Table 3 MISTIC pipeline results compared to the other state-of-the-art approaches

Model

P

R

F1

Acc

Rule-based system

Pattern-matching

0.961

0.710

0.816

0.777

Zero-shot learning

mDeBERTa-v3-base-MNLI-XNLI

0.746

0.810

0.776

0.673

mDeBERTa-v3-base-tasksource

0.712

0.848

0.774

0.653

Comprehend-IT

0.702

1.000

0.825

0.703

XLM-RoBERTa-large-IT-MNLI

0.737

0.814

0.774

0.667

Structured generation

Llama 3.2 3B

0.706

0.595

0.645

0.543

Minerva 3B

0.723

0.648

0.683

0.580

Mixtral 7B

0.720

0.562

0.631

0.540

Few-shot learning

sentence-bert-base-italian-uncased

0.873

0.952

0.911

0.870

MISTIC

0.851

0.981

0.912

0.867