【摘要】 随着互联网时代的高速发展，用户面临信息过载问题，推荐系统应运而生.推荐系统一般分两个阶段，即推荐召回和推荐排序，推荐召回阶段主要用来筛选出一部分候选集以减小推荐排序阶段的计算压力.多兴趣个性化推荐系统对于每一个用户，算法能学习到用户的多种不同的兴趣偏好，然而目前的多兴趣召回算法只考虑了用户短期历史纪录，忽视了用户长期历史纪录中蕴含的丰富信息.针对这一问题，提出一种基于用户长短期历史的多兴趣召回算法，通过不同的神经网络模型结构分别建模用户长短期兴趣偏好，并通过门控融合网络融合用户长短期兴趣偏好，最终得到用户的多个兴趣偏好，实现了个性化推荐召回.在两个公开数据集上的实验证明了模型的有效性.更多还原

【Abstract】 With the rapid development of the internet era,users are facing the problem of information overload,and recommendation systems have emerged. Recommendation systems are generally divided into two stages: the recommendation recall stage and the recommendation ranking stage,with the main purpose of the recommendation recall stage being to select a part of the candidate set to reduce the computing load in the recommendation ranking stage. A multiinterest personalized recommendation system learns various users′ interest preferences for each user. However,current multiinterest recall algorithms only consider users′ short-term history and ignore the rich information contained in users′ long-term history. To address this issue,this paper proposes a multi-interest recall algorithm based on users′ long and short-term history.The algorithm models users′ long and short-term interest preferences through different neural network model structures and uses a gate fusion network to fuse users′ long and short-term interest preferences to ultimately obtain users′ multiple interest preferences,achieving personalized recommendation recall. The effectiveness of the model is demonstrated through experiments on two public datasets.更多还原