【摘要】 在以往提取多兴趣嵌入的序列推荐模型中仅能通过聚类的方法发现少量兴趣概念，忽视项目交互序列中特征级信息对最终推荐结果的影响。针对此问题，对传统的多兴趣序列推荐模型进行改进，提出一种融合项目特征级信息的稀疏兴趣网络序列推荐模型。实验结果表明，相比其它模型，该模型可以更好捕捉用户的多样化偏好并缓解冷启动问题。在给定数据集上，该模型比传统的序列推荐模型在命中率上平均提高了6.4%,归一化折损累计增益平均提高了8.7%。更多还原

【Abstract】 Only a small number of interest concepts can be found by clustering in the previous sequence recommendation model for extracting multi interest embeddings, and these models also ignore the feature level information in the project interaction sequence. Aiming at these problem, the traditional multi-interest sequential recommendation model was optimized and a sparse-interest network sequential recommendation model integrating project feature-level information was proposed. Experimental results show that this model can better capture the diverse preferences of users and alleviate the cold start problem than other models. Compared with traditional sequence recommendation model, the average hit rate of this model is increased by 6.4%, and the average normalized discounted cumulative gain is increased by 8.7% on given data sets.更多还原