LISE: a Logic-based Interactive Similarity Explainer
Author(s)
Date Issued
2025
Type
conferenceObject
Volume
4001
ISSN
1613-0073
Journal
Abstract
This work presents LISE (Logic-based Interactive Similarity Explainer), a system for explaining the similarity of clusters of RDF resources, by identifying common characteristics in their RDF descriptions. LISE follows a pipeline that consists of four main modules: Machine Learning Module, which creates a representation of RDF resources as vector embeddings and clusters them; Logic-Based Module, which, for each cluster, computes a Knowledge Graph (with blank nodes) modeling the common characteristics of resources in the cluster; Natural Language Generation
Module, which translates the computed Knowledge Graphs into human-readable descriptions; and User Interaction and Feedback Loop, which collects user feedback about the relevance of generated explanations. LISE operates in a closed loop, leveraging user feedback to refine embeddings and subsequently improve clustering. It was tested on an RDF dataset containing structured drug-related information, demonstrating promising results in terms of
explainability and interpretability of clustering results.
Conference(s)
8th Workshop on Semantic Web Solutions for Large-Scale Biomedical Data Analytics, SeWeBMeDa 2025