Knowledge Graphs (KGs) are essential in applications like semantic search, question answering, and decision support. They structure knowledge, validate facts, enable inference, and increasingly enhance Large Language Models (LLMs) by grounding outputs in structured, factual data. However, KGs have treated facts as static and timeless, ignoring the temporal nature of many truths. This leads to outdated or incorrect inferences. Temporal Knowledge Graphs (TKGs), like Wikidata and YAGO, address this by modeling the time-bound validity of facts. Multiple recent work has focused on predicting missing temporal facts, yet validating existing temporal information, to ensure the reliability and accuracy of TKGs, remains underexplored. In order to advance this area of research, we introduce the first benchmark designed to evaluate temporal fact validation methods. Derived from Wikidata, this benchmark supports systematic, quantitative, and qualitative comparisons, incorporating diverse assumptions about temporal data (e.g., timestamps, intervals) and KG structures (e.g., density, depth).

In recent years, keen interest towards Knowledge Graphs has increased in both academia and the industry which has led to the creation of various datasets and the development of different research topics. In this paper, we present an approach that discovers differential causal rules in Knowledge Graphs. Such rules express that for two different class instances, a different treatment leads to different outcomes. Discovering causal rules is often the key of experiments, independently of their domain. The proposed approach is based on semantic matching relying on community detection and strata that can be defined as complex sub-classes. An experimental evaluation on two datasets shows that such mined rules can help gain insights into various domains.

International audience

One intrinsic characteristic of knowledge bases (KB), especially those published on the Web of data, is the frequent evolution of their data. Hence, changes that occur may lead to KB inconsistency and therefore, may generate contradictions between the KB facts and the KB axioms. In this paper, we propose an approach that is able to, first, compute and semantically represent the symmetric difference (diff) between two different versions of a KB and, second, use the generated diff to detect changes (addition and deletion) for the corresponding KB axioms. We further propose an experimental assessment of the approach on exisitng knowledge bases such as DBpedia.