基于LightGBM与深度兴趣网络Stacking融合模型的商品推荐算法
Commodity Recommendation Algorithm Based on the Fusion Model of LightGBM and Deep Interest Network Stacking

随着电子商务平台的迅速发展，如何提高用户对平台的忠诚度并稳定客流，进而调整平台运营方向以获得持续的收益，成为当前电子商务平台急需解决的关键问题。常见于电商平台的推荐系统利用用户的购买、收藏、浏览等数据，采用特定的算法向用户推荐商品。本研究提出了一种基于LightGBM与深度兴趣网络Stacking融合模型的商品推荐的新解决方案。该模型根据用户过去一年的交易记录提取相应的商品特征和用户特征，整合协同过滤的多路召回策略与这些特征，并将其作为模型的输入，以预测下单客户可能购买的产品并进行商品推荐。研究结果表明，在测试数据上，相对于其他常用推荐算法，本文提出的模型具有更高的准确性、更快的预测速度和更好的推荐效果。这些研究结果为电子商务企业提供了改进服务的契机，为相关研究和实践提供了有益的参考和借鉴，为商品推荐问题的解决提供了有价值的参考和帮助。
With the rapid development of e-commerce platforms, how to improve user loyalty to the platform and stabilize customer flow, and then adjust the direction of platform operation to obtain sustained revenue, has become a key issue that e-commerce platforms urgently need to solve. Recommenda-tion systems commonly used in e-commerce platforms utilize user’s purchase, collection, browsing and other data to recommend commodities to users using specific algorithms. In this study, we propose a new solution for commodity recommendation based on the fusion model of LightGBM and deep interest network Stacking. The model extracts the corresponding commodity features and us-er features based on the user’s transaction records in the past year, integrates a collaborative fil-tering multiplexed recall strategy with these features, and uses them as inputs to the model in or-der to predict the commodities that the customers placing the order are likely to purchase and make commodity recommendations. The research results show that the model proposed in this paper has higher accuracy, faster prediction speed, and better recommendation effect than other commonly used recommendation algorithms on the test data. These findings provide e-commerce enterprises with opportunities to improve their services, provide useful references and lessons for related research and practice, and provide valuable references and assistance in solving the prob-lem of commodity recommendation.