Enhancing Customer Segmentation Using Large Language Models (LLMs) and Deterministic, Independent-of-Corpus Embeddings (DICE)

Customer segmentation is an instrumental aspect of marketing efforts that helps businesses understand and target their customers better. However, customer segmentation using traditional clustering methods faces challenges in dealing with diverse data types, capturing semantic information in textual features, and capturing minor differences between customers. This paper explores a novel approach that integrates Large Language Models (LLMs) and Deterministic, Independent-of-Corpus Embeddings (DICE) for customer segmentation. LLMs are used to generate sentence embeddings for textual features, while DICE is used to generate numeracy preserving word embeddings for numerical features. The embeddings are then combined and clustered using k-means. The paper compares the performance of the proposed approach with the conventional approach of clustering without embeddings on two real-world datasets from different domains. The results show that the hybrid approach achieves better cluster separation and evaluation metrics, demonstrating the potential of LLMs and DICE for enhancing customer segmentation.