@inproceedings{rizwan-demsar-2026-resource,
title = "Resource-Efficient {LLM}s for Depression Symptoms Screening: Performance and Limitations in Zero Shot Setting",
author = "Rizwan, Muhammad and
Dem{\v{s}}ar, Jure",
editor = {Kokkinakis, Dimitrios and
Themistocleous, Charalambos and
Dias, Ga{\"e}l and
Fraser, Kathleen C. and
{\"O}hman, Fredrik and
Pais, Sebasti{\~a}o},
booktitle = "Proceedings of the Sixth Resources and {P}rocess{I}ng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments in cooperation with the {MENTAL}.ai consortium",
month = may,
year = "2026",
address = "Palma, Mallorca, Spain",
publisher = "European Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.rapid-1.7/",
doi = "10.63317/5fkbms3rp4bo",
pages = "74--81",
abstract = "Depression is the leading cause of global disability and early detection is crucial for effective intervention. Recent advances in large language models (LLMs) offer potential for analyzing text to identify depression symptoms. This work investigates the zero-shot capability of LLMs to recognize nine DSM5 depression symptoms from short-text inputs. We evaluated eight open LLMs with model sizes ranging from 1.5B to 14B parameters using a clinically annotated dataset and assessed both overall agreement and symptom-level performance. Results indicate that while smaller models exhibit limited clinical accuracy, the Qwen 2.5-7B model achieves substantial performance with a Cohen{'}s Kappa of 0.603 and a Macro F1 score of 0.648. Notably, a performance plateau between the 7B and 14B Qwen variants suggests that model scaling alone does not guarantee improved symptom-level classification, establishing Qwen 2.5-7B as a resource-efficient model. Further analysis of the best-performing model revealed strengths in identifying salient symptoms like suicidal thoughts, but limitations in recognizing core symptoms such as depressed mood and anhedonia. Misclassification analysis reveals that the model frequently misclassifies posts expressing `depressed mood' as `no symptom' or vice versa, often overlooking indicators of irritability or social withdrawal. These findings suggest that resource-efficient LLMs can support preliminary symptom screening in zero shot settings, but there is risk of overlooking clinically important symptoms without fine-tuning."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rizwan-demsar-2026-resource">
<titleInfo>
<title>Resource-Efficient LLMs for Depression Symptoms Screening: Performance and Limitations in Zero Shot Setting</title>
</titleInfo>
<name type="personal">
<namePart type="given">Muhammad</namePart>
<namePart type="family">Rizwan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jure</namePart>
<namePart type="family">Demšar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Sixth Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments in cooperation with the MENTAL.ai consortium</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dimitrios</namePart>
<namePart type="family">Kokkinakis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Charalambos</namePart>
<namePart type="family">Themistocleous</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gaël</namePart>
<namePart type="family">Dias</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kathleen</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Fraser</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fredrik</namePart>
<namePart type="family">Öhman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sebastião</namePart>
<namePart type="family">Pais</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>European Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca, Spain</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Depression is the leading cause of global disability and early detection is crucial for effective intervention. Recent advances in large language models (LLMs) offer potential for analyzing text to identify depression symptoms. This work investigates the zero-shot capability of LLMs to recognize nine DSM5 depression symptoms from short-text inputs. We evaluated eight open LLMs with model sizes ranging from 1.5B to 14B parameters using a clinically annotated dataset and assessed both overall agreement and symptom-level performance. Results indicate that while smaller models exhibit limited clinical accuracy, the Qwen 2.5-7B model achieves substantial performance with a Cohen’s Kappa of 0.603 and a Macro F1 score of 0.648. Notably, a performance plateau between the 7B and 14B Qwen variants suggests that model scaling alone does not guarantee improved symptom-level classification, establishing Qwen 2.5-7B as a resource-efficient model. Further analysis of the best-performing model revealed strengths in identifying salient symptoms like suicidal thoughts, but limitations in recognizing core symptoms such as depressed mood and anhedonia. Misclassification analysis reveals that the model frequently misclassifies posts expressing ‘depressed mood’ as ‘no symptom’ or vice versa, often overlooking indicators of irritability or social withdrawal. These findings suggest that resource-efficient LLMs can support preliminary symptom screening in zero shot settings, but there is risk of overlooking clinically important symptoms without fine-tuning.</abstract>
<identifier type="citekey">rizwan-demsar-2026-resource</identifier>
<identifier type="doi">10.63317/5fkbms3rp4bo</identifier>
<location>
<url>https://aclanthology.org/2026.rapid-1.7/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>74</start>
<end>81</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Resource-Efficient LLMs for Depression Symptoms Screening: Performance and Limitations in Zero Shot Setting
%A Rizwan, Muhammad
%A Demšar, Jure
%Y Kokkinakis, Dimitrios
%Y Themistocleous, Charalambos
%Y Dias, Gaël
%Y Fraser, Kathleen C.
%Y Öhman, Fredrik
%Y Pais, Sebastião
%S Proceedings of the Sixth Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments in cooperation with the MENTAL.ai consortium
%D 2026
%8 May
%I European Language Resources Association (ELRA)
%C Palma, Mallorca, Spain
%F rizwan-demsar-2026-resource
%X Depression is the leading cause of global disability and early detection is crucial for effective intervention. Recent advances in large language models (LLMs) offer potential for analyzing text to identify depression symptoms. This work investigates the zero-shot capability of LLMs to recognize nine DSM5 depression symptoms from short-text inputs. We evaluated eight open LLMs with model sizes ranging from 1.5B to 14B parameters using a clinically annotated dataset and assessed both overall agreement and symptom-level performance. Results indicate that while smaller models exhibit limited clinical accuracy, the Qwen 2.5-7B model achieves substantial performance with a Cohen’s Kappa of 0.603 and a Macro F1 score of 0.648. Notably, a performance plateau between the 7B and 14B Qwen variants suggests that model scaling alone does not guarantee improved symptom-level classification, establishing Qwen 2.5-7B as a resource-efficient model. Further analysis of the best-performing model revealed strengths in identifying salient symptoms like suicidal thoughts, but limitations in recognizing core symptoms such as depressed mood and anhedonia. Misclassification analysis reveals that the model frequently misclassifies posts expressing ‘depressed mood’ as ‘no symptom’ or vice versa, often overlooking indicators of irritability or social withdrawal. These findings suggest that resource-efficient LLMs can support preliminary symptom screening in zero shot settings, but there is risk of overlooking clinically important symptoms without fine-tuning.
%R 10.63317/5fkbms3rp4bo
%U https://aclanthology.org/2026.rapid-1.7/
%U https://doi.org/10.63317/5fkbms3rp4bo
%P 74-81
Markdown (Informal)
[Resource-Efficient LLMs for Depression Symptoms Screening: Performance and Limitations in Zero Shot Setting](https://aclanthology.org/2026.rapid-1.7/) (Rizwan & Demšar, RaPID 2026)
ACL