Construction of improved LSTM model for consumer psychological preference mining

Abstract

With the acceleration of digitization, consumers generate massive behavioral data in their daily business activities, which hide complex and dynamic psychological preferences, and are of key significance to the understanding of market trends and the optimization of business decisions. The traditional preference analysis model is difficult to capture the nonlinear dependence and long-term memory effect, so this study constructs an improved LSTM model, which enhances the focusing ability of the key information of the review by introducing the attention mechanism, and realizes the effective identification of consumer functional preferences and emotional preferences. Based on the review data of women’s clothing on large ecommerce platforms, the model uses a double-layer LSTM (Long Short-Term Memory) structure to extract temporal features, and combines with the attention layer to dynamically weight important semantic segments. The empirical results show that the LSTM-Attention model achieves an accuracy of 0.89 and a macro F1 score of 0.87, which is significantly better than benchmark models such as logistic regression and random forest, and the robustness test shows that the performance degradation of the LSTM-Attention model is controllable under data disturbance. The model provides a high precision and interpretable analysis tool for consumer preference mining, and helps to improve the effectiveness of personalized marketing strategies.