A large language model-enhanced few-shot learning algorithm for network complaint intent recognition

GU Ninglun; LUO Zhiyi; WANG Xidong; DING Jianbing; OUYANG Ye

doi:10.11959/j.issn.1000-0801.2025236

您当前的位置：

首页 >

文章列表页 >

A large language model-enhanced few-shot learning algorithm for network complaint intent recognition

Research and Development | 更新时间：2025-11-06

- A large language model-enhanced few-shot learning algorithm for network complaint intent recognition
- Telecommunications Science Vol. 41, Issue 10, Pages: 161-171(2025)
- 作者机构：
  
  1.中国移动通信集团有限公司，北京 100032
  2.亚信科技（中国）有限公司，北京 100193
- 作者简介：
- 基金信息：
- DOI：10.11959/j.issn.1000-0801.2025236
  CLC： TN911.7;TP18
- Received：02 July 2025，
  
  Revised：2025-09-20，
  
  Accepted：23 September 2025，
  
  Published：20 October 2025
- 稿件说明：
移动端阅览
顾宁伦,罗志毅,王希栋等.网络投诉意图识别：一种基于大模型增强的少样本学习方法[J].电信科学,2025,41(10):161-171.

GU Ninglun,LUO Zhiyi,WANG Xidong,et al.A large language model-enhanced few-shot learning algorithm for network complaint intent recognition[J].Telecommunications Science,2025,41(10):161-171.
顾宁伦,罗志毅,王希栋等.网络投诉意图识别：一种基于大模型增强的少样本学习方法[J].电信科学,2025,41(10):161-171. DOI： 10.11959/j.issn.1000-0801.2025236.

GU Ninglun,LUO Zhiyi,WANG Xidong,et al.A large language model-enhanced few-shot learning algorithm for network complaint intent recognition[J].Telecommunications Science,2025,41(10):161-171. DOI： 10.11959/j.issn.1000-0801.2025236.

摘要

针对5G时代日益复杂的网络架构与用户规模增长带来的网络投诉处理挑战，旨在解决实际投诉数据中普遍存在的少样本类别意图识别难题。传统小型自然语言处理（natural language processing，NLP）模型因对数据量和质量高度依赖，在少样本场景下泛化能力不足，识别效果欠佳。大语言模型（large language model，LLM）虽具备强大的通用语义理解能力，但受限于通信领域专业知识匮乏及巨大计算资源需求，难以直接高效应用于此特定场景。为此，提出一种基于大模型增强的少样本网络投诉意图识别算法。该算法创新性地融合了LLM的文本生成优势与小型模型的高效推理特性，通过LLM迭代生成针对少样本类别的高质量模拟训练样本，并引入小模型进行质量评估与反馈，持续优化样本生成过程。该机制显著提升了小模型对少样本投诉意图的识别准确率。基于实际网络投诉数据的测试结果表明，所提方案使3类少样本网络投诉的意图识别准确率提升了21%，并实现了全部8类投诉识别准确率9%的整体提升。

Abstract

Addressing the escalating challenge of network complaint handling in the 5G era

driven by increasingly complex network architectures and growing user bases

the aim is to solve the prevalent issue of few-shot intent recognition for complaint categories with scarce samples. Traditional small-parameter natural language processing (NLP) models exhibit insufficient generalization and suboptimal performance in such scenarios due to their heavy reliance on data quantity and quality. While large language models (LLM) possess strong general semantic understanding

their limited domain-specific knowledge in telecommunications and substantial computational demands hinder direct

efficient application to few-shot network complaint intent recognition. To precisely tackle this

an LLM-enhanced few-shot intent recognition algorithm for network complaints was proposed. This innovative algorithm integrated the powerful text generation capabilities of LLM with the efficient inference characteristics of small models. It leveraged LLM to iteratively generate high-quality simulated training samples for few-shot categories

incorporating a small model for quality evaluation and feedback to continuously optimize the sample generation process. This mechanism significantly boosted small model training and specifically enhanced the recognition accuracy of few-shot complaint intents. Test results on actual network complaint data demonstrated a 21% improvement in intent recognition accuracy for 3 types of few-shot network complaints

and a 9% overall improvement in recognition accuracy across all 8 complaint types.

关键词

Keywords

references

OUYANG Y , WANG L L , YANG A D , et al . Next decade of telecommunications artificial intelligence [J ] . CAAI Artificial Intelligence Research , 2022 , 1 ( 1 ): 28 - 53 .

曾伟 , 钟检荣 , 张玮 , 等 . 基于机器学习的5G用户智能投诉处理方案研究 [J ] . 邮电设计技术 , 2021 ( 5 ): 43 - 48 .

ZENG W , ZHONG J R , ZHANG W , et al . Research on intelligent complaint handling scheme of 5G users based on machine learning [J ] . Designing Techniques of Posts and Telecommunications , 2021 ( 5 ): 43 - 48 .

蒋燕 , 程浩辉 , 何丹 . 基于BERT算法的通信投诉智能处理探索 [J ] . 广东通信技术 , 2024 , 44 ( 2 ): 52 - 55 .

JIANG Y , LIANG H H , HE D . Exploring intelligent processing of communication complaints based on the BERT algorithm [J ] . Guangdong Communication Technology , 2024 , 44 ( 2 ): 52 - 55 .

亚信科技 , 清华大学智能产业研究院 . AIGC(GPT-4)赋能通信行业应用白皮书 [R ] . 2023 .

AsiaInfo, Institute for AI Industry Research, Tsinghua University . A white paper of AIGC (GPT-4) empowering telecom sector [R ] . 2023 .

FINN C , ABBEEL P , LEVINE S . Model-agnostic meta-learning for fast adaptation of deep networks [J ] . arXiv preprint , 2017 : 1703 .03400.

ZHANG H X , ZHANG X F , HUANG H B , et al . Prompt-based meta-learning for few-shot text classification [C ] // Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing . Stroudsburg, PA, USA : ACL , 2022 : 1342 - 1357 .

EDWARDS A , USHIO A , JOSÉ CAMACHO-COLLADOS , et al . Guiding generative language models for data augmentation in few-shot text classification [J ] . arXiv preprint , 2021 : 2111 .09064.

BAYER M , KAUFHOLD M A , REUTER C . A survey on data augmentation for text classification [J ] . ACM Computing Surveys , 2023 , 55 ( 7 ): 1 - 39 .

GAO J , LYU S Q , LIU G R . A hybrid model for few-shot text classification using transfer and meta-learning [J ] . arXiv preprint , 2025 : 2502 .09086.

BRAGG J , COHAN A , LO K , et al . FLEX: unifying evaluation for few-shot NLP [J ] . arXiv preprint , 2021 : 2107 .07170.

AGARWAL R , VIEILLARD N , ZHOU Y , et al . On-policy distillation of language models: learning from self-generated mistakes [J ] . arXiv preprint , 2024 : 2306 .13649.

LIANG C , ZUO S M , ZHANG Q R , et al . Less is more: task-aware layer-wise distillation for language model compression [C ] // Proceedings of the 40th International Conference on Machine Learning . Hawaii : ICML , 2023 : 20852 - 20867 .

ZOU T , LIU Y , LI P , et al . Fusegen: PLM fusion for data-generation based zero-shot learning [J ] . arXiv preprint , 2024 : 2406 .12527.

YAO Z W , YAZDANI AMINABADI R , ZHANG M J , et al . Zeroquant: efficient and affordable posttraining quantization for large-scale transformers [J ] . Advances in Neural Information Processing Systems , 2022 ( 35 ): 27168 - 27183 .

CHENG Y , ZHANG L , LI A . GFL: federated learning on non-IID data via privacy-preserving synthetic data [C ] // Proceedings of 2023 IEEE International Conference on Pervasive Computing and Communications (PerCom) . Piscataway : IEEE Press , 2023 : 61 - 70 .

ZHANG Z , YANG Y H , DAI Y , et al . FedPETuning: when federated learning meets the parameter-efficient tuning methods of pre-trained language models [C ] // Findings of the Association for Computational Linguistics (ACL) , 2023 : 9963 - 9977 .

Views

149

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Research on customer escalated complaints prediction based on LLM-RFE-XGBoost approach

Automated extraction and evaluation of business process entities and relations via pre-trained models

Research and application of intelligent operation and maintenance human-computer interaction system based on large language models for communication networks

Large language model for cloud-network configuration audit and its application in IP network

Multi-hop question answering by integrating large language models and knowledge graphs

Related Author

Jiang Jun

Xu Yu

Liu Fang

Wang Qian

Qiao Yaqian

Li Shan

Wan Xiang

Liu Zesan

Related Institution

China Mobile Online Service Co., Ltd., AI Capability Center

China Mobile Online Service Co., Ltd., Shandong Branch

Electric Power Science Research Institute, State Grid Liaoning Electric Power Co., Ltd.

State Grid Information and Communication Industry Group Corporation

Aostar Information Technologies Co., Ltd.

⁰