微型电脑应用

针对传统神经协同过滤模型辅助信息利用不足的问题，提出一种结合自然语言处理方法的神经协同过滤模型。所提方法以评论数据作为用户特征，嵌入神经协同过滤模型，从而弥补模型因辅助特征缺失，导致的推荐精度与个性化程度低的问题。采用SBERT模型对用户评论文本进行语义提取，通过聚类获得用户辅助特征，再通过嵌入层将其与神经协同过滤模型进行组合。在神经协同过滤框架下，通过学习获得的特征从线性和非线性两方面，模拟用户和项目间的交互关系，并在最后的隐藏层进行结合，给出预测评分。在相同数据集下，将提出的组合模型与其他同类推荐模型进行实验对比。实验结果表明，相比其他同类推荐模型，该组合模型的推荐效果得到了一定提升。

To address the problem of insufficient utilization of auxiliary information in the traditional neural collaborative filtering model, this study proposes a neural collaborative filtering model combined with a natural language processing method. The proposed method uses review data as user features and embeds them into the neural collaborative filtering model, thus compensating for the lack of recommendation accuracy and personalization caused by the lack of auxiliary features in this model. The SBERT model is used to semantically extract the user review text, and the user auxiliary features are obtained through clustering, and then combined with the neural collaborative filtering model through the embedding layer. In the neural collaborative filtering framework, the features obtained by learning from both linear and nonlinear aspects, simulate the interaction between users and items, and combine them in the final hidden layer to give predicted scores. The combined model proposed in this study is experimentally compared with other similar recommendation models under the same data set. The experimental results show that the combined model can effectively improve the recommendation prediction compared with other similar recommendation models.