sergiog95/csabstracts

Dataset of scientific abstracts for the purpose of sentence classification

CS Abstracts Dataset

Dataset of scientific abstracts for the purpose of sentence classification.

The dataset is composed by a total of 654 abstracts, which were collected from the arXiv platform. Then, using crowdsourcing and collective inteligence, the data was annotated, where the sentences were categorized in the following classes:

• Background
• Objective
• Methods
• Results
• Conclusions

All the abstracts are from the Computer Science field (hence the name CS Abstracts). The following table details the composition of the dataset, with respective numbers of abstracts and sentences splitted by training set, validation set and test set:

	Training set	Validation set	Test set
#abstracts	500	77	77
#sentences	3287	824	619

The dataset is split in three files (train.txt, validation.txt and test.txt).

Each entry in the file corresponds to one entry in the dataset and each column is separated by a tab. The first column is the position of the sentence of the abstracts, the second column is the label/category of the sentence of the abstracts, and the third column is the text itself of the abstract sentence.

The dataset was developed in the context of my Master's Thesis in Engineering and Management of Information Systems.

Citation Request

This dataset is made freely available for research purposes. Please include this citation if you plan to use this database:

Gonçalves, S., Cortez, P., & Moro, S. (2019). A deep learning classifier for sentence classification in biomedical and computer science abstracts. In Neural Computing and Applications, In press, http://dx.doi.org/10.1007/s00521-019-04334-2