Traditional recommendation systems often face issues in bridging the semantic gap and seamlessly integrating data from heterogeneous knowledge bases, thus limiting their ability to make contextually relevant and highly personalized recommendations. Our approach addresses key challenges in integrating diverse data sources by resolving terminological inconsistencies and capturing complex relationships between user preferences and item attributes. Experimental results on the ontology-aligned datasets, MovieLens, DBpedia, and Wikidata, constructed by our proposed work demonstrate significant performance improvements comparing with the baseline: precision increases from 0.0230 to 0.1294 (462%), recall from 0.0197 to 0.1471 (646%), and F1-score from 0.0172 to 0.1079 (527%) at top-10 recommendation list.