Advancing Singular Value Decomposition Techniques for Enhanced Data Mining in Recommender Systems

The development of effective algorithms for processing large volumes of information is a very urgent problem for modern information systems. The work examines the problems of content analysis and data mining by recommender systems. Features of automatic systems management using machine learning algorithms are presented. The importance of using recommender systems for more accurate prediction of goods or services of interest to users, determination of interrelationships between various production factors, etc. have been determined. The operation of the FunkSVD algorithm, which allows accurate and fast processing of sparse data sets of a large volume, has been studied. A modification of FunkSVD is proposed to use fewer data about users and items when forming recommendations. The simulation of the proposed algorithm was carried out and it was determined that it works faster than the usual one, while maintaining fairly high accuracy, so it can be used in the processing of large data. It is also proposed to increase the privacy and accuracy of FunkSVD recommendations in distributed systems by using the Fed SVD algorithm. The results of the experiments showed the high accuracy of the proposed algorithm but at the expense of a certain increase in the duration of calculations. According to the conducted research, it was concluded that both modifications can be used in systems with different requirements for data mining and distributed architecture.