In this thesis, we introduce and implement new approaches for the Relation Linking task in a questions-answering context using the Wikidata knowledge graph. We extracted different information from the text question, such as entities and their representation, and we used Sentence- BERT to encode questions. Two categories of approaches for relation linking were developed, baseline models and advanced models. For the baseline models, different approaches are used based on similarity, regression, etc. On the other hand, based on models that have been proven to work well for passage retrieval in Open Domain Question Answering, we developed neural-based methods for relation linking. We use the F1 score to compare the performance of our models among each other as well as to Falcon, a benchmark approach for the Relation Linking task.