Aplicación del método few shot learning al modelo GPT3.5 para la personalización del contenido de las redes sociales

This paper presents an approach based on KDD (Knowledge Discov-ery in Databases) for the development of a web application using the knowledge transfer technique “few shot learning” to the GPT3.5 model. The objective is to generate personalized and high-quality content for business social networks. Data is collected through a logging interface and Excel files containing existing copies and social network interaction metrics. Data preprocessing and transformation techniques are applied to improve the generalizability of the model. The evalua-tion is performed using metrics of similarity, grammaticality, and relevance of the generated content. The results show high scores for precision (0.7177) and recall (0.6890), indicating a substantial similarity between the generated copies and the existing ones. In addition, a grammatical score close to perfection is achieved. Regarding human validation, the results show that users are mostly satisfied with the generated content, which is validated through a survey after the use of the application. These results demonstrate the effectiveness of the proposal in the automated generation of personalized content for business social networks, which can save time and effort for marketers, while improving the quality and consistency of the generated content.