DistilBERT SNIPS Intent Router

A fine‑tuned distilbert-base-uncased model that classifies short user utterances into one of 7 customer‑support intents.

Model Details

Model Description

This model was fine‑tuned on the SNIPS built‑in intents dataset for single‑label text classification. It takes a user query (e.g. “Book me a table for tonight”) and returns one of the predefined intents:

AddToPlaylist
BookRestaurant
GetWeather
PlayMusic
RateBook
SearchCreativeWork
SearchScreeningEvent

Attribute	Value
Developed by	Goutham
Model type	DistilBERT (sequence classification)
Language(s)	English
License	apache-2.0
Fine‑tuned from	`distilbert-base-uncased`
Dataset	SNIPS built‑in intents

Uses

Direct Use

Route user requests in chatbots, voice assistants, or email triage systems into support categories for faster handling.

Out‑of‑Scope Use

Long-form or multi‑sentence inputs; performance may degrade on utterances beyond ~20 words.
Languages other than English.

Bias, Risks, and Limitations

Bias: Trained only on clear, synthetic voice‑assistant style utterances. May misclassify non‑standard phrasing or dialects.
Risks: Misrouting critical user requests (e.g. emergency queries) if phrased unusually.
Limitations:
- Accuracy degrades on very short (“Hi”) or very long (“I’d like to…”) utterances.
- No support for multi‑intent or slot filling.

How to Get Started

from transformers import pipeline

intent_router = pipeline(
    "text-classification",
    model="YOUR_USERNAME/snips-intent-router",
    tokenizer="YOUR_USERNAME/snips-intent-router",
)

# Example
result = intent_router("Book me a table for two at an Italian restaurant tonight")
print(result)
# → [{'label':'BookRestaurant','score':0.99}]