基于改进DeepFM的车险索赔预测模型的研究  

Research on Prediction Model of Auto Insurance Claim based on Improved DeepFM

作　　者：张姝[1] 陶宏才[1] ZHANG Shu;TAO Hongcai(School of Computing&Artificial Intelligence,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]西南交通大学计算机与人工智能学院,四川成都611756

出　　处：《成都信息工程大学学报》2021年第3期311-315,共5页Journal of Chengdu University of Information Technology

基　　金：国家自然科学基金资助项目(61806170)。

摘　　要：广义线性模型因其简单且输出结果具有可解释性被广泛应用于车险索赔预测领域,但不能识别特征之间交互作用从而限制了模型的表现力。DeepFM使用因子分解机和深度神经网络分别捕捉低阶和高阶特征交互,在数据稀疏的实际场景取得了显著效果。在因子分解机的基础上引入域相关的权重,针对特征存在互相干扰的问题提出相应缓解策略,并将轻量级的视觉注意力机制作用于深度神经网络进一步提升模型的表现力。实验结果表明,提出的模型相比于基本的DeepFM模型取得了更好的风险分割效果。Generalized linear model is widely used in the field of auto insurance claim prediction because of simplicity and interpretability.However,its expressiveness is limited because it can’t recognize the interaction between features.DeepFM uses Factorization Machine and Deep Neural Network to capture the interaction of low-order and high-order features respectively,and achieves remarkable results in the real scene with sparse data.This paper introduces the weight of domain correlation based on Factorization Machine,and proposes mitigation strategies to ease the problem of mutual interference between features.The lightweight visual attention mechanism is also applied to the Deep Neural Network to enhance the accuracy of the model.Experimental results show that the proposed model achieves better risk segmentation effect than the basic DeepFM model.

关 键 词：车险索赔 特征交互 DeepFM 注意力机制 

分 类 号：TP303[自动化与计算机技术—计算机系统结构]