Abstract: Aiming at the problems of incomplete information dimension, sparse interaction data and redundant interaction information in recipe recommendation tasks, a recipe recommendation model RecipeFlavor based on hierarchical learning of flavor embedding heterogeneous graph was proposed. First, the flavor molecule dimension was introduced, and a heterogeneous graph was constructed based on users, foods, ingredients and flavor substances of ingredients to effectively represent the connection between the four nodes. Then, a hierarchical learning module based on heterogeneous graph was constructed based on information transmission mechanism, and combined with compressed attention mechanism, different node relationships were regarded as different information channels, key interaction information between nodes was extracted and noise was suppressed. Finally, a contrastive learning module was constructed based on feature-aware noise, positive and negative sample discrimination tasks were introduced in model learning to enhance the information association between users and recipe nodes and improved the model's learning ability for features. Compared with HGAT (Hierarchical Graph Attention Network for Recipe Recommendation), the Area Under the Curve (AUC) of the RecipeFlavor model on the Recipe 1M+ large dataset increased by 1.44 percentage points, and the model Precision (Pre), Hit Rate (HR), Mean Average Precision (MAP) and Normalized Discounted Cumulative Gain (NDCG) of Top@10 increased by 0.76, 6.11, 2.68 and 3.05 percentage points respectively. Experimental results show that the introduction of flavor molecule information expands the learning dimension of recipe recommendation, and the RecipeFlavor model can effectively extract key information in heterogeneous graphs, enhance the correlation between users and recipes, and thus improve the accuracy of recipe recommendations.

Key words: Graph Convolution Network, Heterogeneous Graph Learning, Squeeze Attention Mechanism, Recommendation Algorithm, Contrastive Learning

摘要： 针对食谱推荐任务中信息维度不全面、交互数据稀疏和交互信息冗余的问题，提出了一种基于风味嵌入异构图层次学习的食谱推荐模型RecipeFlavor。首先，引入风味分子维度，基于用户、食物、食材以及食材的风味物质构建异构图，有效表示四种节点之间的联系；其次，基于信息传递机制构建基于异构图的层级学习模块，结合压缩注意力机制，将节点不同关系视为不同的信息通道，提取节点之间的关键交互信息并抑制噪声；最后，基于特征感知噪声构建对比学习模块，在模型学习中引入正负样本区分任务，增强用户和食谱节点之间的信息关联，提升模型对特征的学习能力。RecipeFlavor模型在Recipe 1M+大型数据集上，与HGAT（Hierarchical Graph Attention Network for Recipe Recommendation）相比，曲线下面积（AUC）提升了1.44个百分点，Top@10的模型精度（Pre）、命中率（HR）、平均精度（MAP）以及归一化折损累计增益（NDCG），分别提升了0.76、6.11、2.68、3.05个百分点。实验结果表明，风味分子信息的引入拓展食谱推荐的学习维度，而RecipeFlavor模型能够有效提取异构图中的关键信息，增强用户和食谱之间的关联性，从而提升食谱推荐精度。

关键词: 图卷积网络, 异构图学习, 压缩注意力机制, 推荐算法, 对比学习