Abstract: Recommendation system algorithms, having recently garnered significant attention in the field of digital Earth science, are expected to be widely applied in metallogenic prediction. Traditional metallogenic prediction studies fail to fully mine the various types of semantic information in massive geoscience data. The deep interest evolution network (DIEN), as a recommendation system algorithm, can fully mine semantic information to predict user preferences. Therefore, this study employed the DIEN model as the prediction model and the semantic information extracted from bedrock interpretation as the ore-controlling elements according to the database provided by the Western Australian government. The model was trained to perform metallogenic prediction for the study area. The prediction results indicate that 92.95% of uranium ore occurrences fell within the medium-high probability zone in the prediction map, with some unknown zones also showing high prediction probabilities. After removing known uranium ore occurrences in some zones, the retrained model still yielded medium-high prediction probabilities in these zones. The results suggest that the DIEN can effectively mine semantic information in metallogenic prediction studies, and the DIEN model exhibits strong predictive capacity for the study area, providing a novel approach for metallogenic prediction studies.