Explainable Temporal Fact Validation Through Constraints Discovery in Knowledge Graphs

Thibaut Soulard, Fatiha Saïs, Joe Raad

Abstract: Truth discovery aims to identify the most reliable or truthful information from multiple sources, particularly in scenarios where sources may be contradictory or unreliable. Addressing this challenge is critical for applications like data exploration, integration, and veracity assessment. Knowledge Graphs (KGs) provide a valuable framework for verifying facts against explicit or implicit domain knowledge. However, the temporal validation of facts that consist in determining their truthfulness within specific time intervals, remains an open area of research. In this paper, we propose an explainable approach that exploits temporal constraints discovered within KGs to classify facts based on their temporal validity. Our method involves identifying both simple and complex temporal constraints in KGs that capture the temporal consistency of facts within an entity’s timeline. These constraints are expressed through an extension to time sequences of Allen’s temporal algebra. Using these constraints, we validate facts through both symbolic and machine learning-based approaches, comparing their performance under various hyperparameter settings. Experiments conducted on Wikidata, one of the largest publicly accessible KGs, demonstrate the effectiveness and high accuracy of our temporal constraint-based approach.

External IDs:dblp:conf/esws/SoulardSR25