Adaptive domain transfer for persuasive user and item modeling and cold-start alleviation in next-basket recommendation systems

The Adaptive Persuasive User/Item Information Extraction and Cold-Start Mitigation (APIC) framework is designed to tackle key challenges in next-basket recommendation systems, specifically data sparsity, cold-start scenarios, and evolving customer preferences. Conventional Next Basket Recommender Systems (NBRS) often overlook the complex relationships both within and across shopping baskets, as well as the real-time influence of social media. To address these gaps, this study introduces APIC—a comprehensive and resilient NBRS model.

APIC leverages advanced embedding techniques to enrich the representation of user-item interactions and integrates real-time social media data to enhance recommendation accuracy. To further combat cold-start issues for new users and products, the framework employs domain adaptation strategies. In addition, APIC incorporates an improved Particle Swarm Optimization (PSO) algorithm inspired by the Rainbow Eucalyptus phenomenon, which optimally balances exploration and exploitation during the learning process. This advancement significantly boosts the performance of the attention-based Gated Recurrent Unit (GRU) by enabling it to more effectively capture both localized and broader user behavior patterns.

The effectiveness of APIC is validated on the IJCAI-15 and TaFeng datasets, demonstrating superior performance in comparison to standard NBRS models, particularly in terms of F1-score and NDCG metrics. These results highlight APIC’s potential to deliver highly relevant, timely, and precise recommendations, positioning it as a new benchmark for NBRS effectiveness. Ultimately, APIC’s innovative integration of social media insights and dynamic optimization techniques represents a meaningful advancement in recommendation technology, offering a powerful tool for modern e-commerce platforms.