基于改进词嵌入表示和卷积神经网络的银行智能派单系统的研究

银行智能派单系统的实现和功能完善,对银行提升客户满意度、提高突发事件处理效率、降低人工处理成本等非常重要。针对现有的基于Word2vec和TextCNN模型的银行智能派单系统进行了改进,针对特征词权重表达性弱,特征词类别及位置区分性弱等问题,提出基于改进TF-IDF加权的Word2vec词嵌入表示和卷积神经网络结合的银行智能派单系统:首先利用Word2vec模型得到输入事件单的词嵌入向量;再针对经典TF-IDF方法不具备类别区分性、位置区分性,也没有考虑极端频率特征词代表性的情况,提出改进型TF-IDF算法,计算每个特征词的权重,得到基于改进TF-IDF加权的Word2vec词嵌入表示;最后在卷积神经网络模型中进行训练,通过迭代训练最终得到分类器,利用分类器可对输入事件单信息自动进行系统类别的判断。实验结果表明改进词嵌入表示的银行智能派单系统分类模型的宏查准率、宏查全率、准确率以及宏F1值都得到进一步的提高。

Research on Bank Intelligent Dispatch System Based on Improved Word Embedding Representation and Convolutional Neural Network

This paper improves the existing bank intelligent dispatch system based on the Word2vec and TextCNN models.In order to solve the problems of weak expression of feature word weights,weak classification of feature words and location,etc,propose bank intelligent dispatch system based on improved TF-IDF weighted Word2vec word embedding representation and convolutional neural network:first use the Word2vec model to get the word embedding vector of the input event information,then because the classic TF-IDF method does not have class discrimination,location discrimination,and does not consider the representativeness of extreme frequency feature words,an improved TF-IDF algorithm is proposed in this paper,and the weight of each feature word is calculated to obtain improved TF-IDF weighted Word2vec word embedding representation.Finally,it is trained in a convolutional neural network model and a classifier is obtained.the classifier can be used to automatically determine the system category of the input event information.