Digitala Vetenskapliga Arkivet

Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Exploring Methods to improve Context-Aware text Extraction and Categorization of a Large Language Model
Jönköping University, School of Engineering, JTH, Department of Computer Science and Informatics.
Jönköping University, School of Engineering, JTH, Department of Computer Science and Informatics.
2025 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Abstract [en]

Debriefing training exercises and high-stakes operations often involves analyzing radiocommunicationto better understand interactions and decision-making. This research investigates methods forextracting and categorizing sender-recipient information from transcribed air traffic control (ATC) communication,aiming to support more effective analysis of such data. A quantitative experimental approachwas used to compare the accuracy of zero-shot and few-shot prompting techniques, as well as their finetunedcounterparts using Low-Rank Adaptation (LoRA).Zero-shot prompting provided a strong baseline for evaluating the effectiveness of few-shot andfine-tuned approaches. The few-shot method generated unexpectedly strong results, and its combinationwith a LoRA fine-tuned model achieved the highest accuracy. These improvements highlight how simpleadjustments can enhance a model’s ability to understand domain-specific communication.The findings suggest that these methods could benefit other high-stakes domains, such as militaryoperations, emergency response, and other uniformed services, by supporting faster and more accurateinterpretation of critical communications during both training and real-world events. The results aresupported by Fisher’s exact test, performed to determine statistical significance.

Place, publisher, year, edition, pages
2025. , p. 48
Keywords [en]
Large Language Models, Zero-Shot Prompting, Few-Shot Prompting, LoRA, Categorizing Radiocommunication, Gemma 2 9b it
National Category
Natural Language Processing
Identifiers
URN: urn:nbn:se:hj:diva-69417OAI: oai:DiVA.org:hj-69417DiVA, id: diva2:1985379
External cooperation
Saab AB
Subject / course
JTH, Computer Engineering
Supervisors
Examiners
Available from: 2025-08-04 Created: 2025-07-24 Last updated: 2025-10-13Bibliographically approved

Open Access in DiVA

fulltext(9116 kB)177 downloads
File information
File name FULLTEXT01.pdfFile size 9116 kBChecksum SHA-512
0ad82103ba07aebcd567822fe41ec64b65884b2971171e4cb45d9ab72edd1cbc0e95c4e688004236ebee7e376fb50359379683070d53a00a6871291b9bce07b3
Type fulltextMimetype application/pdf

By organisation
JTH, Department of Computer Science and Informatics
Natural Language Processing

Search outside of DiVA

GoogleGoogle Scholar
Total: 181 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 790 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf