| Resumen |
The growing trend of developing Large Language Models (LLMs) has gained popularity due to their ability to process and generate natural language, which has applications in various industries. However, the use of LLMs to build domain-specific knowledge graphs remains challenging due to the reliance on human experts to define entities and relationships and to address problems such as information granularity, lack of timeliness, etc. This work presents a workflow that integrates three different LLMs (Llama 3, GPT-4o, and Claude 3 Sonnet) to perform a semiautomated construction of knowledge graphs from news related to natural disasters in Mexico. This ongoing work provides a preliminary assessment of the ability of LLMs to represent specific knowledge domains, with the potential to improve accessibility and retrieval of relevant data, thus facilitating future identification of associations and patterns related to natural disasters. The experiments carried out have shown that our workflow enriches the identification and relationships of entities from the news corpus. However, our current evaluation shows that the LLMs used are far from replacing human intervention. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025. |
