Abstract: The large and sparse data in group recommendation often tend to ignore the complex dependencies between user groups and items, so fusing different user preference behavioral embeddings to make the performance of user dependencies on groups more intuitive, and also for the purpose of enhancing the view effect in comparison to obtain more accurate recommendation results, an interpretable comparison meta-path group recommendation framework based on hypergraph is proposed. By aggregating dependencies between groups of user items, it constructs meta-paths to represent different types of interactions between entities to promote entity similarity and more accurately obtain users’ in-group and out-group interactions from the data. By combining interpretable models with contrast learning techniques, it improves the interpretability and performance of the models. By interpreting guided enhancement operations on positive and negative views generated on the model framework combined with self-supervised contrast learning, it addresses the above issues. This experiment validates the effectiveness of the proposed approach by conducting experiments on real datasets.

Key words: group recommendation, hypergraph learning, meta-paths, recommendation system, contrastive learning

摘要： 在群组推荐中庞大且稀疏的数据往往容易忽视用户群组及项目之间的复杂依赖关系，因此融合不同用户偏好行为嵌入，使用户对群组依赖关系的表现更直观，同时为了在对比中增强视图效果，以获得更准确的推荐结果的目的，提出了一个面向超图的可解释性对比元路径群组推荐框架。通过聚合用户项目群组之间的依赖关系，构建元路径表现实体之间的不同类型交互，以促进实体的相似性，更准确地从数据中获取用户的组内、组外交互；通过将可解释性模型与对比学习相结合的技术，以提高模型的可解释性和性能；通过解释引导增强操作在模型框架上生成的正负视图上结合自监督对比学习，来解决上述问题。在真实数据集上进行实验，验证了所提出方法的有效性。

关键词: 群组推荐, 超图学习, 元路径, 推荐系统, 对比学习