| 摘 要: 基于线性模型的矩阵分解推荐算法对信息的特征提取单一，当用户和物品含有大量隐含信息时，无法满足用户需求的个性化推荐。针对此问题，提出一种评分和社会标签融合的卷积神经网络推荐算法，该算法能够根据上下文信息，利用非线性模型提取隐含高阶信息，处理复杂且稀疏的数据。首先，设计由两路由多层感知器和卷积神经网络组成的深层网络结构，分别实现利用社会标签信息和用户评分信息建模用户兴趣和项目信息的潜在特征向量；然后，构建对多层神经网络学习后的结果进行融合的输出层，得出预测结果；最后，运用真实数据集进行实验验证。结果表明，该算法与当前主流的推荐模型相比，能更好地利用社会标签信息进行精准推荐。 |
| 关键词: 评分；社会标签；卷积神经网络；推荐模型 |
|
中图分类号: TP391.3
文献标识码: A
|
| 基金项目: 辽宁省教育厅科学研究计划资助项目(JZR2019009). |
|
| Research on Convolutional Neural Network Recommendation Model Combining Scoring and Social Tags |
|
ZHENG Dongxia
|
(School of Software, Dalian Neusoft University of Information, Dalian 116023, China )
zhengdongxia@neusoft.edu.cn
|
| Abstract: Matrix factorization recommendation algorithm based on linear model extracts single characteristics of information. It cannot meet user's need for personalized recommendation when users and items contain a large amount of hidden information. Aiming at this problem, this paper proposes a convolutional neural network recommendation algorithm combining rating and social tags. The algorithm can extract hidden high-order information using a nonlinear model based on context information, and process complex and sparse data. First, a deep network structure composed of two-route multilayer sensors and convolutional neural networks is designed, which realizes the use of social tag information and user rating information to model the potential feature vectors of user interests and item information. Then, a multi-layer neural network after learning is built, and the fusion output layer is used to obtain the prediction result. Finally, the real data set is used for experimental verification. The results show that compared with the current mainstream recommendation models, the proposed algorithm can better utilize social tag information for accurate recommendation. |
| Keywords: rating; social tag; convolutional neural network; recommendation model |
