基于GRM的电力智能客服语义理解系统设计

Design of semantic understanding system for power intelligent customer service based on GRM

摘要

为了提高电力智能客服系统的自动化、智能化水平,解决当前语义解析精度不高、稳定性不足及响应延迟等问题,研究基于文法规则匹配(grammar rule matching,GRM)网络,引入规则配置和基于随机条件场(conditional random field,CRF)的命名实体识别技术,结合朴素贝叶斯分类器与规则分组配置,实现高效的用户意图和语义信息匹配,完成语句到结构化语义的解析。同时,通过本地缓存优化意图分发机制,降低系统计算复杂度。实验结果显示,优化后系统响应时间降低66.23%,基于GRM与命名实体的语义理解系统较基于模型的系统的F₁分数高出3.79%,其平均F₁分数达89.6%,意图识别准确率超90%。该系统具有更高的业务服务精度和稳定性,能有效适配电力咨询、故障报修等场景,满足用户对电力信息咨询的实时性和准确性要求。

Abstract

To improve the automation and intelligence of the power intelligent customer service system and address issues such as low semantic parsing accuracy, instability, and response latency, this study investigates a grammar rule matching (GRM) network. It introduces rule configuration and named entity recognition technology based on conditional random field (CRF), combining a Naive Bayes classifier with rule grouping configuration to achieve efficient matching of user intent and semantic information, completing the parsing from statements to structured semantics. Simultaneously, local caching optimizes the intent distribution mechanism, reducing system computational complexity. Experimental results show that the optimized system response time is reduced by 66.23%, and the semantic understanding system based on GRM and named entities achieves a 3.79% higher F₁ score than the model-based system, with an average F₁ score of 89.6% and an intent recognition accuracy exceeding 90%. This system offers higher business service accuracy and stability, effectively adapting to scenarios such as power consultation and fault reporting, meeting users′ requirements for real-time and accurate power information consultation.

关键词

文法规则匹配 / CRF算法 / 电力智能客服 / 命名实体识别 / 语义解析

Key words

grammar rule matching / CRF algorithm / power intelligent customer service / named entity recognition / semantic analysis

[1] 张晓慧, 孙德艳, 马永波,等. 情绪识别技术在电力智能客服系统中的应用研究[J]. 电子器件, 2020, 43(5):1061-1065.
[2] 俞学豪, 赵子岩, 马应龙,等. 基于BR和GBDT的电力信息通信客服系统多标签文本分类[J]. 电力系统自动化, 2021, 45(11): 144-151.
[3] 嵇友浪, 朱君, 邹云峰,等. 面向在线客服系统的调度算法研究[J]. 计算机工程与科学, 2020, 42(12):2242-2251.
[4] 覃浩, 刘振华, 苏立伟. 基于多模态融合的电力智能客服系统情绪识别方法[J]. 自动化技术与应用, 2024, 43(4):169-172.
[5] Klaus P, Zaichkowsky J. AI voice bots: a services marketing research agenda[J]. Journal of Services Marketing, 2020, 34 (3):389-398.
[6] Bromuri S, Henkel A P, Iren D, et al. Using AI to predict service agent stress from emotion patterns in service interactions[J]. Journal of Service Management, 2021, 32(4):581-611.
[7] 孙哲涛, 于红, 宋奇书,等. 基于规则匹配与深度学习AbTransformer的渔业标准表格信息抽取方法[J]. 大连海洋大学学报, 2023, 38(1):140-148.
[8] 王一海. 基于深度学习的电子商务语音客服中的客户情感研究[J]. 电子器件, 2020, 43(5):1173-1179.
[9] 吴石松, 董召杰. 基于深度语义匹配模型的智能客服系统设计[J]. 自动化技术与应用, 2024, 43(7):176-180.
[10] Johnson D, Larochelle H, Tarlow D. Learning graph structure with a finite-state automaton layer[J]. Advances in Neural Information Processing Systems, 2020(33):3082-3093.
[11] Li Y, Li C, Li X, et al. A comprehensive review of Markov random field and conditional random field approaches in pathology image analysis[J]. Archives of Computational Methods in Engineering, 2022, 29(1):609-639.
[12] 隆峻, 神显豪, 丁小军,等. 量子遗传算法优化加权朴素贝叶斯复合语言文本分类[J]. 济南大学学报(自然科学版), 2022, 36(2):136-141.

基金资助

国网河南省电力项目(B217X022J006)