MsgSense SMS Classifier

This model predicts a composite label in the format:

<score>_<sms_type_id>

• score (first digit): message importance used by app policy.
• sms_type_id (second part): category id from SmsClassificationTypeEntity mapping.

Input Format

Sender: <clean_address> | Message: <normalized body>

Output Decoding

Example output label: 3_37

• score = 3
• sms_type_id = 37 (SERVICE_NOTIFICATION)

Usage

from transformers import pipeline

pipe = pipeline(
    "text-classification",
    model="imShub10/msgsense-sms-bert-base-cleanaddr-fulldata-20260424",
)

text = "Sender: DAIKIN | Message: Your service request is scheduled."
print(pipe(text))

BERT-base

• accuracy: 0.95339
• f1 (weighted): 0.95367
• f1_macro: 0.89484
• precision: 0.95624
• recall: 0.95339
• test_loss: 0.45452
• train_loss: 0.10400
• training_time_min: 70.2
• model_size_mb: 418.5