CLTS: A Chinese Long Text Summarization Dataset

Currently, there are two issues in automatic summarization dataset for Chinese: (1) There are fewer Chinese datasets. (2) There is a lack of Chinese long text datasets, and most existing datasets contain short texts, which can only be used to generate a title from short text.

CLTS is a new Chinese long text summarization dataset, extracted from the Chinese news website ThePaper.cn. The resulting version of the dataset contains more than 180,000 long-sequence pairs, where each article consists of multiple paragraphs and each summary consists of multiple sentences.

The dataset is introduced in CLTS: A New Chinese Long Text Summarization Dataset.

The comparison result of CLTS to other datasets is shown in the table below. For Chinese dataset, length is the number of Chinese characters, while for English, it is the number of words.

We select some samples from the dataset and you can see them in samples.txt

We split the corpus into three parts, including training, validation and test set. The data can be downloaded from the link here. And the password is iucp.