Following the principles of Linked Data (LD), data providers are producing thousands of interlinked datasets in multiple domains including life science, government, social networking, media and publications. Federated query engines allow data consumers to query several datasets through a federation of SPARQL endpoints. However, data providers just receive subqueries resulting from the decomposition of the original federated query. Consequently, they do not know how their data are crossed with other datasets of the federation. In this paper, we propose FETA, a Federated quEry TrAcking system for LD. We consider that data providers collaborate by sharing their query logs. Then, from a fed-erated log, FETA infers Basic Graph Patterns (BGPs) containing joined triple patterns, executed among endpoints. We experimented FETA with logs produced by FedBench queries executed with Anapsid and FedX federated query engines. Experiments show that FETA is able to infer BGPs of joined triple patterns with a good precision and recall.