A User Purchase Motivation-Aware Product Recommender System

Abstract

Retailers struggle to match recommendations to why customers buy. We introduce a practical framework that distinguishes two core, actionable purchase motivations, stable preference and exploratory intent, and present STB, a data-efficient measure that infers which motivation drives each item purchase using only transaction sequences and item attributes. Building on STB, we develop UPSTAR, a motivation-aware recommender that separates users’ behavior into stable-preference and exploratory subsequences and fuses their signals for next-item prediction. Across three real-world e-commerce data sets, UPSTAR substantially improves accuracy and, importantly, advances the system’s ability to surface genuinely exploratory items that drive discovery and cross-category sales. For practitioners, our method enables more targeted marketing: promote reliable items to preference-driven buyers while exposing exploratory buyers to curated novelty, improving conversion and long-term engagement. For platform policy and operations, motivation-aware recommendations support inventory planning, personalized promotions, and responsible diversification of exposure without requiring surveys or extensive auxiliary data. Implementation requires only existing transaction logs and item metadata, making it immediately deployable for large-scale retail systems.