datasets

Datasets

Information about the datasets in this directory.

The structure of this directory is as follows:

📦datasets
 ┣ 📂KGs
 ┃ ┣ 📂fusion_bilingual_KGs
 ┃ ┃ ┣ 📂ILLs_fusion
 ┃ ┃ ┃ ┣ 📜merged_ILLs_KG_en_fr.txt
 ┃ ┃ ┃ ┣ 📜merged_ILLs_KG_en_zh.txt
 ┃ ┃ ┃ ┗ 📜merged_ILLs_KG_zh_fr.txt
 ┃ ┃ ┗ 📂NMN_fusion
 ┃ ┃ ┃ ┣ 📜merged_NMN_KG_en_fr.txt
 ┃ ┃ ┃ ┣ 📜merged_NMN_KG_en_zh.txt
 ┃ ┃ ┃ ┗ 📜merged_NMN_KG_zh_fr.txt
 ┃ ┗ 📂sampled_monolingual_KGs
 ┃ ┃ ┣ 📜Sampled_en.txt
 ┃ ┃ ┣ 📜Sampled_fr.txt
 ┃ ┃ ┗ 📜Sampled_zh.txt
 ┣ 📂Questions
 ┃ ┣ 📂en-fr
 ┃ ┃ ┣ 📂2-hop
 ┃ ┃ ┃ ┣ 📂rdf_version
 ┃ ┃ ┃ ┃ ┣ 📜en_fr_2h_en_question_rdf
 ┃ ┃ ┃ ┃ ┣ 📜en_fr_2h_fr_question_rdf
 ┃ ┃ ┃ ┃ ┗ 📜en_fr_2h_zh_question_rdf
 ┃ ┃ ┃ ┣ 📜en_fr_2h_en_question.txt
 ┃ ┃ ┃ ┣ 📜en_fr_2h_fr_question.txt
 ┃ ┃ ┃ ┗ 📜en_fr_2h_zh_question.txt
 ┃ ┃ ┗ 📂3-hop
 ┃ ┃ ┃ ┣ 📂rdf_version
 ┃ ┃ ┃ ┃ ┣ 📜en_fr_3h_en_question_rdf
 ┃ ┃ ┃ ┃ ┣ 📜en_fr_3h_fr_question_rdf
 ┃ ┃ ┃ ┃ ┗ 📜en_fr_3h_zh_question_rdf
 ┃ ┃ ┃ ┣ 📜en_fr_3h_en_question.txt
 ┃ ┃ ┃ ┣ 📜en_fr_3h_fr_question.txt
 ┃ ┃ ┃ ┗ 📜en_fr_3h_zh_question.txt
 ┃ ┣ 📂en-zh
 ┃ ┃ ┣ 📂2-hop
 ┃ ┃ ┃ ┣ 📂rdf_version
 ┃ ┃ ┃ ┃ ┣ 📜en_zh_2h_en_question_rdf
 ┃ ┃ ┃ ┃ ┣ 📜en_zh_2h_fr_question_rdf
 ┃ ┃ ┃ ┃ ┗ 📜en_zh_2h_zh_question_rdf
 ┃ ┃ ┃ ┣ 📜en_zh_2h_en_question.txt
 ┃ ┃ ┃ ┣ 📜en_zh_2h_fr_question.txt
 ┃ ┃ ┃ ┗ 📜en_zh_2h_zh_question.txt
 ┃ ┃ ┗ 📂3-hop
 ┃ ┃ ┃ ┣ 📂rdf_version
 ┃ ┃ ┃ ┃ ┣ 📜en_zh_3h_en_question_rdf
 ┃ ┃ ┃ ┃ ┣ 📜en_zh_3h_fr_question_rdf
 ┃ ┃ ┃ ┃ ┗ 📜en_zh_3h_zh_question_rdf
 ┃ ┃ ┃ ┣ 📜en_zh_3h_en_question.txt
 ┃ ┃ ┃ ┣ 📜en_zh_3h_fr_question.txt
 ┃ ┃ ┃ ┗ 📜en_zh_3h_zh_question.txt
 ┃ ┗ 📂zh-fr
 ┃ ┃ ┣ 📂2-hop
 ┃ ┃ ┃ ┣ 📂rdf_version
 ┃ ┃ ┃ ┃ ┣ 📜zh_fr_2h_en_question_rdf
 ┃ ┃ ┃ ┃ ┣ 📜zh_fr_2h_fr_question_rdf
 ┃ ┃ ┃ ┃ ┗ 📜zh_fr_2h_zh_question_rdf
 ┃ ┃ ┃ ┣ 📜zh_fr_2h_en_question.txt
 ┃ ┃ ┃ ┣ 📜zh_fr_2h_fr_question.txt
 ┃ ┃ ┃ ┗ 📜zh_fr_2h_zh_question.txt
 ┃ ┃ ┗ 📂3-hop
 ┃ ┃ ┃ ┣ 📂rdf_version
 ┃ ┃ ┃ ┃ ┣ 📜zh_fr_3h_en_question_rdf
 ┃ ┃ ┃ ┃ ┣ 📜zh_fr_3h_fr_question_rdf
 ┃ ┃ ┃ ┃ ┗ 📜zh_fr_3h_zh_question_rdf
 ┃ ┃ ┃ ┣ 📜zh_fr_3h_en_question.txt
 ┃ ┃ ┃ ┣ 📜zh_fr_3h_fr_question.txt
 ┃ ┃ ┃ ┗ 📜zh_fr_3h_zh_question.txt
 ┣ 📂Templates
 ┃ ┣ 📂en
 ┃ ┃ ┣ 📜en_pattern_body_en.txt
 ┃ ┃ ┣ 📜en_pattern_body_fr.txt
 ┃ ┃ ┣ 📜en_pattern_body_zh.txt
 ┃ ┃ ┣ 📜en_pattern_en.txt
 ┃ ┃ ┣ 📜en_pattern_fr.txt
 ┃ ┃ ┗ 📜en_pattern_zh.txt
 ┃ ┣ 📂fr
 ┃ ┃ ┣ 📜fr_pattern_body_en.txt
 ┃ ┃ ┣ 📜fr_pattern_body_fr.txt
 ┃ ┃ ┣ 📜fr_pattern_body_zh.txt
 ┃ ┃ ┣ 📜fr_pattern_en.txt
 ┃ ┃ ┣ 📜fr_pattern_fr.txt
 ┃ ┃ ┗ 📜fr_pattern_zh.txt
 ┃ ┗ 📂zh
 ┃ ┃ ┣ 📜zh_pattern_body_en.txt
 ┃ ┃ ┣ 📜zh_pattern_body_fr.txt
 ┃ ┃ ┣ 📜zh_pattern_body_zh.txt
 ┃ ┃ ┣ 📜zh_pattern_en.txt
 ┃ ┃ ┣ 📜zh_pattern_fr.txt
 ┃ ┃ ┗ 📜zh_pattern_zh.txt
 ┗ 📜README.md

Questions

This directory contains all the parallel questions corresponding to the three bilingual KGs of "en-fr","en-zh" and "zh-fr".

• All datasets are categorized by the type of questions into different sub-directories.
• We name each dataset (text file) by its attributes. For example, en_fr_2h_en_question indicates the following attributes for this dataset:
  • en_fr means the triples used to construct this dataset are from the English and French versions of DBPedia.
  • 2h denotes that this question requires 2-hop reasoning.
  • en_question means all the questions are in English.
• We provide each dataset in two formats, one is the standard RDF-format (in rdf_version directories), the other is a CSV-like format (.txt files) with the following conventions:
  • Each line is a data record.
  • Each data record can be separated by a first-level separator (a TAB character " ") into the text of the question, the answer and the reasoning path.
  • The reasoning path can be separated by a second-level separator ("#") into triples.

Templates

This directory contains parallel templates (patterns) for relations in English, Chinese and French, in case you find them useful.

KGs

This directory contains all the KGs used in question generation.

• sampled_monolingual_KGs contains the sampled KGs using the ICS approach.
• fusion_bilingual_KGs contains bilingual KGs that are created through entity alignments, either with IILs (ground truth) or NMN model predictions.