Financial markets are heavily influenced by textual data, with financial news playing a pivotal role in predicting market states. Although BERT has proven effective in general text analysis, it often fails to capture domain-specific semantics and the complex structural relationships among financial entities. To address these limitations, this study proposes an ontology-enriched BERT model that integrates financial ontologies, graph-based embeddings (Node2Vec), and topological features such as degree centrality, betweenness centrality, and clustering coefficient. This integration enhances the model’s ability to capture semantic and structural interdependencies, improving market state classification. Experimental results demonstrate that the proposed framework achieves 91.2% classification accuracy and reduces prediction loss to 0.308, outperforming baseline models. These findings underscore the importance of combining semantic enrichment and topological insights for financial forecasting, offering valuable contributions to risk management, portfolio optimization, and decision-making processes in both academic and industry contexts.