This repository contains a mini-benchmark for the LIMES: a Java tool for Link Discovery. The repository includes configuration and data. The Makefile serves as an interface and offers steps required to generate results. However, some initial manual steps are required (setting up triple store, initializing triple store repositories). The code may require some parametrization and minor improvements in order to make the solution reusable.

GraphDB is the database system used for these experiments. TED Notices adhering to ePO model are used as the test data.

experiment code explained

eN-Y-Z where:

• N: experiment number
• Y: data size (xs / s / m / a)
• extra description

experiments

1. 01: heuristic-based
   • metric: ExactMatch(x.legal_name, y.legal_name)
   • properties: "epo:hasLegalName AS nolang->lowercase RENAME legal_name"
2. 02: ML-based
   • ML algorithm: Wombat simple
   • properties: unsupervised

Steps to run an experiment

Note that the config files were archived in ./experiments subdirectories together with generated artifacts. In order to re-run an experiment, you need to copy a particular config file to ./config directory.

1. setup your GraphDB database, create repositories and import data (from ./data directory)
2. download limes: make get_limes
3. adjust the config file (config/config.json). Enter an URL to your SPARQL endpoint.
4. generate XML config file: make gen_config
5. run link discovery process (LIMES): make run_ld
6. copy experiment results to a dedicated subdirectory under experiments: make archive_experimentDIR_NAME=e01-xs
7. inspect results in the new experiments subdirectory
8. clean LIMES cache before running another experiment: make clean_cache

Data source

Files stored in ./data directory were produced by using SPARQL queries based on these stored in ./queries. Source of the data: https://fuseki.staging.meaningfy.ws/#/dataset/mdr_dataset/query The triple store contains TED-SWS MDR data. The MDR annotations has been filtered out.