Gossiping Chinese Corpus

PTT Chinese corpus

? News: We used a similar data collection strategy to train a large-scale language model for Traditional Chinese. For model performance and usage, please refer to Kyara

---

Hi, here is the PTT Chinese corpus. I used certain assumptions and methods to simplify each article into a question and answer pairing, where the questions come from the title of the article, and the replies are the tweets of the article. Unfortunately, the noise of this data set is still a bit loud at present. If you have a better way to extract the question and answer pairs of articles, or if you find any parts of this data set that can be improved, please contact me. Also I wish you all smooth development:>

Information description

There are two data sets, which can be obtained from PTT-Gossiping-Corpus or from the data folder of this project.

Gossiping-QA-Dataset.txt

Collected articles from PTT Bagua Edition from 2015 to June 2017. Each line is a question and answer pairing, and the question and answer are separated by a tab ( t ) section, for example

 matlab有什麼炫砲一點的圖？	一樣的圖改一改顏色，有點半透明感覺更唬爛炫
有沒有情人節吃什麼cp值最高的八卦	吃屎啊廢話 免費的一餐
姆咪一個人守得住街亭嗎?	引來一堆肥宅穢土轉生 有機會喔
有沒有被落石砸到該反省的八卦	蔡英文執政就故意誇大報導 東森不意外
情人節該帶女朋友去哪慶祝？	用了一整年 對她好一點  送專業乾洗店吧
為什麼 聖結石 會被酸而 這群人 不會？	質感 劇本 成員 都差很多好嗎 不要拿腎結石來污辱這群人
為什麼慶祝228會被罵可是慶端午不會？	因為屈原不是台灣人，是楚國人。
有沒有戰神阿瑞斯的八卦?	爵士就是阿瑞斯 男主角最後死了
理論與實務最脫節的系	哪個系不脫節...你問最不脫節的簡單多了...
為什麼PTT這麼多人看棒球	肥宅才看棒球　系壘一堆胖子
為什麼達摩祖師傳那麼好看?	達摩從頭到尾都是被動 (別人問他問題

There are currently 418,202 Q&A matches, but not all matches are valid because some articles do not have tweets. Replies to such questions will be marked as沒有資料(650 in total). Please be careful when using them.

Gossiping-QA-Dataset-2_0.csv

The new version of the data set is expanded from Gossiping-QA-Dataset.txt, adding some articles from 2018 and 2019, and contains a total of 774,114 question and answer pairs. The data format is adjusted to csv, which contains two columns: question and answer . It can be easily loaded with pandas:

 import pandas as pd
qa_corpus = pd.read_csv("data/Gossiping-QA-Dataset-2_0.csv")
qa_corpus.head()

# question	answer
# 173763	今年房市是多還是空？	有台北市的房子 再來討論
# 216985	有沒有台灣高登拉姆齊-劉一帆的八卦?	餐飲界賈伯斯
# 679552	什麼時候會覺得台灣治安很差？	台灣一年輾死三千多人
# 349323	有沒有早上起來運動很舒服的八卦？	肥宅每天都在運動,手部運動
# 42228	迪士尼還有什麼片子適合拍成真人版	想說我剛剛不是進sex版啊...

Citation

If you think the dataset is helpful for your research, please cite our project:

 @misc{
    kai-chou yang_2019,
    title={PTT-Gossiping-Corpus},
    url={https://www.kaggle.com/dsv/676336},
    DOI={10.34740/DVS/676336},
    publisher={Kaggle},
    author={Kai-Chou Yang},
    year={2019}
}