Upload open2ch.py

open2ch.py

@@ -0,0 +1,358 @@
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""おーぷん2ちゃんねる対話コーパス"""
+
+from typing import Iterable
+
+import csv
+import os
+
+import datasets as ds
+
+
+_CITATION = """\
+@inproceedings{open2chdlc2019,
+  title={おーぷん2ちゃんねる対話コーパスを用いた用例ベース対話システム},
+  author={稲葉 通将},
+  booktitle={第87回言語・音声理解と対話処理研究会(第10回対話システムシンポジウム), 人工知能学会研究会資料 SIG-SLUD-B902-33},
+  pages={129--132},
+  year={2019}
+}
+"""
+
+_DESCRIPTION = """\
+おーぷん2ちゃんねるの「なんでも実況(ジュピター)」「ニュー速VIP」「ニュース速報+」の3つの掲示板をクロールして作成した対話コーパスの HuggingFace Datasets 変換版です。
+"""
+
+_HOMEPAGE = "https://github.com/1never/open2ch-dialogue-corpus"
+
+_LICENSE = "Apache-2.0"
+
+_URLS = {
+    "corpus": "https://media.githubusercontent.com/media/1never/open2ch-dialogue-corpus/master/corpus.zip",
+    "ranking": "https://raw.githubusercontent.com/1never/open2ch-dialogue-corpus/master/ranking.zip",
+    "ng_words": "https://raw.githubusercontent.com/1never/open2ch-dialogue-corpus/master/data/ng_words.txt",
+}
+
+
+# Load corpus data from file
+def load_corpus(filepath: str, ng_words: list[str] = []) -> Iterable[list]:
+    with open(filepath, "r", encoding="utf-8") as f:
+        reader = csv.reader(f, delimiter="\t")
+
+        for row in reader:
+            dialogue = []
+            should_skip = False
+
+            for i, content in enumerate(row):
+                for word in ng_words:  # NG ワード含んでいたらスキップ
+                    if word in content:
+                        should_skip = True
+                        break
+                if should_skip:
+                    break
+
+                if i % 2 == 0:
+                    dialogue.append(
+                        {"speaker": 1, "content": content.replace("__BR__", "\n")}
+                    )
+                else:
+                    dialogue.append(
+                        {"speaker": 2, "content": content.replace("__BR__", "\n")}
+                    )
+
+            if should_skip:
+                continue
+
+            yield dialogue
+
+
+# Load ranking data from file
+def load_ranking(filepath: str, split: str) -> Iterable[dict]:
+    with open(os.path.join(filepath, f"{split}.tsv"), "r", encoding="utf-8") as f:
+        reader = csv.reader(f, delimiter="\t")
+
+        data = {
+            "dialogue": [],
+            "next": {},
+            "random": [],
+        }
+        dialogue_len = 0
+
+        for i, row in enumerate(reader):
+            if i % 10 == 0:
+                dialogue_len = len(row)
+                for j, content in enumerate(row[1:-1]):  # [0] は使わない
+                    if j % 2 == 0:
+                        data["dialogue"].append(
+                            {"speaker": 1, "content": content.replace("__BR__", "\n")}
+                        )
+                    else:
+                        data["dialogue"].append(
+                            {"speaker": 2, "content": content.replace("__BR__", "\n")}
+                        )
+                data["next"] = {
+                    "speaker": 1 if dialogue_len % 2 == 1 else 2,
+                    "content": row[-1].replace("__BR__", "\n"),
+                }
+
+            elif i % 10 >= 1 and i % 10 <= 8:
+                data["random"].append(row[-1].replace("__BR__", "\n"))
+            elif i % 10 == 9:
+                data["random"].append(row[-1].replace("__BR__", "\n"))
+
+                yield data
+
+                data = {
+                    "dialogue": [],
+                    "next": {},
+                    "random": [],
+                }
+                dialogue_len = 0
+
+
+# Load NG words from file
+def load_ng_words(filepath) -> list[str]:
+    data = []
+    with open(filepath, "r", encoding="utf-8") as f:
+        for line in f:
+            data.append(line.strip())
+    return data
+
+
+class Open2chDataset(ds.GeneratorBasedBuilder):
+    OVER_SIZE_LIMIT = 6_000_000
+    csv.field_size_limit(OVER_SIZE_LIMIT)
+
+    VERSION = ds.Version("1.0.0")
+
+    BUILDER_CONFIGS = [
+        ds.BuilderConfig(
+            name="all-corpus",
+            version=VERSION,
+            description="All concatenated corpus data",
+        ),
+        ds.BuilderConfig(
+            name="livejupiter",
+            version=VERSION,
+            description="A subset of corpus data from livejupiter board",
+        ),
+        ds.BuilderConfig(
+            name="news4vip",
+            version=VERSION,
+            description="A subset of corpus data from news4vip board",
+        ),
+        ds.BuilderConfig(
+            name="newsplus",
+            version=VERSION,
+            description="A subset of corpus data from newsplus board",
+        ),
+        ds.BuilderConfig(
+            name="all-corpus-cleaned",
+            version=VERSION,
+            description="Cleaned version of all concatenated corpus data",
+        ),
+        ds.BuilderConfig(
+            name="livejupiter-cleaned",
+            version=VERSION,
+            description="Cleaned version of a subset of corpus data from livejupiter board",
+        ),
+        ds.BuilderConfig(
+            name="news4vip-cleaned",
+            version=VERSION,
+            description="Cleaned version of a subset of corpus data from news4vip board",
+        ),
+        ds.BuilderConfig(
+            name="newsplus-cleaned",
+            version=VERSION,
+            description="Cleaned version of a subset of corpus data from newsplus board",
+        ),
+        ds.BuilderConfig(
+            name="ranking",
+            version=VERSION,
+            description="Dataset for response ranking task",
+        ),
+    ]
+
+    DEFAULT_CONFIG_NAME = "all-corpus"
+
+    def _info(self):
+        if self.config.name in ["all-corpus", "all-corpus-cleaned"]:
+            features = ds.Features(
+                {
+                    "dialogue": ds.Sequence(
+                        {
+                            "speaker": ds.Value("int8"),  # 1 or 2
+                            "content": ds.Value("string"),
+                        }
+                    ),
+                    "board": ds.Value("string"),
+                }
+            )
+        elif self.config.name in [
+            "livejupiter",
+            "news4vip",
+            "newsplus",
+            "livejupiter-cleaned",
+            "news4vip-cleaned",
+            "newsplus-cleaned",
+        ]:
+            features = ds.Features(
+                {
+                    "dialogue": ds.Sequence(
+                        {
+                            "speaker": ds.Value("int8"),  # 1 or 2
+                            "content": ds.Value("string"),
+                        }
+                    ),
+                }
+            )
+        elif self.config.name == "ranking":
+            features = ds.Features(
+                {
+                    "dialogue": ds.Sequence(
+                        {
+                            "speaker": ds.Value("int8"),  # 1 or 2
+                            "content": ds.Value("string"),
+                        }
+                    ),
+                    "next": {  # 正しい返答
+                        "speaker": ds.Value("int8"),  # 1 or 2
+                        "content": ds.Value("string"),
+                    },
+                    "random": ds.Sequence(  # ランダムに選ばれた返答
+                        ds.Value("string"),
+                    ),
+                }
+            )
+        else:
+            raise ValueError("Invalid config name")
+
+        return ds.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        corpus_url = _URLS["corpus"]
+        ng_words_url = _URLS["ng_words"]
+
+        if self.config.name in ["all-corpus", "livejupiter", "news4vip", "newsplus"]:
+            data_dir = dl_manager.download_and_extract(corpus_url)
+
+            return [
+                ds.SplitGenerator(
+                    name=ds.Split.TRAIN,
+                    gen_kwargs={
+                        "filepath": os.path.join(data_dir, "corpus"),
+                    },
+                ),
+            ]
+        elif self.config.name in [
+            "all-corpus-cleaned",
+            "livejupiter-cleaned",
+            "news4vip-cleaned",
+            "newsplus-cleaned",
+        ]:
+            data_dir = dl_manager.download_and_extract(corpus_url)
+            ng_words_dir = dl_manager.download(ng_words_url)
+
+            return [
+                ds.SplitGenerator(
+                    name=ds.Split.TRAIN,
+                    gen_kwargs={
+                        "filepath": os.path.join(data_dir, "corpus"),
+                        "ng_words_path": ng_words_dir,
+                    },
+                ),
+            ]
+        elif self.config.name == "ranking":
+            ranking_url = _URLS["ranking"]
+            data_dir = dl_manager.download_and_extract(ranking_url)
+
+            return [
+                ds.SplitGenerator(
+                    name=ds.Split.TRAIN,
+                    gen_kwargs={
+                        "filepath": os.path.join(data_dir, "ranking"),
+                        "split": "dev",
+                    },
+                ),
+                ds.SplitGenerator(
+                    name=ds.Split.TEST,
+                    gen_kwargs={
+                        "filepath": os.path.join(data_dir, "ranking"),
+                        "split": "test",
+                    },
+                ),
+            ]
+
+    def _generate_examples(
+        self,
+        filepath: str,
+        split: str = "train",
+        ng_words_path: str | None = None,
+    ):
+        livejupiter_path = os.path.join(filepath, "livejupiter.tsv")
+        news4vip_path = os.path.join(filepath, "news4vip.tsv")
+        newsplus_path = os.path.join(filepath, "newsplus.tsv")
+
+        ng_words = load_ng_words(ng_words_path) if ng_words_path is not None else []
+
+        if self.config.name == "all-corpus":
+            for i, data in enumerate(load_corpus(livejupiter_path, ng_words)):
+                yield f"livejupiter-{i}", {
+                    "dialogue": data,
+                    "board": "livejupiter",
+                }
+
+            for i, data in enumerate(load_corpus(news4vip_path, ng_words)):
+                yield f"news4vip-{i}", {
+                    "dialogue": data,
+                    "board": "news4vip",
+                }
+
+            for i, data in enumerate(load_corpus(newsplus_path, ng_words)):
+                yield f"newsplus-{i}", {
+                    "dialogue": data,
+                    "board": "newsplus",
+                }
+
+        elif self.config.name.startswith("livejupiter"):
+            for i, data in enumerate(load_corpus(livejupiter_path, ng_words)):
+                yield f"livejupiter-{i}", {
+                    "dialogue": data,
+                }
+
+        elif self.config.name.startswith("news4vip"):
+            for i, data in enumerate(load_corpus(news4vip_path, ng_words)):
+                yield f"news4vip-{i}", {
+                    "dialogue": data,
+                }
+
+        elif self.config.name.startswith("newsplus"):
+            for i, data in enumerate(load_corpus(newsplus_path, ng_words)):
+                yield f"newsplus-{i}", {
+                    "dialogue": data,
+                }
+
+        elif self.config.name == "ranking":
+            for i, data in enumerate(load_ranking(filepath, split)):
+                yield f"ranking-{split}-{i}", data
+
+        else:
+            raise ValueError("Invalid config name")