@inproceedings{zhang-etal-2026-domain,
title = "Domain Knowledge Annealing: A Structured Local-to-Global Paradigm for Constructing Expert {LLM}s",
author = "Zhang, Yikang and
Li, Zuchao and
Wang, Ping and
Zhang, Lefei and
Zhao, Hai and
Du, Bo",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.121/",
pages = "2665--2682",
ISBN = "979-8-89176-439-2",
abstract = "Expertise in vertical domains requires more than exposing LLMs to vast corpora; it demands the systematic learning of complex logic. Standard fine-tuning often treats data as an unstructured collection, failing to capture the dependencies needed for high-level reasoning. We propose Domain Knowledge Annealing (DKA), a curriculum learning paradigm designed to improve domain proficiency. Inspired by physical annealing, DKA organizes training from local to global: it starts by injecting discrete concepts via entity-centric samples and progressively advances to synthesizing complex, inter-document relationships. This progression allows the model to integrate isolated facts into a coherent knowledge system. Experiments on Linguistics and Law benchmarks show that DKA outperforms standard strategies, achieving state-of-the-art results among comparably sized models in domain-specific reasoning."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-domain">
<titleInfo>
<title>Domain Knowledge Annealing: A Structured Local-to-Global Paradigm for Constructing Expert LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yikang</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zuchao</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ping</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lefei</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hai</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bo</namePart>
<namePart type="family">Du</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>Expertise in vertical domains requires more than exposing LLMs to vast corpora; it demands the systematic learning of complex logic. Standard fine-tuning often treats data as an unstructured collection, failing to capture the dependencies needed for high-level reasoning. We propose Domain Knowledge Annealing (DKA), a curriculum learning paradigm designed to improve domain proficiency. Inspired by physical annealing, DKA organizes training from local to global: it starts by injecting discrete concepts via entity-centric samples and progressively advances to synthesizing complex, inter-document relationships. This progression allows the model to integrate isolated facts into a coherent knowledge system. Experiments on Linguistics and Law benchmarks show that DKA outperforms standard strategies, achieving state-of-the-art results among comparably sized models in domain-specific reasoning.</abstract>
<identifier type="citekey">zhang-etal-2026-domain</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.121/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>2665</start>
<end>2682</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Domain Knowledge Annealing: A Structured Local-to-Global Paradigm for Constructing Expert LLMs
%A Zhang, Yikang
%A Li, Zuchao
%A Wang, Ping
%A Zhang, Lefei
%A Zhao, Hai
%A Du, Bo
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F zhang-etal-2026-domain
%X Expertise in vertical domains requires more than exposing LLMs to vast corpora; it demands the systematic learning of complex logic. Standard fine-tuning often treats data as an unstructured collection, failing to capture the dependencies needed for high-level reasoning. We propose Domain Knowledge Annealing (DKA), a curriculum learning paradigm designed to improve domain proficiency. Inspired by physical annealing, DKA organizes training from local to global: it starts by injecting discrete concepts via entity-centric samples and progressively advances to synthesizing complex, inter-document relationships. This progression allows the model to integrate isolated facts into a coherent knowledge system. Experiments on Linguistics and Law benchmarks show that DKA outperforms standard strategies, achieving state-of-the-art results among comparably sized models in domain-specific reasoning.
%U https://aclanthology.org/2026.aacl-long.121/
%P 2665-2682
Markdown (Informal)
[Domain Knowledge Annealing: A Structured Local-to-Global Paradigm for Constructing Expert LLMs](https://aclanthology.org/2026.aacl-long.121/) (Zhang et al., AACL-IJCNLP 2026)
ACL