Constructing text-centric knowledge graphs based on deep learning: a focus on named entity recognition and relation extraction

Knowledge graphs (KGs) are structured representations of knowledge widely used in natural language processing (NLP) for tasks such as information retrieval, recommendation systems, and data analysis. This study presents a framework for constructing knowledge graphs from textual data by focusing on named entity recognition (NER) and relation extraction (RE) using pre-trained language models. The proposed framework leverages BERT for deep contextualized embeddings, addressing challenges such as linguistic ambiguity, implicit relationships, and complex syntactic structures in text. To enhance performance, the framework integrates supervised learning with rule-based methods and employs techniques such as data augmentation and hyperparameter tuning. Experimental evaluation on diverse text corpora demonstrates that the BERT-based approach significantly outperforms traditional rule-based systems and BiLSTM-CRF models, achieving higher precision, recall, and F1-scores in both NER and RE tasks. The extracted entities and relationships are organized into a Neo4j-based graph database, enabling intuitive visualization and network analysis. This research not only advances NLP methodologies for knowledge graph construction but also highlights the potential applications in domains such as smart cities and smart education, where structured knowledge representations can facilitate decision-making, information management, and data-driven insights. The proposed approach provides a scalable and robust solution for extracting actionable knowledge from unstructured textual data.