Knowgly

Entity Indexing and Retrieval in Knowledge Graphs, via an information-based and fully unsupervised index schema construction

García, S., Bobed, C. (2025). Information-Aware Entity Indexing in Knowledge Graphs to Enable Semantic Search. In: Curry, E., et al. The Semantic Web. ESWC 2025. Lecture Notes in Computer Science, vol 15718. Springer, Cham.

https://doi.org/10.1007/978-3-031-94575-5_12

1. Requirements
2. Compiling
3. Usage
4. Supplemental materials
5. Licensing

The system consists of a Java project implementing the whole pipeline presented in the paper. There are also several helper scripts written in both Python and bash for evaluation and handling external systems, such as Terrier or Galago. The requirements are the following:

• Maven

• openJDK >=17. pom.xml assumes openJDK 20, but can be easily changed in the maven.compiler.source and maven.compiler.target settings.

• If using galago (recommended):

  • A valid installation of galago under ../galago/galago-3.16/ and openJDK 8. It is also possible to use newer versions. See how_to_setup_galago.txt on ../galago/ for more information.

• If using pyTerrier:

  • A valid installation of pyTerrier under ../terrier/. See how_to_setup_terrier.txt on ../terrier/ for more information.

• If using elastic:

  • An accesible local or remote elastic instance. We have tested our system under version 8.6.2. See configuration/examples/elasticEndpointConfiguration.json for more details.

• If using Lucene:

  • Nothing. The Lucene libraries are already included in Knowgly, and it will use them to create a local index.

• If performing evaluations

  • A compilled executable of https://github.com/usnistgov/trec_eval, and a python environment with numpy and scikit-learn installed. See Knowgly/evaluation/metrics_testing/README.txt for more details.

To compile the system, simply run compile.sh. It will compile and move to this folder all required .jar files, with all dependencies statically linked.

If you want to:

• Run Knowgly's metrics generation and indexing pipelines:

  • Freely edit the demo shown in the Main.java file, and run execute.sh (or use Knowgly as a library).
  • A simple demo on how to run each pipeline is already provided in the file.

• Perform individual queries:

  • Please refer to the demo shown in the Main.java file

• Perform a full evaluation (multiple queries and .run file generation):

  • Execute the RunEvaluator.jar file, which has been prepared as a CLI tool for any system. Our evaluation scripts employ this executable too.

• Perform Coordinate Ascent (Note: We allow all systems, but it has only been tested on galago):

  • Run the ca*.py scripts in the CA folder. There are currently scripts for 3 and 5 fields.

• Evaluate .run files:

  • See Knowgly/evaluation/metrics_testing/README.txt for more details.

• Build the datasets we used for evaluation (DBpedia and IMDb):

  • See Knowgly/datasets/README.txt for more details.

Aside from Knowgly's implementation, we also provide additional documentation, as mentioned throughout the paper:

• dataset_analysis.md: An overview of the dataset details and Predicate-Type distributions.
• implementation_details.md: An analysis of Importance Metrics Calculation times for both datasets (DBpedia and IMDb) and additional clustering (KMeans) details.
• fields_and_weight_scheme_analysis.md: An analysis of the effect of different numbers of fields and an overview of alternative weight scheme strategies, such as directly assigning normalized centroid values.

Additionally, the paper's figures are also available. A small subset of VDoc templates are also available in example_vdocs, and the best results are available as .run files in best_runs.

This software is licensed under the GNU Affero General Public License v3.0