Sistemas de recomendación basados en filtrado colaborativoaceleración mediante computación reconfigurable y aplicaciones predictivas sensoriales

Abstract

Recommender systems are widely used in product recommendation on different platforms. In this area of knowledge, the PhD thesis focuses on two complementary lines of research: design of hardware accelerators for computing and behavior-based prediction. The algorithms that work on recommendations are sophisticated and can require very high computational efforts when working in environments with many users and data. This circumstance motivated research on the acceleration of algorithm computation to obtain recommendation results in reasonable times, by means of FPGA technology, using highlevel synthesis languages as a modeling tool and parallelization strategies. Based on the accelerated algorithms, an innovative application of recommender systems was proposed for a prediction problem in a sensor infrastructure. In this case, recommender systems are proposed as a prediction tool to determine the values of environmental parameters as a function of the human activity performed on spaces, which are monitored by wireless sensors. In addition, a third study was addressed as a consequence of the previous approach, related to the optimal selection of the test data needed to evaluate the prediction algorithms.