This study addresses the challenge of accurately representing temporal information in Knowledge Bases (KBs), crucial for enhancing the reliability of time-sensitive data in resources like Wikidata. The primary focus is on temporal scoping within KBs, emphasizing the integration of time intervals into factual data. Using Wikidata as the main KB, facts are represented as triples (RDF triples) consisting of a subject, predicate, and object. In the initial experimental phase, we implemented a neural network model to perform temporal reasoning on sentences from Wikipedia where both triple components were entities. The model’s performance was evaluated based on its ability to predict the temporal scope of facts, comparing results to Wikidata’s start and end intervals, the gold standard. In the subsequent phase, we adapted the model to handle cases with one entity and one numeric object, focusing on three specific relations. This required expanding data collection to include Google searches for numeric values alongside Wikipedia. To improve semantic understanding, we replaced GloVe embeddings with S-BERT encodings and shifted from a probabilistic to a classification model. An attention-based sliding window mechanism further enhanced temporal scope prediction across relations, broadening the model’s applicability and improving temporal accuracy in KBs.