HGV4Risk: Hierarchical Global View-guided Sequence Representation Learning for Risk Prediction

Risk prediction, usually achieved by learning representations from patient’s physiological sequence or user’s behavioral data, and has been widely applied in healthcare finance. Despite that some recent time-aware deep methods have led to superior performances such representation tasks, improvement is limited due a lack of guidance hierarchical global view. To address this issue, we propose novel end-to-end H ierarchical G lobal V iew-guided (HGV) framework. Specifically, the Global Graph Embedding (GGE) module proposed learn sequential clip-aware temporal correlation graph at instance level. Furthermore, following way key-query attention, harmonic β -attention ( -Attn) also developed for making trade-off between decay observation significance channel level adaptively. Moreover, both can be coordinated heterogeneous information aggregation under Experimental results on risk prediction benchmark SMEs credit overdue task real-world industrial scenario MYBank, Ant Group, illustrated model achieve competitive performance compared with other known baselines. The code released public available at: https://github.com/LiYouru0228/HGV.