Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
IEICE Technical Committee Submission System
Conference Paper's Information		 Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2025-01-24 10:40
The Llama 3 fine-tuning model for depression detection from conversation
Kenyu Ikeuchi, Taishiro Kishimoto, Fumiya Nakai, Mondher Bouazizi, Taichi Okunishi, Chuheng Zheng, Momoko Kitazawa, Toshiro Horigome, Tomoaki Ohtsuki (Keio Univ.) SeMI2024-77
Abstract (in Japanese) (See Japanese page) 
(in English) Depression has a significant impact on society. Early detection is essential because early treatment can mitigate its damages. Models that use state-of-the-art large language models, such as Llama 3, and employ prompt engineering have been reported to detect depression with high accuracy. On the other hand, while fine-tuning techniques offer shorter inference times, they require substantial memory for training and have not been reported to achieve high detection accuracy in classification or regression models.

This study aims to detect depression more rapidly with accuracy comparable to models that use prompt engineering, using fine-tuning techniques. First, the parameters of the large language model were quantized to reduce the memory required for training. Instead of introducing classification or regression layers, the original structure of the large language model was preserved and trained as a generative model.

Validation using the DAIC-WOZ dataset resulted in an F1 score of 84%, achieving depression detection accuracy similar to models using prompt engineering, while enabling faster inference. Additionally, when evaluated using the private dataset PROMPT, the model achieved an F1 score of 82%, demonstrating that its high accuracy in detecting depression is not solely reliant on open datasets that might have been pre-trained.
Keyword (in Japanese) (See Japanese page) 
(in English) Depression detection / Large Language Model / Fine-tuning / Quantization / / / /  
Reference Info. IEICE Tech. Rep., vol. 124, no. 353, SeMI2024-77, pp. 88-93, Jan. 2025.
Paper # SeMI2024-77 
Date of Issue 2025-01-16 (SeMI) 
ISSN Online edition: ISSN 2432-6380
Copyright
and
reproduction		 All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034)
Download PDF SeMI2024-77

Conference Information
Committee SeMI  
Conference Date 2025-01-23 - 2025-01-24 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To SeMI 
Conference Code 2025-01-SeMI 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) The Llama 3 fine-tuning model for depression detection from conversation 
Sub Title (in English)  
Keyword(1) Depression detection  
Keyword(2) Large Language Model  
Keyword(3) Fine-tuning  
Keyword(4) Quantization  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Kenyu Ikeuchi  
1st Author's Affiliation Keio University (Keio Univ.)
2nd Author's Name Taishiro Kishimoto  
2nd Author's Affiliation Keio University (Keio Univ.)
3rd Author's Name Fumiya Nakai  
3rd Author's Affiliation Keio University (Keio Univ.)
4th Author's Name Mondher Bouazizi  
4th Author's Affiliation Keio University (Keio Univ.)
5th Author's Name Taichi Okunishi  
5th Author's Affiliation Keio University (Keio Univ.)
6th Author's Name Chuheng Zheng  
6th Author's Affiliation Keio University (Keio Univ.)
7th Author's Name Momoko Kitazawa  
7th Author's Affiliation Keio University (Keio Univ.)
8th Author's Name Toshiro Horigome  
8th Author's Affiliation Keio University (Keio Univ.)
9th Author's Name Tomoaki Ohtsuki  
9th Author's Affiliation Keio University (Keio Univ.)
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2025-01-24 10:40:00 
Presentation Time 20 minutes 
Registration for SeMI 
Paper # SeMI2024-77 
Volume (vol) vol.124 
Number (no) no.353 
Page pp.88-93 
#Pages 6 
Date of Issue 2025-01-16 (SeMI) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan