Deep segmentation of retail customers based on improved DEC and multimodal semantic representation

With the advancement of digital transformation, the retail industry has accumulated a vast amount of customer data, particularly customer review data, which provides valuable insights into customer behavior and sentiment. Traditional customer segmentation methods mainly rely on market research and manual analysis. However, as the volume and complexity of data continue to grow, these traditional approaches struggle to meet the demands of precise segmentation and personalized marketing. As a result, machine learning-based customer segmentation methods have become a research focus. In particular, clustering algorithms are capable of identifying potential customer groups from large-scale datasets and providing a scientific basis for personalized marketing and product recommendations. With recent advances in natural language processing, especially the application of Bidirectional Encoder Representations from Transformers (BERT) models in text data processing, research on customer segmentation based on review data has gained increasing attention. Most current studies still focus on traditional clustering algorithms such as K-means and hierarchical clustering. However, these methods face limitations when dealing with high-dimensional sparse data and complex textual information. In addition, existing analyses of review texts often rely on traditional bag-of-words or Term Frequency–Inverse Document Frequency (TF-IDF) methods, which fail to fully capture the deep semantic information within the reviews. To address these challenges, this paper proposes an improved Deep Embedded Clustering (DEC) algorithm, incorporating BERT and Latent Dirichlet Allocation (LDA) models for vectorized representation and clustering analysis of review texts. This approach effectively overcomes the limitations of existing methods and enhances the accuracy and practicality of customer segmentation.