Analysis of the impact of cross-platform behaviour on recommendation quality

The rapid growth in the number of digital platforms and the diversity of online services create new challenges for the development of recommender systems that must factor in cross-platform user behaviour to ensure the accuracy and privacy of recommendations. The purpose of this study was to determine how combining cross-platform behavioural data can improve the accuracy of recommender systems. To this end, the study analysed modern machine learning algorithms and Big Data processing methods that enable the efficient integration of information from various sources. The study used clustering and neural network algorithms to identify patterns of user behaviour in cross-platform environments. The findings obtained suggest that the integration of cross-platform data improves the accuracy of personalised recommendations by 15-30%, which exceeds the performance of conventional, single-platform approaches. Furthermore, it was found that the analysis of social interactions and network effects can greatly improve the efficiency of recommender systems in a cross-platform environment, as it factors in additional aspects of user interaction. The study also addressed privacy aspects, offering an overview of modern approaches to protecting personal data while maintaining high quality recommendations. Within the framework of the experimental part of the study, a prototype cross-platform recommender system integrating data from three popular online platforms was developed and implemented. Testing the system on real data showed an average 27% increase in the accuracy of personalised recommendations and a 35% reduction in the number of irrelevant offers compared to conventional single-platform approaches. Furthermore, the implementation of the developed privacy protection system based on differential privacy allowed maintaining the high quality of recommendations while ensuring an adequate level of protection of users’ personal data. The practical value of the study lies in the application of a cross-platform approach to increase the competitiveness of recommender systems in various digital ecosystems