TableNet: An Approach for Determining Fine-grained Relations for Wikipedia Tables

This repository contains the code for extracting tables from Wikipedia, and additionally contains the models for determining if for a pair of tables one of the relations subPartOf or equivalent holds.

HTML Article Extraction

To extract the HTML content of the Wikipedia articles, please use the following command:

java -cp *COMPILED_PROJECT*.jar HTMLTableExtractor -option crawl_table_articles -seed SEED_FILE -out OUT_FILE

• SEED_FILE: should consist of a list of articles that you are interested in extracting the HTML content. Each line contains an article name.
• OUT_FILE: the output file where the extracted HTML content is stored.

Table Extraction

To extract the tables, please use the following command:

java -cp *COMPILED_PROJECT*.jar HTMLTableExtractor -option parse_tables -in INPUT_FILE -out OUT_FILE

• INPUT_FILE: should point to the HTML content of the Wikipedia articles. The HTML content is JSON escaped, such that each line consists of an article. You can use the previous operation (*HTML Article Extraction*) to extract the HTML content.
• OUTPUT_FILE: the file where the extracted tables will be outputed in JSON format.

Table Alignment

We have uploaded all the datasets for the TableNet evaluation as well as the extracted tables at table data. The TableNet code for alignment of tables can be found here.

Citation

Please cite the following work, when using this dataset or approach.

@article{fetahu2019tablenet,
  title={TableNet: An Approach for Determining Fine-grained Relations for Wikipedia Tables},
  author={Fetahu, Besnik and Anand, Avishek and Koutraki, Maria},
  booktitle={Proceedings of the 2019 World Wide Web Conference on World Wide Web, {WWW} 2019, San Francisco, USA, May 13-17},
  year={2019}
}