Lightweight Deep Learning Model for Marketing Strategy Optimization and Characteristic Analysis

Abstract

The business model of traditional market is declining day by day, and people's consumption cognition has risen to a new level with the leap in science and technology. Enterprises need to adjust and optimize their marketing strategies in time according to the new consumption characteristics, so as to smoothly adapt to the environmental changes in the Internet age. This paper briefly analyzes the relationship between sales development and psychology and constructs a fusion model that can predict preferences with the help of neural network structure of the deep learning method. Describe user portraits and characteristics, analyze users' purchasing behavior and credit literacy, and push related products combined with a hash algorithm to achieve accurate e-commerce marketing purposes. The results show that (1) the model constructed in this paper and five different models are used for multi-modal recognition analysis: the accuracy is 79.56%, the recall rate is 77.43%, F1 is 0.785, and the error value can be reduced to about 0.18 by epoch iteration; the model is superior and has great use value. (2) Using the model to extract user attribute features and predict certain preferences, 13 topics and weight ratios are obtained for users of a certain platform, and the portrait model of each user is constructed. (3) According to the portrait optimization, 8 different marketing strategies are obtained, and the marketing effect is remarkable, fluctuating between 69% and 82%, and the income situation is also satisfactory. The final model design is reasonable and the data performance is good, which provides an intelligent and efficient dynamic strategy service for enterprises.