融合画像约束和潜在特征的深度推荐算法
A Deep Recommendation Algorithm Integrating Portrait Constraint and Latent Feature

基于深度学习的推荐算法已成为推荐系统领域的研究趋势。然而，大多数现有工作仅考虑单一的用户与物品交互数据，限制了算法的预测性能。本文提出一种画像约束的编码方式，并融合隐因子模型中的潜在特征，丰富了推荐算法的输入以提升评分预测的准确性。该算法利用矩阵分解得到潜在特征初始化用户与物品的嵌入，然后通过线性注意力机制增强模型对画像特征的敏感度，最后结合深度神经网络进行评分预测。通过本文算法与其他基线算法在MovieLens与Netflix数据集上进行对比，该算法与基线算法相比显著提高了评分预测的精度，并在推荐列表排序性能等方面表现出色。本文的研究揭示了加入用户与物品的画像约束和潜在特征，可以有效提升推荐系统的性能。
Recommendation algorithms based on deep learning have emerged as a prominent research direction in the field of recommender systems. However, most existing studies focus primarily on single user-item interaction data, which limits the prediction performance of these models. This article proposes a method for encoding portrait constraints and integrates potential features from hidden factor models, increasing the input of recommendation algorithms to improve the accuracy of rating prediction. This algorithm utilizes matrix decomposition to initialize the embedding of latent features between users and items, and then enhances the sensitivity of the model to portrait features through linear attention mechanism. Finally, it combines deep neural networks for rating prediction. By comparing the algorithm proposed in this article with other baseline algorithms on the MovieLens and Netflix datasets, it was found that this algorithm significantly improved the accuracy of rating prediction compared to the baseline algorithm, and performed well in terms of recommendation list ranking performance. This study reveals that incorporating portrait constraints and latent features of users and items can effectively improve the performance of recommendation systems.