SOCIOCOM at the NTCIR-18 RadNLP Main task: Zero-Shot LLM Approaches for Lung Cancer Staging

Key Points

• The aim is to classify lung cancer stages from radiology reports using advanced language models.
• Utilized GPT-4o model for inference.
• Employed prompt engineering techniques to enhance model performance.
• Participated in the NTCIR-18 RadNLP 2024 main task for evaluation.
• Achieved an accuracy of 0.5648 on Japanese test data.
• Demonstrated robustness of the closed-source models in classification tasks.