基于注意力门控神经网络的社会化推荐算法

针对社会化推荐算法中存在的推荐准确率不高的问题，提出了一种多头注意力门控神经网络（MAGN）算法。具体来说，采用门控神经网络对输入的用户和用户-朋友对进行融合得到联合嵌入，利用注意力记忆网络来获取不同朋友在不同方面对用户的影响，利用多头注意力来获取在不同方面对用户影响程度偏高的几位朋友。采用门控神经网络将朋友影响和用户自身兴趣偏好进行混合，继而基于混合兴趣偏好对项目进行推荐。在两个公开真实数据集上进行实验进一步验证了所提方法的有效性。

Social Recommendation Algorithm Based on Attention Gated Neural Network

To address the problem of low recommendation accuracy in social recommendation algorithms,a multi-headed attention gating neural network(MAGN)algorithm is proposed.Specifically,a gated neural network is used to fuse the input user and the user’s friend pairs to obtain a joint embedding,then an attentional memory network is used to obtain the influence of different friends on the user in different aspects,and then multi-headed attention is used to obtain several friends who have a high degree of influence on the user in different aspects.Finally,a gated neural network is used to mix friend influence with the user’s own interest preferences,and items are then recommended based on mixed interest preferences.Experiments on two publicly real available datasets further validate the effectiveness of the proposed method.