Metodología para la Generación de Explicaciones para Sistemas de Recomendación Sensibles al Contexto

A great deal of the research in the Recommender Systems area is centered in the study of recommendation techniques; these techniques depend on the context, the effectiveness of the technique as well as on the metrics in order to evaluate them. However, the study of the explanation styles in recommender systems has become relevant because they have shown to improve the user's experience.Using explanation styles in a recommender system helps the user to more rapidly understand the information that is provided by a RS and to decide if there is enough evidence to take a recommender as valid helping the user in the decision making process. In addition, explanation styles propose several objectives such as transparency, effectiveness, satisfaction, persuasion, efficiency, and trust among others.In this thesis it was designed a methodology to build explanatory information that depends on the recommender technique and on its objectives. This explanation methodology provides a guide for textual explanations using templates with variable fields. Such methodology was evaluated in a Context-Aware Recommender System (T-Guia, González 2012) in its first version which does not have a service to generate explanations.In order to evaluate the methodology developed in this thesis, real users answered questionnaires; the evaluation of the explication techniques was carried out by means of a Web prototype which allowed to carry out a user-centered evaluation UCE in order to measure the impact of the explications in the trust metrics. The results showed that the explications by means of explanatory templates in combination with images obtained better evaluations in comparison with explications presented by means of mind maps and concept maps. The results obtained are the following: Textual explanations with a comprehension of 84.71% and trust of 85.25%, explications by means of mind maps with a comprehension of 82.00 and trust of 80.42% and the explications by means of conceptual maps with a comprehension of 76.10% and trust of 75.50%.