data_process

Ontological Schema Construction

Original Data Preparation and Introduction

1. download the ontology file of NELL from here and put it into the folder ori_data;

2. relations in datasets and their triples, including splits for training, validation and testing

   • NELL: NELL.train_tasks.json (training relations), NELL.dev_tasks.json (validation relations) and NELL.test_tasks.json (testing relations);
   • Wiki: relation file and triple file, Wiki.train_tasks.json (training relations), Wiki.dev_tasks.json (validation relations) and Wiki.test_tasks.json (testing relations), and Wiki.relation2ids_1 with all relations in Wikidata-ZS

Data Processing

We run the following scripts to construct ontological schemas for NELL-ZS and Wikidata-ZS:

For NELL-ZS

1. extract_RDFS_literals.py: extract RDFS axioms and literal information of relations and concepts, output rdfs_triples.txt and literals.txt

2. extract_OWL_axioms.py: extract 6 kinds of relation characteristics, output owl1.txt

3. extract_OWL_composition.py: extract relation compositions, output owl2_composition.txt

For Wikidata-ZS

1. For efficiency, we first run lookup_rels_info.py to lookup the relation information in Wikidata. Each relation has a queried json file named by its name and saved in ori_data/WikidataRelsInfo/

2. Then, we run parse_rels_info.py to parse the json file to extract RDFS axioms, including relation hierarchy, relation domain and range, and output rdfs_triples_sp_domain_range.txt

3. The subclass triples can be looked up by running lookup_sc_triples.py and output rdfs_triples_sc.txt

4. Next, we look up the textual meta data (including label names and descriptions) for relations and concepts, run lookup_literals.py and output literals.txt

5. Run parse_rels_info_owl.py for extract the relation characteristics semantics and inverse triples, output owl1.txt

6. Mine rules from relations' facts to extract relation compositions, the steps are as follows:

   • We first gather the facts of relations in Wikidata-ZS by running the function prepare_data() in extract_OWL_composition.py and save facts in a CSV file wiki_triples.csv
   • Then, we mine rules from these facts using AMIE, a detailed illustration for using AMIE is here
   • For the mined rules, we save them in a CSV file, we have provided the rules mined by ours in output_data/Wiki/mined_rules.csv
   • Finally, we process these rules with extract_OWL_composition.py and output the composition axioms owl2_composition.txt.

7. Run output2CSV.py to save the ontological schemas in the form of RDF triples. Note we set different parameters to output schema graphs with different semantic settings. Taking NELL-ZS as an example:

   • generate schema graph with all semantics by running python output2CSV.py --dataset NELL --all
   • generate schema graph with semantics in RDFS by running python output2CSV.py --dataset NELL --rdfs
   • generate schema graph with semantics in RDFS and text by running python output2CSV.py --dataset NELL --rdfs --literal

8. Run output2RDF.py to save KGs with Turtle and XML files.