Research on User Preference Modeling and Data Enhancement Based on Generative Adversarial Networks in Precision Recommendation for E-Commerce Platforms

Abstract

With the advent of the intelligent internet era, e-commerce has also experienced explosive growth, leading to intensified competition among various e-commerce platforms. To accurately identify user preferences on e-commerce platforms and eliminate irrelevant items, this paper proposes a user-demand-based generative adversarial recommendation algorithm based on generative adversarial networks (GANs). This algorithm consists of multiple generators, with a discriminative model evaluating the generated items and providing feedback to the generative model to continuously improve it, ultimately recommending items with high similarity to the generated items. Comparative experiments were conducted on two datasets, and the model training was completed using the proposed model. Experimental simulations were conducted on the MovieLens-100K dataset. The experimental results indicate that the proposed algorithm can effectively improve accuracy. UR-GAN has advantages such as fast convergence speed and a stable training process, making the practical value of the recommendation system based on UR-GAN relatively high.