Research on User Profile Multi Model Fusion Recommendation System Based on Spark

In order to improve the real-time, accuracy and diversity of items recommended by e-commerce platform, and improve users' satisfaction and loyalty to recommended items, this paper proposes a method of combining recommendation based on user profile characteristics and real-time recommendation based on user profile behavior to realize the recommendation function. First of all, in the recommendation based on user profile features, the improved TF-IDF and Word2Vec models are weighted and fused, and the user's interest preference values are deeply mined by combining with the text classification convolutional neural network TextCNN, and the Top-N ranking recommendations are made. Secondly, in the real-time recommendation based on user profile behavior, Flume+Kafka is used to collect real-time data, and Spark Streaming obtains real-time data and then combines alternating least squares ALS algorithm and priority ranking algorithm to realize real-time recommendation function. In the last two models, modulation fusion strategy is recommended to improve the recommendation effect of personalized items and the satisfaction and loyalty of users to items. The feasibility and practicability of multi-model fusion recommendation system are proved by experiments.