ConsRec: Learning Consensus Behind Interactions for Group Recommendation
ConsRec: Learning Consensus Behind Interactions for Group Recommendation
复制标题
DOI：
10.1145/3543507.3583277
复制
发表时间：
2023-02
期刊：
影响因子：
--
通讯作者：
Xixi Wu;Yun Xiong;Yao Zhang;Yizhu Jiao;Jiawei Zhang;Yangyong Zhu;Philip S. Yu
中科院分区：
文献类型：
--
作者：
Xixi Wu;Yun Xiong;Yao Zhang;Yizhu Jiao;Jiawei Zhang;Yangyong Zhu;Philip S. Yu
由于群体活动在日常生活中变得非常普遍，因此迫切需要为一组用户生成推荐，称为群体推荐任务。现有的群体推荐方法通常通过聚集不同成员的兴趣来推断群体的偏好。实际上，群体的最终选择是成员之间的妥协，最终达成一致。然而，现有的个人信息聚合缺乏整体的群体层面的考虑，未能捕捉到共识信息。此外，它们的特定聚合策略要么遭受高计算成本，要么变得过于粗粒度，无法进行精确的预测。为了解决上述局限性，在本文中，我们专注于探索群体行为数据背后的共识。为了全面捕捉群体共识，我们创新地设计了三个不同的视图，提供相互补充的信息，使多视图学习，包括成员级聚合，项目级口味，和群体级的固有偏好。为了融合和平衡多视图信息，进一步提出了自适应融合组件。至于成员级聚合，不同于现有的线性或专注的策略，我们设计了一种新的超图神经网络，允许有效的超图卷积运算，以产生富有表现力的成员级聚合。我们在两个真实世界的数据集上评估了我们的ConsRec，实验结果表明我们的模型优于最先进的方法。一个广泛的案例研究也验证了共识建模的有效性。
Since group activities have become very common in daily life, there is an urgent demand for generating recommendations for a group of users, referred to as group recommendation task. Existing group recommendation methods usually infer groups’ preferences via aggregating diverse members’ interests. Actually, groups’ ultimate choice involves compromises between members, and finally, an agreement can be reached. However, existing individual information aggregation lacks a holistic group-level consideration, failing to capture the consensus information. Besides, their specific aggregation strategies either suffer from high computational costs or become too coarse-grained to make precise predictions. To solve the aforementioned limitations, in this paper, we focus on exploring consensus behind group behavior data. To comprehensively capture the group consensus, we innovatively design three distinct views which provide mutually complementary information to enable multi-view learning, including member-level aggregation, item-level tastes, and group-level inherent preferences. To integrate and balance the multi-view information, an adaptive fusion component is further proposed. As to member-level aggregation, different from existing linear or attentive strategies, we design a novel hypergraph neural network that allows for efficient hypergraph convolutional operations to generate expressive member-level aggregation. We evaluate our ConsRec on two real-world datasets and experimental results show that our model outperforms state-of-the-art methods. An extensive case study also verifies the effectiveness of consensus modeling.