基于矩阵分解的记忆网络模型

Memory Network Model Based on Matrix Decomposition

现有的两类推荐模型(基于循环神经网络的推荐模型和基于协同过滤的推荐模型)将用户编码在一个低维想两种,限制了用户偏好的记忆能力。为增大模型记忆空间,匹配用户需求,提出基于混合矩阵分解的记忆网络模型,在基于矩阵分解的框架下混合使用深度神经网络和记忆网络,得到两个层级的用户表示。

The existing two kinds of recommendation models(the recommendation model based on recurrent neural network and the recommendation model based on collaborative filtering)encode users in one low dimension,which limits the memory ability of users'preferences.In order to enlarge the memory space of the model and match the needs of users,a memory network model based on hybrid matrix decomposition is proposed.In the framework of matrix decomposition,deep neural network and memory network are mixed to obtain two levels of user repre⁃sentation.