == 'train']
+df_valid = df[df['split'] == 'validation']
+stats = []
+for r in df['relation_type'].unique():
+    _df_t = df_train[df_train['relation_type'] == r]
+    _df_v = df_valid[df_valid['relation_type'] == r]
+    stats.append({
+        'relation_type': r,
+        'positive (train)': 0 if len(_df_t) == 0 else _df_t['positives'].values[0],
+        'negative (train)': 0 if len(_df_t) == 0 else _df_t['negatives'].values[0],
+        'positive (validation)': 0 if len(_df_v) == 0 else _df_v['positives'].values[0],
+        'negative (validation)': 0 if len(_df_v) == 0 else _df_v['negatives'].values[0],
+    })
+
+df = pd.DataFrame(stats).sort_values(by=['relation_type'])
+df.index = df.pop('relation_type')
+sum_pairs = df.sum(0)
+df = df.T
+df['SUM'] = sum_pairs
+df = df.T
+
+df.to_csv('stats.csv')
+with open('stats.md', 'w') as f:
+    f.write(df.to_markdown())
+
+

process.py

@@ -0,0 +1,74 @@
+import json
+import os
+import gzip
+import requests
+
+import pandas as pd
+
+urls = {
+    'dev1': 'https://home.ttic.edu/~kgimpel/comsense_resources/dev1.txt.gz',
+    'dev2': 'https://home.ttic.edu/~kgimpel/comsense_resources/dev2.txt.gz',
+    'test': 'https://home.ttic.edu/~kgimpel/comsense_resources/test.txt.gz'
+}
+exclude = ['NotCapable', 'NotDesires']
+
+
+def wget(url, cache_dir: str = './cache'):
+    """ wget and uncompress data_iterator """
+    os.makedirs(cache_dir, exist_ok=True)
+    filename = os.path.basename(url)
+    path = f'{cache_dir}/{filename}'
+    if os.path.exists(path):
+        return path.replace('.gz', '')
+    with open(path, "wb") as f:
+        r = requests.get(url)
+        f.write(r.content)
+    with gzip.open(path, 'rb') as f:
+        with open(path.replace('.gz', ''), 'wb') as f_write:
+            f_write.write(f.read())
+    os.remove(path)
+    return path.replace('.gz', '')
+
+
+def read_file(file_name):
+    with open(file_name) as f_reader:
+        df = pd.DataFrame([i.split('\t') for i in f_reader.read().split('\n') if len(i) > 0],
+                          columns=['relation', 'head', 'tail', 'flag'])
+    df_positive = df[df['flag'] == '1']
+    df_negative = df[df['flag'] == '0']
+    df_positive.pop('flag')
+    df_negative.pop('flag')
+    return df_positive, df_negative
+
+
+if __name__ == '__main__':
+    test_p, test_n = read_file(wget(urls['test']))
+    dev1_p, dev1_n = read_file(wget(urls['dev1']))
+    train_p = pd.concat([test_p, dev1_p])
+    train_n = pd.concat([test_n, dev1_n])
+    with open(f'dataset/train.jsonl', 'w') as f:
+        for relation, df_p in train_p.groupby('relation'):
+            if len(df_p) < 2:
+                continue
+            if relation in exclude:
+                continue
+            df_n = train_n[train_n['relation'] == relation]
+            f.write(json.dumps({
+                'relation_type': relation,
+                'positives': df_p[['head', 'tail']].to_numpy().tolist(),
+                'negatives': df_n[['head', 'tail']].to_numpy().tolist()
+            }) + '\n')
+
+    dev2_p, dev2_n = read_file(wget(urls['dev2']))
+    with open(f'dataset/valid.jsonl', 'w') as f:
+        for relation, df_p in dev2_p.groupby('relation'):
+            if len(df_p) < 2:
+                continue
+            if relation in exclude:
+                continue
+            df_n = dev2_n[dev2_n['relation'] == relation]
+            f.write(json.dumps({
+                'relation_type': relation,
+                'positives': df_p[['head', 'tail']].to_numpy().tolist(),
+                'negatives': df_n[['head', 'tail']].to_numpy().tolist()
+            }) + '\n')

stats.csv

@@ -0,0 +1,24 @@
+relation_type,positive (train),negative (train),positive (validation),negative (validation)
+AtLocation,383,1768,97,578
+CapableOf,195,1790,73,600
+Causes,71,1797,26,595
+CausesDesire,9,1793,11,595
+CreatedBy,2,1796,0,0
+DefinedAs,0,0,2,595
+Desires,16,1794,12,595
+HasA,67,1814,17,595
+HasFirstSubevent,2,1796,0,0
+HasLastSubevent,2,1796,3,593
+HasPrerequisite,168,1803,57,592
+HasProperty,94,1801,39,605
+HasSubevent,125,1798,40,609
+IsA,310,1764,98,603
+MadeOf,17,1793,7,593
+MotivatedByGoal,14,1796,11,595
+NotCapableOf,15,1793,0,0
+NotDesires,4,1795,4,592
+PartOf,34,1801,7,593
+ReceivesAction,18,1793,8,593
+SymbolOf,0,0,2,596
+UsedFor,249,1815,81,588
+SUM,1795,35896,595,11305

stats.md

@@ -0,0 +1,25 @@
+| relation_type    |   positive (train) |   negative (train) |   positive (validation) |   negative (validation) |
+|:-----------------|-------------------:|-------------------:|------------------------:|------------------------:|
+| AtLocation       |                383 |               1768 |                      97 |                     578 |
+| CapableOf        |                195 |               1790 |                      73 |                     600 |
+| Causes           |                 71 |               1797 |                      26 |                     595 |
+| CausesDesire     |                  9 |               1793 |                      11 |                     595 |
+| CreatedBy        |                  2 |               1796 |                       0 |                       0 |
+| DefinedAs        |                  0 |                  0 |                       2 |                     595 |
+| Desires          |                 16 |               1794 |                      12 |                     595 |
+| HasA             |                 67 |               1814 |                      17 |                     595 |
+| HasFirstSubevent |                  2 |               1796 |                       0 |                       0 |
+| HasLastSubevent  |                  2 |               1796 |                       3 |                     593 |
+| HasPrerequisite  |                168 |               1803 |                      57 |                     592 |
+| HasProperty      |                 94 |               1801 |                      39 |                     605 |
+| HasSubevent      |                125 |               1798 |                      40 |                     609 |
+| IsA              |                310 |               1764 |                      98 |                     603 |
+| MadeOf           |                 17 |               1793 |                       7 |                     593 |
+| MotivatedByGoal  |                 14 |               1796 |                      11 |                     595 |
+| NotCapableOf     |                 15 |               1793 |                       0 |                       0 |
+| NotDesires       |                  4 |               1795 |                       4 |                     592 |
+| PartOf           |                 34 |               1801 |                       7 |                     593 |
+| ReceivesAction   |                 18 |               1793 |                       8 |                     593 |
+| SymbolOf         |                  0 |                  0 |                       2 |                     596 |
+| UsedFor          |                249 |               1815 |                      81 |                     588 |
+| SUM              |               1795 |              35896 |                     595 |                   11305 |