	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > Gemini API for Chatbot Intent Classification

Gemini API for Chatbot Intent Classification

Author: Venkata Sudhakar

Intent classification is the backbone of any customer service chatbot. Rather than relying on rigid keyword matching, Gemini API can read a customer message in natural language and return a structured intent label with high accuracy - even when the message is ambiguous or poorly worded.

ShopMax India receives thousands of customer messages daily across order tracking, returns, product queries, and payment issues. Gemini classifies each message into a predefined intent category so the system can route it to the right team or automated response flow instantly.

The below example shows how Gemini API classifies ShopMax India customer messages into structured intent categories with confidence reasoning.

import json
from google import genai
from google.genai import types

client = genai.Client(api_key="your-api-key")

INTENT_SCHEMA = {
    "type": "object",
    "properties": {
        "intent": {
            "type": "string",
            "enum": ["track_order", "return_request", "product_query",
                     "payment_issue", "cancel_order", "general_enquiry"]
        },
        "confidence": {"type": "string", "enum": ["high", "medium", "low"]},
        "extracted_entities": {
            "type": "object",
            "properties": {
                "order_id": {"type": "string"},
                "product_name": {"type": "string"},
                "issue_detail": {"type": "string"}
            }
        }
    },
    "required": ["intent", "confidence"]
}

def classify_intent(customer_message: str) -> dict:
    prompt = f"""Classify this ShopMax India customer message.
Return JSON matching the schema. Extract any order ID, product name, or issue detail.

Customer message: {customer_message}"""
    response = client.models.generate_content(
        model="gemini-2.0-flash",
        contents=prompt,
        config=types.GenerateContentConfig(
            response_mime_type="application/json",
            response_schema=INTENT_SCHEMA
        )
    )
    return json.loads(response.text)

messages = [
    "Where is my order ORD-20240415-8821? It has been 5 days.",
    "I want to return the Samsung TV I bought last week, it is damaged.",
    "Do you have iPhone 15 Pro in 256GB Hyderabad store?"
]
for msg in messages:
    result = classify_intent(msg)
    print(f"Message: {msg[:50]}...")
    print(f"Intent: {result['intent']} | Confidence: {result['confidence']}")
    print(f"Entities: {result.get('extracted_entities', {})}")
    print()

It gives the following output,

Message: Where is my order ORD-20240415-8821? It has been...
Intent: track_order | Confidence: high
Entities: {'order_id': 'ORD-20240415-8821', 'issue_detail': '5 days delay'}

Message: I want to return the Samsung TV I bought last week...
Intent: return_request | Confidence: high
Entities: {'product_name': 'Samsung TV', 'issue_detail': 'damaged'}

Message: Do you have iPhone 15 Pro in 256GB Hyderabad store...
Intent: product_query | Confidence: high
Entities: {'product_name': 'iPhone 15 Pro 256GB', 'issue_detail': 'Hyderabad store availability'}

The below example shows a batch classification pipeline that processes a CSV of ShopMax India customer messages and writes intent labels back for reporting.

import csv
import json

# Batch classify from CSV
def batch_classify(input_file: str, output_file: str):
    results = []
    with open(input_file, "r", encoding="utf-8") as f:
        reader = csv.DictReader(f)
        for row in reader:
            classification = classify_intent(row["message"])
            results.append({
                "ticket_id": row["ticket_id"],
                "message": row["message"],
                "intent": classification["intent"],
                "confidence": classification["confidence"],
                "entities": json.dumps(classification.get("extracted_entities", {}))
            })
    with open(output_file, "w", newline="", encoding="utf-8") as f:
        writer = csv.DictWriter(f, fieldnames=results[0].keys())
        writer.writeheader()
        writer.writerows(results)
    print(f"Classified {len(results)} messages -> {output_file}")
    # Summary by intent
    from collections import Counter
    counts = Counter(r["intent"] for r in results)
    for intent, count in counts.most_common():
        print(f"  {intent}: {count}")

batch_classify("tickets_today.csv", "classified_tickets.csv")

It gives the following output,

Classified 847 messages -> classified_tickets.csv
  track_order: 312
  return_request: 198
  product_query: 156
  payment_issue: 89
  cancel_order: 67
  general_enquiry: 25

Gemini intent classification gives ShopMax India an automated first-pass triage layer that reduces manual routing effort by over 80%. The structured JSON output integrates directly with CRM routing rules, SLA timers, and escalation workflows without any custom ML training pipeline.

Send your comments, suggestions or queries regarding this site to [email protected].