Uma abordagem de sistema de recomendação orientada pelo aprendizado sem fim
Gotardo, Reginaldo Aparecido
MetadataMostrar registro completo
Recommender Systems have a very well defined function: recommend something to someone. Through Artificial Intelligence techniques, more particularly from areas such as Data Mining and Machine Learning, it is possible to build recommendation systems. These systems will analyze large amounts of data and will inform users about some items that will probably interest them. However, some limitations of the recommender systems, which are sometimes, caused by the Mining or Learning models themselves or by the lack of available data make them computationally expensive or inaccurate. Besides, recommender systems in real environments are dynamic: data change over time or with new ratings, new users, new items or when user updates previous ratings. The Never Ending-Learning Approach (NEL) aims at a self-supervised and self-reflexive learning to mainly maximize learning of a system based on data from several sources, algorithms that can cooperate to make a better knowledge base considering the dynamic of real learning problems: learning improves along the time. As mentioned before, recommender systems are dynamic and depend on data between user and items. In order to minimize this dependency and to provide meaningful and useful results to users, this work presents a Recommender System approach guided by NEL Principles. Results show that it is possible to minimize or delay the data dependency through classifiers coupling techniques and concept deviation control. Due to that, it is possible to start with little data from a recommender system that will be dynamic and will receive new information. These new information will help even more in controlling the concept deviation and promoting the most useful recommendations. Then, this thesis presents how the Recommender System guided by NEL principles can contribute to the state of the art in recommender systems and implement a system with practical results through the Never-Ending Learning Approach.