In this paper, we measure the effectiveness of recommendations using the provider coverage of reinforcement learning algorithms under new media videos, using parameters smaller than the state space to estimate the value function. The reinforcement learning algorithm determines the population's state and maximum fitness by specifying the number of gene updates per generation. Based on the sample characteristics of the mixed cross-section data, the least squares benchmark regression sets the category of agricultural products and finds that the growth rate of cross-border e-commerce in agricultural products decreases from 28.57% to 11.66%, the highest logistics and distribution accuracy is 11.987, and the cross-border e-commerce in agricultural products has a coverage rate of about 5.57%. It shows that utilizing a reinforcement learning algorithm ensures the maintenance of the efficiency of e-commerce applications and can achieve a comprehensive change in sales.