3. 3.3. Performance evaluation results of improved collaborative filtering recommendation algorithm The score prediction results of the collaborative filtering recommendation algorithm using the fusion customer trust model come from two parts: the nearest neighbor of customer trust and the nearest neighbor of customer score similar. In order to determine the influence of the threshold number ρ of the two sets of intersection elements on the recommendation result, an experimental verification was performed. During the experiment, the alpha value in the scoring prediction formula is set to 0.8, the number of recommendations is N=15, the value of the threshold ρ is constantly changed, and the change of the MAE is observed. The experimental results are shown in Fig. 4. 3.4. The results of the prediction formula weight α In order to study the influence of the nearest neighbors of customer scoring on the recommendation results in the improved scoring prediction formula of this research model, cross-trial tests were used to analyze the value of α values. In the experiment, the recommended number was selected as N=15, ρ=16, and the results of the experiment were counted. The results obtained on this basis are shown in Fig. 5. 3.5. Performance comparison of different recommended algorithms The improved algorithm is compared with the collaborative filtering recommendation algorithm and the recommendation algorithm based on the customer trust model to verify the effectiveness of the improved algorithm. In the calculation of trust degree, the trust model proposed in this paper is used for calculation. The number of neighbors is from 4 to 30, with an interval of 4. The experimental results are shown in Fig. 6.