You can deploy this model and start earning money today!

jibinsajujoseph/
email-intent-classifier

Name: email-intent-classifier
Availability: InStock

safetensors

roberta

Views

No views yet

Email Intent Classifier

Fine-tuned distilroberta-base for SaaS customer-support email intent classification.

Labels

ID	Intent	Department
0	`login_issue`	Technical Support
1	`billing_refund`	Billing
2	`subscription_change`	Billing
3	`bug_report`	Engineering
4	`feature_request`	Product
5	`integration_api`	Developer Support
6	`performance_issue`	Technical Support
7	`security_concern`	Security

Model Comparison

Model	Accuracy	Macro F1
TF-IDF + Logistic Regression	1.0000	1.0000
DistilRoBERTa (fine-tuned)	0.9975	0.9975

Training

Base model : distilroberta-base
Dataset : 8,000 training / 1,000 validation / 1,000 test (synthetic)
Max seq len : 256
Max epochs : 4 (Early Stopping patience=2)
Batch size : 32
LR : 2e-05

Inference

python

from transformers import pipeline
import json

meta = json.load(open('label_metadata.json'))
clf  = pipeline('text-classification', model='./email-intent-classifier', top_k=3)

results = clf(email_text)
top     = results[0]   # highest-confidence prediction
intent  = top['label']
dept    = meta['intent_metadata'][intent]['department']
conf    = top['score']

jibinsajujoseph/email-intent-classifier