摘要
电力投诉工单中往往存在长文本数据,这对工单分类模型的构建是一种挑战。以提升工单分类准确度为目的,提出了一种基于分级信息融合的电力投诉工单分类模型来提高模型分析长文本的能力。使用Word2vec方法对句中的单词进行处理,进而得到单词向量和句子矩阵。利用双向长短时记忆网络(BiLSTM)来学习单词间的依赖关系,同时运用TextCNN学习句子间的相互关联。将各级学习到的深度语义特征利用多层感知机(MLP)实现特征层融合。所提出模型在包含3万真实电力投诉工单样本的数据集上进行实验,5类投诉的平均分类正确率为0.921,平均宏-F_(1)分数为0.901,正确率相较于TextCNN、BiLSTM以及深度置信网络(DBN)分别提升了1.9%、5.3%和13.5%,能够完成投诉工单分类任务。
There are long text data of power customer service tickets,which is a challenge to the construction of the model to classification power customer service tickets.Therefore,this paper proposes a classification model based on hierarchical information fusion to improve the analysis ability of long text.The Word2vec method is used to process the words in the sentences,and then the word vector and sentence matrix are obtained.The bidirectional long-term and short-term memory network(BiLSTM)is used to learn the dependence between words,and the TextCNN is used to learn the correlation between sentences.The multi-layer perceptron(MLP)is used to extract the deep semantic features,which are learned at all levels to achieve feature layer fusion.The proposed model is tested on a dataset containing thirty thousand real power customer service ticket samples,the average classification accuracy of the five types of service tickets is 0.921,and the average macro-F1 score is 0.901.The results show that compared with TextCNN,BiLSTM,and deep belief network(DBN),the recognition accuracy of the proposed method is improved by 1.9%,5.3%,and 13.5%,respectively,which can give an outstanding performance on the classification of power customer service tickets.
作者
张莉
王颖
赵阳
崔涵翔
刘娟
ZHANG Li;WANG Ying;ZHAO Yang;CUI Hanxiang;LIU Juan(Customer Service Center of State Grid Corporation of China,Tianjin 300306;Beijing China-Power Puhua Information Technology Co.,Ltd.,Beijing 100031)
出处
《微型电脑应用》
2023年第11期87-90,共4页
Microcomputer Applications
