init

conceptnet.py

@@ -0,0 +1,72 @@
+import json
+import datasets
+
+logger = datasets.logging.get_logger(__name__)
+_DESCRIPTION = """[ConceptNet5](https://ojs.aaai.org/index.php/AAAI/article/view/11164)"""
+_NAME = "conceptnet"
+_VERSION = "1.0.0"
+_CITATION = """
+@inproceedings{speer2017conceptnet,
+  title={Conceptnet 5.5: An open multilingual graph of general knowledge},
+  author={Speer, Robyn and Chin, Joshua and Havasi, Catherine},
+  booktitle={Thirty-first AAAI conference on artificial intelligence},
+  year={2017}
+}
+"""
+
+_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'],
+}
+
+
+class ConceptNetHighConfidenceConfig(datasets.BuilderConfig):
+    """BuilderConfig"""
+
+    def __init__(self, **kwargs):
+        """BuilderConfig.
+        Args:
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(ConceptNetHighConfidenceConfig, self).__init__(**kwargs)
+
+
+class ConceptNetHighConfidence(datasets.GeneratorBasedBuilder):
+    """Dataset."""
+
+    BUILDER_CONFIGS = [
+        ConceptNetHighConfidenceConfig(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[str(i)]})
+                for i in [datasets.Split.TRAIN, datasets.Split.VALIDATION]]
+
+    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 +1,30 @@
+import pandas as pd
+from datasets import load_dataset
+
+data = load_dataset('relbert/conceptnet')
+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')
+df.to_csv('stats.csv')
+with open('stats.md', 'w') as f:
+    f.write(df.to_markdown())
+
+

process.py

@@ -1,7 +1,7 @@
 import os
 import json
 from tqdm import tqdm
-
+import numpy as np
 from datasets import load_dataset
 
 export_dir = 'dataset'
@@ -30,23 +30,33 @@ def check(example):
 
 dataset = dataset.filter(lambda example: check(example))
 relations = list(set(dataset["rel"]))
+all_word = [os.path.basename(i) for i in dataset['arg1'] + dataset['arg2']]
+t, c = np.unique(all_word, return_counts=True)
+freq = {_t: _c for _t, _c in zip(t, c)}
+
+
+def freq_filter(example):  # filter by entity frequency
+    if freq[os.path.basename(example['arg1'])] < 5:
+        return False
+    if freq[os.path.basename(example['arg2'])] < 5:
+        return False
+    return True
+
 
-for r in tqdm(relations):
-
-    _dataset = dataset.filter(lambda example: example['rel'] == r)
-    _dataset = _dataset.shuffle(0)
-    train_size = int(len(_dataset) * 0.7)
-
-    with open(f"{export_dir}/train.{os.path.basename(r)}.jsonl", 'w') as f:
-        f.write(json.dumps({
-            'relation_type': os.path.basename(r),
-            'positives': [[i['arg1'], i['arg2']] for i in _dataset[:train_size]],
-            'negatives': []
-        }))
-
-    with open(f"{export_dir}/valid.{os.path.basename(r)}.jsonl", 'w') as f:
-        f.write(json.dumps({
-            'relation_type': os.path.basename(r),
-            'positives': [[i['arg1'], i['arg2']] for i in _dataset[train_size:]],
-            'negatives': []
-        }))
+with open(f"{export_dir}/train.jsonl", 'w') as f_train:
+    with open(f"{export_dir}/valid.jsonl", 'w') as f_valid:
+        for r in tqdm(relations):
+            _dataset = dataset.filter(lambda example: example['rel'] == r).shuffle(0)
+            pairs = [[os.path.basename(i['arg1']), os.path.basename(i['arg2'])] for i in _dataset if freq_filter(i)]
+            train_size = int(len(_dataset) * 0.7)
+            f_train.write(json.dumps({
+                'relation_type': os.path.basename(r),
+                'positives': pairs[:train_size],
+                'negatives': []
+            }))
+            if len(pairs[train_size:]) > 0:
+                f_valid.write(json.dumps({
+                    'relation_type': os.path.basename(r),
+                    'positives': pairs[train_size:],
+                    'negatives': []
+                }))