The consumer behavior of shopping and entertainment on multimedia platforms has more multidimensional data properties than before, which offers more information to create brand user profiles and formulate recommendation strategies in the internet age. The paper chooses Brand H as the object of study and simulates and evaluates static data and rating data of 20,000 users in China to serve as a reference model of brand user profiling. To enable the user data clustering and profiling, the K-Means clustering algorithm is chosen as the data aggregation algorithm. Simultaneously, a new recommendation model is presented, which combines the traditional stacking, but with increased dimensions of input features and weighting of various features. It results in better-stacking-based recommendation model with enhanced prediction accuracy. When building user profiles and providing personalized recommendations to consumers regarding Brand H, the recommended improved stacking recommendation model had a maximum average reciprocal hit rate of 0.281 and a maximum average precision of 0.738, indicating that it can be applied effectively in practice.
