基于多路交叉的用户金融行为预测

针对通过挖掘用户的金融行为来改善金融领域的服务模式和服务质量的问题,本文提出了一种基于多路交叉特征的用户金融行为预测算法.根据数据包含的属性构建训练的特征,基于因子分解机模型(FM)利用下游行为预测任务对金融数据的特征进行预训练,获取数据特征的隐含向量.引入特征交叉层对金融数据的高阶特征进行提取,解决FM线性模型只能提...

Prediction of user financial behavior based on multi-way crossing

To improve the service mode and service quality in the financial field by mining the financial behaviors of users, a user financial behavior prediction algorithm based on multi-way crossing (MCUP) is proposed in this paper. First, the training features are constructed based on the attributes contained in the data. Based on the FM model, the downstream behavior prediction tasks are used to pre-train the features of the financial data, and the hidden vectors of the features are obtained. Second, the feature cross-layer is introduced to extract high-order features of financial data, overcoming the disadvantage that the FM linear model can only extract low-order features. Then, the residual network structure is used to extract high-order features of financial data, solving the gradient disappearance problem caused by the too deep network layer. Finally, a unified multi-tower model integrated by the FM, feature cross network, and residual network is used to predict user financial behavior, blending low-order and high-order features. Experimental results show that the proposed algorithm can achieve a better accuracy rate in predicting user financial behavior.