User preference information recommendation based on DCGNN and GNNPK algorithms

With the rapid popularization of the Internet and mobile devices, the problem of information overload faced by users is becoming increasingly serious, resulting in traditional recommendation systems being inadequate in meeting the diverse needs of users. Especially when dealing with challenges such as user dynamic preferences, data noise, and cold start, the limitations of these systems become increasingly obvious, and more advanced algorithms are urgently needed to improve the personalization and accuracy of recommendations. To this end, the study first used a dual channel graph neural network as the basis and introduced a personalized knowledge aware attention network for denoising user preference information. On this basis, knowledge graph technology was introduced to deepen the understanding and feature extraction of user preference information. Finally, a novel user preference information recommendation model was developed. The experiment findings indicated that the highest recommendation accuracy of the new model in Tok-5, Tok-10, and Tok-15 environments was 88%, 90%, and 84%, respectively. The lowest recommendation errors for music, movies, books, and travel were 3.4%, 5.1%, 2.8%, and 4.3%, respectively, and the lowest recommendation time was 0.3 minutes. The highest recommended coverage rate could reach 93.4%, the lowest resource consumption rate could reach 18.7%, and the highest user satisfaction rate was 95.6%. It can be seen from this that the proposed recommendation model performs well in handling user preference information. This means that this method has good performance and potential in practical applications, providing new methods and ideas for optimizing the recommendation system, and has important academic value and practical application significance.