Table 5 Interpretability analysis of the model (where \(\alpha\) represents the scaling factor of LoRA).
From: Knowledge graph construction for intelligent cockpits based on large language models
Methods | Evaluation metrics | |||
|---|---|---|---|---|
BLEU4 | ROUGE-1 | ROUGE-2 | ROUGE-L | |
GLM-TripleGen (in-context learning) | 68.15 | 80.86 | 63.42 | 57.60 |
GLM-TripleGen (rank=4) | 92.42 | 95.70 | 95.76 | 92.41 |
GLM-TripleGen (rank=16) | 92.67 | 95.87 | 95.92 | 92.52 |
GLM-TripleGen (rank=32) | 93.29 | 96.07 | 93.13 | 93.04 |
GLM-TripleGen (\(\alpha\)= 8) | 93.17 | 96.49 | 93.49 | 92.86 |
GLM-TripleGen (\(\alpha\)= 32) | 92.72 | 96.32 | 93.58 | 93.34 |
GLM-TripleGen w/o CoT prompting | 60.87 | 72.15 | 58.64 | 53.27 |
GLM-TripleGen | 93.56 | 96.73 | 93.74 | 93.20 |
