基于ALBERT-BiGRU模型的文本情绪分类研究  

Research on Text Emotion Classification Based on ALBERT-BiGRU Model

作　　者：李艳艳 冯柳鑫 徐梦舟 王笑一 张展鹏 Li Yanyan;Feng Liuxin;Xu Mengzhou;Wang Xiaoyi;Zhang Zhanpeng(State Grid Corporation of China Customer Service Center Southern Branch,Nanjing 211100)

机构地区：[1]国家电网有限公司客户服务中心南方分中心,江苏南京211100

出　　处：《中阿科技论坛（中英文）》2024年第12期80-84,共5页China-Arab States Science and Technology Forum

基　　金：国网客服中心南中心2024年基于全渠道多环节客户服务数据挖掘的电费智能催收项目(6B9931240009)。

摘　　要：针对国家电网客服中心繁杂的客户诉求问题,为将用户电话诉求对应的情绪识别出来,文章基于海量的95598电话语音文本数据集,构建了文本情绪分类模型。结合预训练语言模型ALBERT和双向门控循环单元设计ALBERT-BiGRU模型,利用ALBERT内部多层双向的Transformer架构学习高质量的动态语义特征,并引入BiGRU模型构造文本情绪分类器进行训练,提取不同抽象层次的语义信息特征。对照实验结果显示,相较于其他模型,ALBERTBiGRU文本情绪分类模型能有效提高分类精准度,在95598客户诉求情绪分类场景中具有较好的应用价值。In response to the complex customer demands of the State Grid Customer Service Center,this article constructs a text emotion classification model based on the massive 95598 telephone voice text dataset to recognize the emotions corresponding to user telephone demands.Design an ALBERT-BiGRU model by combining pre trained language model ALBERT and bidirectional gated recurrent unit.Utilize the multi-layer bidirectional Transformer architecture inside ALBERT to learn high-quality dynamic semantic features,and introduce the BiGRU model to construct a text emotion classifier for training,extracting semantic information features at different levels of extraction.The comparative experimental results show that compared with other models,the ALBERT-BiGRU text emotion classification model can effectively improve classification accuracy and has good application value in the 95598 customer demand emotion classification scenario.

关 键 词：ALBERT BiGRU 文本情绪 语音文本 数据挖掘 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]