Research on the Construction of a Multi-Behavior Recommendation Algorithm Model for E-Commerce Platforms Based on Deep Learning

In multiple behavior scenarios, a combination of recurrent neural networks and convolutional neural networks is used to build recommendation models that integrate user behaviors and product characteristics in order to improve the performance of recommendation algorithms. This project takes 1,000,000 user behaviors and 100,000 products in the public dataset as the research object, adopts the temporal coding method to encode multiple behaviors such as clicking, purchasing, and bundling, and combines feature extraction such as text and images to achieve multi-dimensional information fusion. The experiment proved that the algorithm was significantly improved in terms of correct rate, recall rate, F1 score, etc., and the F1 value was 0.79; the average number of clicks and user satisfaction score of the top ten products in the ranking increased to 52% and 4.5 respectively, the study reveals that deep learning can better capture users' behavioral associations, and enhance users' personalization and accuracy rate.