jd-coderepos / task-dataset-metric-extraction

Parses pdfs for Tasks, Datasets, and Metrics using GROBID (Adapted version of https://github.com/IBM/science-result-extractor)

Task Dataset Metric Software Extraction

This program produces the test data for classification over a set of predefined task#dataset#metrics#software labels.

Given input a pdf file, it scrapes the text from the filw using the Grobid parser, subsequently generating the test data file for input to the neural network classifier.

Steps to run the program

1. Clone this repository (https://github.com/jenlindadsouza/task-dataset-metric-extraction)
2. The program jar file can be found at https://github.com/jenlindadsouza/task-dataset-metric-extraction/tree/master/build/libs
        - It is called tdm-1.0.jar
3. Usage: java -jar tdm-1.0.jar <pdf-file-path> <resources-dir> Where the pdf-file-path is the input pdf article to extract task, dataset, metric, software elements from and resources-dir should point to the directory at https://github.com/jenlindadsouza/task-dataset-metric-extraction/tree/master/src/main/resources
4. To run directly from the source code, the build file will have to be updated for the local dependencies for jar files. See here: https://github.com/jenlindadsouza/task-dataset-metric-extraction/blob/master/build.gradle
5. Either the jars can be obtained from source. For convenience, they are also made available here https://drive.google.com/drive/folders/1ax7ah8AInoD-_7amx27VKaSrFV73ggEk?usp=sharing

Acknowledgement:

This program reuses code modules from IBM's science-result-extractor (https://github.com/IBM/science-result-extractor). A reference url to their paper on the ACL anthology is https://www.aclweb.org/anthology/P19-1513