Hierarchical User Intent Graph Network forMultimedia Recommendation
This work addresses the challenge of enhancing multimedia recommendation by modeling hierarchical user intents, representing an incremental advancement in graph-based recommendation systems.
The paper tackles the problem of learning multi-level user intents from item co-interaction patterns to improve recommendation performance, resulting in significant improvements over state-of-the-art methods like MMGCN and DisenGCN on three public datasets.
In this work, we aim to learn multi-level user intents from the co-interacted patterns of items, so as to obtain high-quality representations of users and items and further enhance the recommendation performance. Towards this end, we develop a novel framework, Hierarchical User Intent Graph Network, which exhibits user intents in a hierarchical graph structure, from the fine-grained to coarse-grained intents. In particular, we get the multi-level user intents by recursively performing two operations: 1) intra-level aggregation, which distills the signal pertinent to user intents from co-interacted item graphs; and 2) inter-level aggregation, which constitutes the supernode in higher levels to model coarser-grained user intents via gathering the nodes' representations in the lower ones. Then, we refine the user and item representations as a distribution over the discovered intents, instead of simple pre-existing features. To demonstrate the effectiveness of our model, we conducted extensive experiments on three public datasets. Our model achieves significant improvements over the state-of-the-art methods, including MMGCN and DisenGCN. Furthermore, by visualizing the item representations, we provide the semantics of user intents.