RF-NTPP:基于规则融合的神经点过程模型OA

Neural temporal point process(NTPP)models are a classical approach for modeling and predicting event sequences.Addressing issues such as high-frequency event bias caused by class imbalance in existing neural point process models,this paper proposes a rule-fused neural temporal point process model(RF-NTPP).A two-stage rule mining strategy(TSRM)is employed:Logistic regression is used to mine a set of rules that are significantly associated with the target event,and the rules are refined based on model feedback.Rule fusion module RFM(rulefusion module)is used to embed these rules into the network representation,guiding the model to focus on modeling low-frequency and rare events.Temporal positional encoding enhancement module(TPEE)is introduced,enabling adaptive perception of complex time patterns.RF-NTPP is evaluated on three real-world medical datasets Stroke,Coroheart,and Sepsis.It achieves RMSEs of 0.88,1.05,and 0.31 for predicting the timing of low-frequency events,and classification accuracies of 65%,48%,and 69%respectively.All results outperform those of existing mainstream models.