Heterogeneous Graph Augmented Multi-Scenario Sharing Recommendation with Tree-Guided Expert Networks

Sharing recommendation is becoming ubiquitous at almost every e-commerce website, where a user will be recommended a list of users when he wants to share something with others. With the tremendous growth of online shopping users, sharing recommendation confronts several distinct difficulties: 1) how to establish a unified recommender model for large numbers of sharing scenarios; 2) how to handle with long-tail even cold start scenarios with limited training data; 3) how to incorporate social influence in order to make more accurate recommendations.

To tackle with the above challenges, we firstly build multiple expert networks to integrate different scenarios. During model training one specific scenario can learn to differentiate importance of each expert network automatically based on corresponding context information. With respect to the long-tail issue, we propose to maintain a complete scenario tree such that each scenario can utilize context knowledge from root node to leaf node to select the expert networks. At the same time, making use of the tree-based full path message contributes to alleviating training data sparsity problem. Moreover, we construct a large-scale heterogeneous user-to-user graph which is derived from various social behaviors at e-commerce websites. Then a novel scenario-aware multi-view graph attention network is leveraged to augment user representations socially. In addition, an auxiliary inconsistency loss is applied to balance the load of expert networks, along with main click-through rate (CTR) prediction loss, the whole framework is trained in an end-to-end fashion. Both offline experiments and online A/B test results demonstrate the superiority of proposed approach over a bunch of state-of-the-art models.