CSL: A Large-scale Chinese Scientific Literature Dataset

Proceedings of the 29th International Conference on Computational Linguistics, pages 3917–3923 October 12–17, 2022. 3917 CSL: A Large-scale Chinese Scientific Literature Dataset Yudong Li1,2, Yuqing Zhang1∗, Zhe Zhao3, Linlin Shen2, Weijie Liu3, Weiquan Mao3, and Hui Zhang4 1 China University of Geosciences (Beijing), School of Information Engineering 2 Shenzhen University, School of Computer Science and Software Engineering 3 Tencent AI Lab 4 Information Technology Center for National Science and Technology Infrastructure, Beijing, China Abstract Scientific literature serves as a high-quality corpus, supporting a lot of Natural Language Processing (NLP) research. However, existing datasets are centered around the English lan- guage, which restricts the development of Chi- nese scientific NLP. In this work, we present CSL, a large-scale Chinese Scientific Literature dataset, which contains the titles, abstracts, keywords and academic fields of 396k papers. To our knowledge, CSL is the first scientific document dataset in Chinese. The CSL can serve as a Chinese corpus. Also, this semi- structured data is a natural annotation that can constitute many supervised NLP tasks. Based on CSL, we present a benchmark to evalu- ate the performance of models across scien- tific domain tasks, i.e., summarization, key- word generation and text classification. We analyze the behavior of existing text-to-text models on the evaluation tasks and reveal the challenges for Chinese scientific NLP tasks, which provides a valuable reference for fu- ture research. Data and code are available at https://github.com/ydli-ai/CSL . 1 Introduction With the increase in the publication of papers, Nat- ural Language Processing (NLP) tools that assist users in writing, searching, and archiving scien- tific literature have grown increasingly important. For instance, paper/citation recommendation (Beel et al., 2016; Cohan et al., 2020), topic classification (Beltagy et al., 2019) and summarization (Cohan and Goharian, 2018) systems have been developed. The construction of these automatic systems pri- marily relies on academic resources such as large- scale corpus (Lo et al., 2020; Saier and Färber, 2020), citation graphs (Sinha et al., 2015; Tang et al., 2008; Zhang et al., 2019) and supervised scientific datasets (Li et al., 2016; Jurgens et al., 2018). These resources, however, are mostly cen- tered around the English language, which restricts Title Analysis on the Subjective Well-being of Farmers in Weihai City Abstract To investigate the subjective well-being of farmers in Weihai City and its influencing factors, this paper randomly sampled 4550 farmers by their gender, age, marriage, educational level, occupation, poverty or not. We use a Multivariate Logistic Regression model to evaluate their subjective well-being ... Keywords Farmers; Subjective Well-Being; Multivariate Logistic Regression Category Management Discipline Public Administration Summarization c (a) CSL Example (b) NLP Tasks d Keyword Generation Text Classification A K T A T T A K A K A K A d T T A c K T d Figure 1: (a) An example of paper meta-information (translated into English). (b) Examples of NLP tasks constructed from CSL. The arrow indicates the input and output of the task, for example, “A → T” represents the task feeding abstract to produce …

Content truncated to 500 words