Learning personalized preference: A segmentation strategy under consumer sparse data

Personalization is crucial to consumer satisfaction and repeat purchases. Learning personalized preferences from consumer data is often challenged by data sparsity because each consumer's history only accounts for a small part of the total data. Since consumer segmentation and personalized preference are often highly correlated and can be mutually reinforcing, the segmentation strategy to alleviate the data sparsity problem is proposed. The proposed strategy utilizes a Nonparametric Hierarchal Bayesian method to integrate Preference learning and consumer Segmentation into a unified model, named as NHBPS, that estimates the number of segmentations and preferences in a data-adaptive way. The NHBPS model is applied to a real-world dataset containing consumer behavior history and demographics. The analysis results show the following: (1) at least 17.50% improvement in consumer segmentation compared to benchmark methods. (2) personalized preference distributions and insightful interpretations are learned successfully. (3) the NHBPS model outperforms benchmarks in predicting consumer behaviors and demographics.