基于信息融合的商品推荐算法设计与实现

Design and Implementation of Product Recommendation Algorithm Based on Information Fusion

在大数据时代,推荐系统提高了信息获取和分发的效率,但现有的推荐方法和效果还有很大的提升空间,需要更多的投入和研究。文章在召回阶段使用多路融合,融合后的列表对ItemCF算法和FM列表中的项目得分进行平均加权,使得模型在候选集数量N相同的情况下,获得了比单一召回方式更高的召回率。在排序阶段使用优化过的深度兴趣网络(Deep Interest Network,DIN)模型,对比其他模型,其曲线下面积(Area Under Curve,AUC)指标显著提升,验证了优化过的DIN模型性能的优越性。

The recommendation system has improved the efficiency of information acquisition and distribution in the age of big data, but the existing recommendation methods and effects still have much room for improvement, requiring more investment and research. In the recall stage, this paper uses multi way fusion. The fused list weighs the average scores of the items in the ItemCF algorithm and FM list, so that the model can obtain a higher recall rate than the single recall method when the number of candidate sets N is the same. In the sorting stage, the optimized Deep Interest Network(DIN) model is used. Compared with other models, its AUC index is significantly improved, which verifies the superiority of the optimized DIN model.