SMM4H-2024 Task 2 Japanese RE

Overview

This is a relation extraction model created by fine-tuning daisaku-s/medtxt_ner_roberta on SMM4H 2024 Task 2b corpus.

Tag set:

CAUSED
TREATMENT_FOR

Usage

from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

text = "サンプルテキスト"
model_name = "yseop/SMM4H2024_Task2b_ja"
id2label = ['O', 'CAUSED', 'TREATMENT_FOR']

with torch.inference_mode():
    model = AutoModelForSequenceClassification.from_pretrained(model_name).eval()
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    encoded_input = tokenizer(text, return_tensors='pt', max_length=512)
    output = re_model(**encoded_input).logits
    class_id = output.argmax().item()
    print(id2label[class_id])

Results

Relation	tp	fp	fn	precision	recall	f1
CAUSED\|DISORDER\|DISORDER	1	163	38	0.0061	0.0256	0.0099
CAUSED\|DISORDER\|FUNCTION	0	70	13	0	0	0
CAUSED\|DRUG\|DISORDER	9	196	105	0.0439	0.0789	0.0564
CAUSED\|DRUG\|FUNCTION	2	59	7	0.0328	0.2222	0.0571
TREATMENT_FOR\|DISORDER\|DISORDER	0	12	0	0	0	0
TREATMENT_FOR\|DISORDER\|FUNCTION	0	3	0	0	0	0
TREATMENT_FOR\|DRUG\|DISORDER	0	15	91	0	0	0
TREATMENT_FOR\|DRUG\|FUNCTION	0	0	1	0	0	0
all	12	518	255	0.0226	0.0449	0.0301