基于深度神经网络的带有评论文本的推荐模型

Deep Neural Networks Model of Texts Using Reviews for Recommendation

最近的研究表明,与传统的推荐方法相比,在有评论的文本情况下,深度学习方法可以更加有效地提高推荐系统的性能。例如,DeepCoNN模型使用神经网络来学习目标用户编写的所有评论文本的一种潜在表示形式,以及目标项目所有评论文本的第二种潜在表示形式,然后将这些潜在形式组合起来,以获得推荐任务的最新性能。评论文本的大部分预测价值来自目标用户对目标项目的评论。基于此,笔者首先介绍了一种将这些信息用于推荐的方法,即使目标用户对目标项目的评论不可用时也能用于推荐。该模型通过引入一个表示目标用户-目标项对的附加潜在层来扩展DeepCoNN模型;然后,在训练时将这个层加入正则化,使其类似于目标用户对目标项的另一个潜在表示。实验证明扩展版本比原版本的技术水平有了很大的改进。

Recent research has shown that deep learning methods can improve the performance of recommendation systems compared to traditional recommendation methods,especially in the case of texts with comments.For example,the recent model DeepCoNN uses a neural network to learn a potential representation of the text of all comments written by the target user,and a second potential representation of the text of all comments of the target item,and then combine these potential forms to represent Get the latest performance for recommended tasks.We found(unfortunately)that most of the predicted value of the review text comes from the target user’s comments on the target project.Then,we introduced a way to use this information for recommendations,even if the target user’s comments on the target item are not available.Our model is an extended version of the DeepCoNN model that extends the DeepCoNN model by introducing an additional potential layer that represents the target user-target item pair.Then,we add this layer to the regularization during training to make it similar to another potential representation of the target item by the target user.We have shown that this extended version has been greatly improved over the previous technical level.