We offer to use the linked customers across public media sites and e-commerce sites (users who have public media accounts and have made purchases on e-commerce websites) as a link to map users’ public media functions to another function representation for item suggestions. In specific, we recommend learning both users’ and products’ function representations (called customer embeddings and item embeddings, respectively) from data gathered from e-commerce sites using repeated sensory systems and then apply a customized slope boosting trees and shrubs method to transform users’ public media functions into customer embeddings. We then develop a feature-based matrix factorization strategy which can make use of the learned customer embeddings for cold-start item suggestions. Trial results on a large dataset constructed from the biggest China micro blogging service SINA WEIBO and the biggest China B2C e-commerce website JINGDONG have shown the potency of our suggested structure. Further improvement of suggested recommended the system, we need to improve this application to genetic programming method (GP) for real-time function removal in on the internet public social networking sites. Our experimental assessment accomplishes efficiency in cross-sectional progress in on the internet public networks.