IFP-DeepFM: Integrating Improved FP-Growth and Attention-Based Deep Factorization Machines for User Purchase Behavior Modeling and E-commerce Recommendation

Abstract

This study examines how intelligent recommendation algorithms can be used to analyze user purchasebehavior on e-commerce platforms and to support the development of more effective marketing strategies.First, a systematic review of existing recommendation algorithms is conducted, which shows thattraditional methods have limitations when processing large-scale and complex user data. To address theseissues, an Improved Frequent Pattern-growth (IFP-growth) algorithm is introduced. By incorporatingtime-window parameters, the algorithm mines frequent patterns from user behavior logs and extractstimely and relevant purchasing features. In parallel, an Attention-enhanced Deep Factorization Machine(DeepFM) is adopted, integrating the strengths of Factorization Machines and Deep Neural Networks.The attention mechanism further refines feature-interaction representations, enabling more accurate andpersonalized recommendations. To validate the effectiveness and robustness of the proposed model,experiments are conducted using real behavioral data from a major e-commerce platform over a threemonth period, comprising approximately 1.2 million behavior logs, 80,000 active users, and 100,000products. The IFP-DeepFM model is compared with traditional collaborative filtering, IFP-growth, andstandard DeepFM algorithms. The results show that IFP-DeepFM outperforms DeepFM in precision,recall, F1-score, and the area under the precision–recall curve by 6.81%, 11.9%, 9.51%, and 11.5%,respectively. In practical metrics such as Click-Through Rate (CTR) and Conversion Rate (CVR), themodel also achieves improvements of 10.4% and 8.9%. Based on these results, this study outlines severalmarketing strategy implications for e-commerce platforms, including personalized recommendations,targeted advertising, and user behavior prediction. These strategies enhance user satisfaction andplatform performance while providing data-driven support for operational decision-making. Theproposed IFP-DeepFM model demonstrates a practical approach to analyzing user purchase behaviorand refining marketing strategies in large-scale e-commerce settings.