Thesis Aplicación de técnicas de procesamiento de lenguaje natural para la creación de un sistema de recomendación de productos
Loading...
Date
2024-03
Authors
Journal Title
Journal ISSN
Volume Title
Program
DEPARTAMENTO DE ELECTRÓNICA. INGENIERÍA CIVIL TELEMÁTICA
Campus
Campus Casa Central Valparaíso
Abstract
En esta era de la información digital, las empresas que proveen servicios de búsqueda de información consideran un proceso fundamental la retención de los usuarios que utilizan sus sistemas. Desde empresas de retail hasta redes sociales, buscan que los usuarios se mantengan el mayor tiempo posible dentro de sus plataformas, y para ello buscan personalizar la experiencia de usuario sobre sus sistemas. Uno de los enfoques que se utiliza comúnmente es la entrega de recomendaciones asociadas al comportamiento del usuario dentro de los sistemas utilizados.
Utilizando una combinación de técnicas del procesamiento del lenguaje natural, como la extracción de palabras clave, o la generación de embeddings, se propone un sistema de recomendación robusto, capaz de entregar recomendaciones relevantes mediante el cálculo de la distancia coseno entre los embeddings asociados al usuario y a los productos obtenidos desde modelos derivados de BERT mediante K-NN, para finalmente evaluar la inclusión de TD-IDF para el reordenamiento y mejora de las recomendaciones entregadas.
Dentro de las pruebas realizadas, el sistema mostró altos valores para métricas de ranking, como un 76% para nDCG y un 79% para MRR. Estos resultados sugieren que la aplicación de técnicas de lenguaje natural y aprendizaje automático pueden ser efectivas para personalizar la experiencia de usuario en una variedad de contextos, beneficiando tanto a la empresa como a los usuarios finales.
In this era of digital information, companies that provide information search services consider user retention to be a fundamental process. From retail companies to social networks, they aim to keep users engaged as long as possible within their platforms. To achieve this, they seek to personalize the user experience on their systems. One of the approaches commonly used is to provide recommendations based on user behavior within the systems they use. Using a combination of natural language processing techniques, such as keyword extraction and the generation of embeddings, a robust recommendation system is proposed. The system is capable of providing relevant recommendations by calculating the cosine distance between the user-related embeddings and the product-related embeddings derived from BERT models via KNN. The implementation of TF-IDF is also evaluated to reorder and improve the recommendations given. In the tests conducted, the system showed high values for ranking metrics, such as 76% for nDCG and 79% for MRR. These results suggest that the application of natural language processing and machine learning techniques can be effective in personalizing the user experience in various contexts, benefiting both the company and end-users.
In this era of digital information, companies that provide information search services consider user retention to be a fundamental process. From retail companies to social networks, they aim to keep users engaged as long as possible within their platforms. To achieve this, they seek to personalize the user experience on their systems. One of the approaches commonly used is to provide recommendations based on user behavior within the systems they use. Using a combination of natural language processing techniques, such as keyword extraction and the generation of embeddings, a robust recommendation system is proposed. The system is capable of providing relevant recommendations by calculating the cosine distance between the user-related embeddings and the product-related embeddings derived from BERT models via KNN. The implementation of TF-IDF is also evaluated to reorder and improve the recommendations given. In the tests conducted, the system showed high values for ranking metrics, such as 76% for nDCG and 79% for MRR. These results suggest that the application of natural language processing and machine learning techniques can be effective in personalizing the user experience in various contexts, benefiting both the company and end-users.
Description
Keywords
Sistema de recomendación, Métricas de Ranking, Microservicios, BERT, Extracción de palabras clave