Error-Aware Knowledge Distillation via Targeted Revision for Customer-Service Summarization

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Lee, Hee-Jin, Guo, Zhen, Jin, Luchao, Goudarzi, Morteza Moazami
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915820803719168
author Lee, Hee-Jin
Guo, Zhen
Jin, Luchao
Goudarzi, Morteza Moazami
author_facet Lee, Hee-Jin
Guo, Zhen
Jin, Luchao
Goudarzi, Morteza Moazami
contents We introduce an Analyze-Revise-Finetune (ARF) pipeline that enables smaller open-source language models (LLMs) to surpass substantially larger proprietary models in customer service summarization tasks. The pipeline first analyzes and categorizes common errors in summaries produced by a teacher model (GPT-3.5), then performs a targeted revision using a compact editor model (Llama 3.1 70B) to generate high-quality, refined training data. Fine-tuning smaller student models (e.g., Llama 3.1 8B, QWen3 4B) on this refined data resulted in superior summarization performance compared to GPT-3.5. The ARF pipeline improves cost efficiency and data privacy while maintaining competitive accuracy, illustrating a generalizable framework for enhancing open-source LLMs across diverse downstream applications.
format Preprint
id arxiv_https___arxiv_org_abs_2511_03005
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Error-Aware Knowledge Distillation via Targeted Revision for Customer-Service Summarization
Lee, Hee-Jin
Guo, Zhen
Jin, Luchao
Goudarzi, Morteza Moazami
Computation and Language
We introduce an Analyze-Revise-Finetune (ARF) pipeline that enables smaller open-source language models (LLMs) to surpass substantially larger proprietary models in customer service summarization tasks. The pipeline first analyzes and categorizes common errors in summaries produced by a teacher model (GPT-3.5), then performs a targeted revision using a compact editor model (Llama 3.1 70B) to generate high-quality, refined training data. Fine-tuning smaller student models (e.g., Llama 3.1 8B, QWen3 4B) on this refined data resulted in superior summarization performance compared to GPT-3.5. The ARF pipeline improves cost efficiency and data privacy while maintaining competitive accuracy, illustrating a generalizable framework for enhancing open-source LLMs across diverse downstream applications.
title Error-Aware Knowledge Distillation via Targeted Revision for Customer-Service Summarization
topic Computation and Language
url https://arxiv.org/abs/2511.03005