Métodos de clasificación basados en asociación aplicados a sistemas de recomendación

[EN]Current e-commerce systems continually need to provide personalization when their content is shown. In this sense, recommender systems make suggestions and provide information of items available in the system. Nowadays, there is a vast amount of methods, including data mining techniques that can be employed for personalization in recommender systems. However, such methods are still quite vulnerable to some limitations and shortcomings related to recommender environment. Classification based on association methods, also named associative classification methods, consist of an alternative data mining technique, which combines concepts from classification and association in order to allow association rules to be employed in a prediction context. In this work we propose the use of associative classifiers in recommender systems in order to enhance the recommendation process and to shorten limitations presented within it. To do so, we firstly present a bibliographic revision concerning the associative classification area and, subsequently, we describe concepts and methods related to recommender systems. Within this context, in this work we have developed a hybrid recommender methodology, which encloses characteristics of the two main approaches of recommender methods: collaborative filtering and content-based. This methodology also includes an associative classification algorithm that inherits features from fuzzy logic, that allows it to enhance its effectiveness and recommendation quality. In this way, the methodology takes advantage from the strengths of both approaches and, therefore, minimizes recommender systems limitations. In order to analyze the behaviour and to validate the developed methodology, we have accomplished a comparative study using data gathered from real recommender systems. Such study analyzes results from associative classification and traditional classification algorithms, furthermore it analyzes the behaviour of the algorithm proposed as a part of the methodology. Finally, we describe the implementation of the proposed methodology in a real recommender system, where critical scenarios that usually occur in a recommendation context were emulated. The results of the comparative study and the analysis of the emulated scenarios have demonstrated that the associative classification and the proposed methodology can be successfully applied in recommender systems and are able to supply benefits to them as well

Author

• Pinho Lucas, Joel

• Tesis y disertaciones académicas
• Universidad de Salamanca (España)
• Academic dissertations
• Informática
• Inteligencia artificial
• Bases de datos

Identifier

hdl:10366/83342

OAI identifier oai:identifier

oai:gredos.usal.es:10366/83342