Table 4 The performance of LLMs under imbalanced length of doctor-patient conversations.

From: A Multimodal Depression Consultation Dataset of Speech and Text with HAMD-17 Assessments

Models

Imbalance

Training

Precision

Recall

F1

Qwen2.5-7B-Instruct

Large

No

0.427

0.430

0.429

Large

Yes

0.609

0.613

0.611

Small

No

0.387

0.388

0.387

Small

Yes

0.543

0.544

0.543

LLaMA3.1-8B-Instruct

Large

No

0.448

0.451

0.449

Large

Yes

0.584

0.588

0.586

Small

No

0.396

0.397

0.397

Small

Yes

0.571

0.571

0.571

  1. "Imbalance” represents the ratio of the length of the patient’s transcribed text to that of the doctor’s transcribed text. “Training” indicates whether to fine-tune the parameters of LLMs.
Back to article page