Leopard - A baseline approach to attribute prediction and validation for knowledge graph population.

In this paper, we report on the participation of Leopard to the Semantic Web Challenge at the 16th International Semantic Web Conference. Leopard is a baseline approach to predict and validate attributes for knowledge graph population. The approach was designed as a baseline for the challenge. It combines diverse text extraction methods with a simple precision ranking and utilizes sources from the multilingual Document Web as well as from the multilingual Data Web. Despite being designed to be a baseline, Leopard achieved the second-best score in both challenge tasks (53.42% F1-Score and 53.09% AUC) behind IBM’s system Socrates (55.40% F1-Score and 68.01% AUC). Our approach is open source and can be found at https://github.com/dice-group/Leopard.