A Neuro-Symbolic Approach For Sentiment Analysis In Financial News: Integrating Ontologies Into The Natural Language Processing Pipeline

Neuro-symbolism combines structured semantic knowledge with data-driven learning to mitigate the limitations of purely neural or purely symbolic models. Many real-world domains involve complex knowledge that cannot be captured by a single form of reasoning. This work proposes a hybrid neuro-symbolic approach for financial sentiment analysis, integrating a UFO-based financial event ontology into an NLP and deep learning pipeline. The ontology is engineered with the Methontology framework and formalized in OWL, enabling explicit modeling of financial events, their causes, and consequences. Semantic embeddings are constructed by combining TF-IDF and Word2Vec representations with OWL2Vec-based ontological features, yielding document vectors that jointly encode lexical and symbolic information. These embeddings are used to train SVM, Random Forest, CNN, and LSTM classifiers on short financial headlines. Ontology-enhanced models consistently outperform baselines, with the TF-IDF + SVM configuration reaching 75% accuracy and substantial F1-score gains. The results indicate that semantic enrichment improves predictive performance, robustness, and interpretability, highlighting the potential of neuro-symbolic methods for financial engineering applications.