Knowledge Graphs (KGs) are essential structures for capturing relations among entities in various artificial intelligence applications. However, learning and reasoning on KGs is often impeded by their sparsity.
This thesis investigates the role of negative sampling, a technique that incorporates false examples, in enhancing learning and reasoning with KGs. The goal is to develop a negative sampling technique, which incorporates entity types, relation types, and graph topology to generate more meaningful negative samples. This work emphasizes the importance of advanced negative sampling in improving the performance of KG-based systems and offers practical implications for real-world applications.