Thesis TÉCNICAS DE EXPANSIÓN DE VECINDARIOS CON APLICACIÓN A FILTRADO COLABORATIVO
Loading...
Date
2016
Journal Title
Journal ISSN
Volume Title
Program
Campus
Campus San Joaquín, Santiago
Abstract
Los Sistemas de Recomendación ayudan a las personas a tomar decisiones ante la gran cantidad de información existente. Filtrado Colaborativo es una de las técnicas más populares,el cual utiliza la información de usuarios con gustos similares para realizar recomendaciones.Se divide en dos categorías: métodos basados en memoria y basados en modelos. Los métodos basados en memoria gozan de gran popularidad debido a su simplicidad y buen rendimiento.Sin embargo, su desempeño decae en situación de escasez de datos (cold-start).Este documento tiene por objetivo mejorar el desempeño ante situación de escasez de información mediante expansión de vecindarios. Para esto se estudian las técnicas más representativas de filtrado colaborativo. Luego, se realiza un estudio exhaustivo y comparativo de los métodos propuestos, mediante experimentos en el área de recomendación de productos. Los resultados muestran que expansión de segundo orden amplía el espacio de búsqueda mejorandoel problema cobertura limitada y escasez de datos que adolecen los métodos basados en memoria, proporcionando mayor diversidad en las recomendaciones. Finalmente, se concluye en base a los resultados obtenidos y se mencionan posibles formas de extender este trabajo.
Recommender Systems aid people with decision making when large amounts of information are involved. Collaborative Filtering is one of the most popular techniques, it uses information from users with similar taste to make recommendations. It’s divided in two categories:memory-based and model-based. Memory-based models are highly popular because of their simplicity and performance. However, their efficacy drops when there’s lack of (coldstart) data. This document’s objective is to improve the performance for this situation using neighborhood expansions. To do this, the most representative collaborative filtering techniques are studied. Then, an exhaustive comparative study of the proposed methods is done with experiments in the area of product recommendation. The results of which show that the second order expansion widens the search space improving the limited coverage problem andlack of data that disrupt the memory based methods, providing more diversity in the recommendations.Finally, conclusions are drawn from the results found, and possible extensions
Recommender Systems aid people with decision making when large amounts of information are involved. Collaborative Filtering is one of the most popular techniques, it uses information from users with similar taste to make recommendations. It’s divided in two categories:memory-based and model-based. Memory-based models are highly popular because of their simplicity and performance. However, their efficacy drops when there’s lack of (coldstart) data. This document’s objective is to improve the performance for this situation using neighborhood expansions. To do this, the most representative collaborative filtering techniques are studied. Then, an exhaustive comparative study of the proposed methods is done with experiments in the area of product recommendation. The results of which show that the second order expansion widens the search space improving the limited coverage problem andlack of data that disrupt the memory based methods, providing more diversity in the recommendations.Finally, conclusions are drawn from the results found, and possible extensions
Description
Catalogado desde la version PDF de la tesis.
Keywords
EXPANSION DE VECINDARIOS, FILTRADO COLABORATIVO, SISTEMAS DE RECOMENDACION