Integrating Ontological Reasoning with Statistical and Deep Models for User Similarity Evaluation in Social Media

Understanding user similarity on social and academic networks is a cornerstone for personalized recommendation , community detection, and knowledge discovery. However, most existing approaches rely solely on statistical or deep neural representations that capture surface correlations but overlook structured semantics. To address this gap, we propose an ontology-enhanced hybrid framework that integrates ontological reasoning with statistical and deep contextual embeddings to achieve interpretable and semantically consistent user similarity evaluation. Our framework unifies classical methods (TF–IDF, LSA/LDA), probabilistic topic modeling (ETM), and deep transformer embeddings (BERT + PCA) under an ontology-based reasoning layer that encodes hierarchical, associative, and multi-hop relationships among entities such as users, posts, topics, and devices. Ontology reasoning is employed to infer latent relations, reduce semantic noise, and provide explicit interpretability through property chains and SWRL rules. Comprehensive experiments on five heterogeneous datasets—Twitter, Facebook Social Circles, DBLP, AMiner, and Mendeley—demonstrate that the proposed Ontology–Hybrid model consistently outperforms baseline methods across all major metrics (Accuracy, F1, NDCG@10, MIH@10, and RWPH@10), achieving up to 8% F1 improvement and higher semantic diversity in top-k results. Despite moderate computational overhead, runtime–accuracy analysis shows that ontology-guided reasoning achieves the most efficient accuracy gain per unit time among semantic models.This research contributes a reproducible, extensible methodology that bridges symbolic reasoning and data-driven learning. By embedding domain ontolo-gies into hybrid embedding pipelines, the study offers both quantitative improvements and explainable insights-advancing the frontier of interpretable AI for user modeling, social-media analytics, and scholarly network mining.