Análisis De Sentimiento y Minería De Opiniones En Entornos Digitales Mediante Deep Learning: evolución desde modelos secuenciales hasta modelos de lenguaje de gran escala (LLMs)

Abstract

Sentiment analysis in digital environments has undergone a rapid methodological evolution, transitioning from classic sequential models to deep learning architectures capable of understanding profound human linguistic context. To examine this progression, a systematic review was conducted applying the PRISMA protocol, evaluating high-impact empirical studies (2020-2025) extracted from Scopus, IEEE Xplore, Web of Science, etc. The findings validate the efficacy of these technologies in real-world scenarios, highlighting their use in e-commerce satisfaction monitoring and public health trend assessment. However, the review reveals a priority research gap: the urgency of integrating algorithmic explainability (XAI) from the system's conception and democratizing these tools for low-resource languages. It is concluded that the current state of the art lies in combining Transformers and Large Language Models (LLMs) with parameter-efficient adaptation techniques (such as LoRA), offering organizations an optimal balance between maximum predictive accuracy and computational viability.