https://repositorio.uide.edu.ec/handle/37000/7601 2026-11-03T08:14:57Z https://repositorio.uide.edu.ec/handle/37000/10117 Title: Ajuste fino de un modelo de embeddings ligero pa ra la recuperación semántica de jerga coloquial en co mercio electrónico mediante bases de datos vectoriales Authors: Moromenacho Aguirre, Juan David; Vizcaino Imacaña, Fernanda Paulina (tutor) Abstract: Este trabajo evalúa el fine-tuning de un modelo de embeddings ligero(MiniLM-L6-v2)paralare cuperación semántica de productos en comercio electrónico cuando las consultas se formulan en español coloquial. Los buscadores basados en palabras clave no logran asociar consultas informales como “estuche pal celu xiami” con descripciones técnicas como “Carcasa protec tora Xiaomi Redmi Note 12”. Todo el estudio se realizó con datos sintéticos y en condiciones experimentales controladas: se generó un conjunto base de 1656 pares ancla–positivo, par ticionado en entrenamiento, validación y prueba de forma disjunta y ampliado por sustitución de variantes coloquiales a 9684 pares de entrenamiento, y se ajustó el modelo con Multiple Negatives Ranking Loss durante 20 épocas, con selección del mejor punto de control por pér dida de validación...This work evaluates the fine-tuning of a lightweight embedding model (MiniLM-L6-v2) for se manticproduct retrieval in e-commerce whenqueriesare written in colloquialSpanish.Keyword based search fails to map informal queries such as “estuche pal celu xiami” to technical des criptions such as “Carcasa protectora Xiaomi Redmi Note 12”. The whole study was carried out on synthetic data under controlled experimental conditions: a base set of 1,656 anchor–positive pairs was generated, split into disjoint training, validation and test partitions, and expanded by colloquial-variant substitution to 9,684 training pairs; the model was fine-tuned with Multiple Negatives Ranking Loss over 20 epochs, selecting the best checkpoint by validation loss... 2026-01-01T00:00:00Z https://repositorio.uide.edu.ec/handle/37000/9247 Title: Diseño de un escape room en realidad virtual para preservar las tradiciones y cultura de la comunidad shuar Authors: Carrión Campoverde, Leonardo Andrés; Cardenas, Charlie (tutor) Abstract: This article presents the design and implementation of a escape room called “Arutam: The Call of the Spirit,” which is a virtual reality game developed with Unreal Engine 5 for implementation on the Meta Quest 3 platform. The main objective is to contribute to the preservation and dissemination of the Shuar people's worldview through an immersive experience based on symbolic narrative and interactive mechanics. The development followed an iterative approach, integrating narrative design, human-computer interaction, and programming using Blueprints and C++. Qualitative results suggest that virtual reality promotes immersion and cultural understanding, positioning the video game as a viable technological tool for the preservation of intangible cultural heritage. 2026-01-01T00:00:00Z https://repositorio.uide.edu.ec/handle/37000/9246 Title: Implementación de Modelos Predictivos para la Gestión del Proceso de Polinización y Germinación de Semillas en la Empresa Ecuagenera Authors: Portilla Cartuche, Arlette Stefanny; Palacios Morocho, Milton Ricardo (tutor) Abstract: The study develops and implements a predictive system based on machine learning techniques for managing orchid pollination and germination processes at the Ecuadorian company Ecuagenera. These processes exhibit high biological and seasonal variability, which has historically forced the company to rely on manual records and empirical experience, generating operational uncertainty and production losses. The research adopts a quantitative and experimental approach, using historical data collected in the laboratory and applying feature engineering techniques to incorporate temporal, biological, and operational variables. Different supervised learning algorithms were evaluated, including Random Forest, XGBoost, and LightGBM, selecting the models with the best predictive performance according to metrics such as MAE, RMSE, and coefficient of determination (R²). The results show that Random Forest offers a high level of accuracy in predicting germination time, while XGBoost performs better in predicting the times associated with the pollination and germination process. 2026-01-01T00:00:00Z https://repositorio.uide.edu.ec/handle/37000/8725 Title: Desarrollo e implementación de un Chatbot basado en LLM para brindar información a la comunidad universitaria de la Universidad Internacional del Ecuador, campus Loja Authors: Jiménez Orellana, Santiago Michael; Palacios Morocho, Milton Ricardo (tutor) 2025-01-01T00:00:00Z