Intent-Aware Sequential Recommendation for Small Retail: A Hybrid CatBoost-GRU Framework Using Survey and Transaction Data

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Small physical retailers face a fundamental challenge: making personalized decisions with sparse, heterogeneous customer data.Unlike large e-commerce platforms, they lack the interaction volume required by conventional machine learning and recommendation systems.This paper proposes an intent-aware recommendation framework that jointly models customer purchase readiness and next-product preference using survey and transactional data from a micro-retail juice bar (Sundew Juicebar), supplemented with synthetically augmented records that preserve the original behavioral and demographic distributions.Purchase intent is formulated as a customer-level classification task, achieving a mean AUC of 0.801 under cross-validation, while next-item prediction is addressed using sequential and co-occurrence-based methods.The results show that classical recommendation techniques remain competitive in menu-constrained environments, while sequential models capture complementary behavioral patterns.By integrating intent prediction with recommendation through a weighted fusion mechanism, the proposed framework improves Accuracy@1 by 5.00% and NDCG@3 by 0.89% over a standalone sequential model.Beyond predictive performance, the framework provides interpretable decision support, enabling small-business operators to understand and act on model outputs.The results demonstrate that combining behavioral and attitudinal signals yields practical value in real-world micro-retail settings, even under severe data constraints.