@inproceedings{haralovic-etal-2026-crisiskd,
title = "{C}risis{KD}: Five-Stage Knowledge Distillation for Aspect-Level Sentiment and Emotion Analysis in Crisis Discourse",
author = {Haralovic, Marko and
Akca, Onat Dogan and
Y{\"u}cet{\"u}rk, Salih Eren and
Li, Minsi and
Theune, Mari{\"e}t},
editor = "Bak, JinYeong and
van der Goot, Rob and
Jang, Hyeju and
Buaphet, Weerayut and
Ramponi, Alan and
Xu, Wei and
Ritter, Alan",
booktitle = "Proceedings of the 11th Workshop on Natural User-generated Text",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wnut-1.4/",
pages = "49--64",
ISBN = "979-8-89176-470-5",
abstract = "Identifying the targets of emotional words and phrases in crisis situations, particularly health-related crises, is essential for understanding public concerns across cultural and linguistic contexts. We propose CrisisKD, a five-stage teacher{--}student knowledge distillation framework for aspect-level sentiment and emotion analysis on unannotated social media data. A teacher large language model generates aspect-level labels and reasoning traces to supervise a smaller student model across five tasks: aspect extraction, syntactic parsing, opinion extraction, sentiment classification, and emotion classification. Using this framework, we construct and release a dataset containing 50,615 aspect-level annotations, along with the annotation pipeline and fine-tuned student models as open-source resources. The resulting student model supports end-to-end aspect-based sentiment analysis and emotion detection at substantially lower inference cost than the teacher, making the approach suitable for practitioners with limited access to annotated data and large-scale computational resources. Because CrisisKD requires only raw text and a dependency parser, the framework is architecturally portable to other crisis domains and to languages supported by suitable multilingual parsers."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="haralovic-etal-2026-crisiskd">
<titleInfo>
<title>CrisisKD: Five-Stage Knowledge Distillation for Aspect-Level Sentiment and Emotion Analysis in Crisis Discourse</title>
</titleInfo>
<name type="personal">
<namePart type="given">Marko</namePart>
<namePart type="family">Haralovic</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Onat</namePart>
<namePart type="given">Dogan</namePart>
<namePart type="family">Akca</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Salih</namePart>
<namePart type="given">Eren</namePart>
<namePart type="family">Yücetürk</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minsi</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mariët</namePart>
<namePart type="family">Theune</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 11th Workshop on Natural User-generated Text</title>
</titleInfo>
<name type="personal">
<namePart type="given">JinYeong</namePart>
<namePart type="family">Bak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rob</namePart>
<namePart type="family">van der Goot</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hyeju</namePart>
<namePart type="family">Jang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weerayut</namePart>
<namePart type="family">Buaphet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alan</namePart>
<namePart type="family">Ramponi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alan</namePart>
<namePart type="family">Ritter</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-470-5</identifier>
</relatedItem>
<abstract>Identifying the targets of emotional words and phrases in crisis situations, particularly health-related crises, is essential for understanding public concerns across cultural and linguistic contexts. We propose CrisisKD, a five-stage teacher–student knowledge distillation framework for aspect-level sentiment and emotion analysis on unannotated social media data. A teacher large language model generates aspect-level labels and reasoning traces to supervise a smaller student model across five tasks: aspect extraction, syntactic parsing, opinion extraction, sentiment classification, and emotion classification. Using this framework, we construct and release a dataset containing 50,615 aspect-level annotations, along with the annotation pipeline and fine-tuned student models as open-source resources. The resulting student model supports end-to-end aspect-based sentiment analysis and emotion detection at substantially lower inference cost than the teacher, making the approach suitable for practitioners with limited access to annotated data and large-scale computational resources. Because CrisisKD requires only raw text and a dependency parser, the framework is architecturally portable to other crisis domains and to languages supported by suitable multilingual parsers.</abstract>
<identifier type="citekey">haralovic-etal-2026-crisiskd</identifier>
<location>
<url>https://aclanthology.org/2026.wnut-1.4/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>49</start>
<end>64</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T CrisisKD: Five-Stage Knowledge Distillation for Aspect-Level Sentiment and Emotion Analysis in Crisis Discourse
%A Haralovic, Marko
%A Akca, Onat Dogan
%A Yücetürk, Salih Eren
%A Li, Minsi
%A Theune, Mariët
%Y Bak, JinYeong
%Y van der Goot, Rob
%Y Jang, Hyeju
%Y Buaphet, Weerayut
%Y Ramponi, Alan
%Y Xu, Wei
%Y Ritter, Alan
%S Proceedings of the 11th Workshop on Natural User-generated Text
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-470-5
%F haralovic-etal-2026-crisiskd
%X Identifying the targets of emotional words and phrases in crisis situations, particularly health-related crises, is essential for understanding public concerns across cultural and linguistic contexts. We propose CrisisKD, a five-stage teacher–student knowledge distillation framework for aspect-level sentiment and emotion analysis on unannotated social media data. A teacher large language model generates aspect-level labels and reasoning traces to supervise a smaller student model across five tasks: aspect extraction, syntactic parsing, opinion extraction, sentiment classification, and emotion classification. Using this framework, we construct and release a dataset containing 50,615 aspect-level annotations, along with the annotation pipeline and fine-tuned student models as open-source resources. The resulting student model supports end-to-end aspect-based sentiment analysis and emotion detection at substantially lower inference cost than the teacher, making the approach suitable for practitioners with limited access to annotated data and large-scale computational resources. Because CrisisKD requires only raw text and a dependency parser, the framework is architecturally portable to other crisis domains and to languages supported by suitable multilingual parsers.
%U https://aclanthology.org/2026.wnut-1.4/
%P 49-64
Markdown (Informal)
[CrisisKD: Five-Stage Knowledge Distillation for Aspect-Level Sentiment and Emotion Analysis in Crisis Discourse](https://aclanthology.org/2026.wnut-1.4/) (Haralovic et al., WNUT 2026)
ACL