Challenges in Working with Unstructured Data in the LLM Era: NER Processes Using Graph Neural Networks

The application of neural networks and large language models (LLMs) in the financial technology (fintech) sector has significantly enhanced capabilities of providing better services based on unstructured documents, mainly thanks to the astonishing interactive capabilities of large language models (LLMs). However, in these contexts, precisely recognizing named entities (for example, the beneficiary of an insurance policy) and creating reliable, domain-dependent, private knowledge graphs is more crucial than conversational capabilities. Moreover, optimizing these models for dynamic data environments and ensuring Explainability presents notable challenges. This paper aims to present our approach for the creation of a knowledge graph that uses graph neural nets (GNN) to provide a better named-entity recognition (NER) process, to identify entities in the text precisely and to solve other issues in knowledge graph management, like the co-ference resolution problem or focusing on dynamic data environments and the critical need for Explainability of modern AI-based solutions. The paper presents our preliminary results on one of the various pipeline steps that creates the navigable knowledge graph.