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

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Anthropic Claude API > Claude Text Classification and Sentiment Analysis

Claude Text Classification and Sentiment Analysis

Author: Venkata Sudhakar

Text classification and sentiment analysis are among the highest-ROI applications of Claude for e-commerce. For ShopMax India, automatically classifying 5,000 daily customer reviews as positive, negative, or neutral - and tagging them by topic (delivery, product quality, price, service) - enables the product team to spot quality issues within hours of delivery rather than waiting for weekly manual review cycles. Claude outperforms traditional sentiment models on nuanced Indian English, including mixed Hindi-English (Hinglish) expressions common in customer reviews.

Classification tasks work best with Claude when you provide explicit label definitions and examples in the prompt. For multi-label classification (a review can be about both delivery and product quality), ask Claude to return all applicable labels rather than just one. Using Haiku for classification is cost-effective since these tasks need fast inference at high volume rather than deep reasoning. Returning confidence scores alongside labels helps route borderline cases for human review.

The following example shows ShopMax India classifying customer reviews by sentiment and topic, with confidence scoring and priority flagging for negative reviews needing urgent attention:

import anthropic
import json
import os

client = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))

REVIEWS = [
    {"id": "R001", "text": "Samsung TV delivered to Mumbai on time but remote was missing from box"},
    {"id": "R002", "text": "Picture quality is outstanding! Very happy with my purchase. 5 stars"},
    {"id": "R003", "text": "Worst experience. AC stopped working after 2 days. Nobody picking up support calls"},
    {"id": "R004", "text": "Decent product but delivery took 8 days instead of promised 3 days in Bangalore"},
    {"id": "R005", "text": "Ekdum best washing machine! Hard water bhi saaf kar deta hai. Highly recommend"},
]

def classify_review(review_text: str) -> dict:
    prompt = (
        "Classify this ShopMax India customer review. Return JSON with: "
        "sentiment (POSITIVE/NEGATIVE/NEUTRAL), "
        "topics (list from: DELIVERY, PRODUCT_QUALITY, PRICE, CUSTOMER_SERVICE, PACKAGING), "
        "confidence (0.0-1.0), "
        "urgent (true if immediate action needed, e.g. defective product or safety issue). "
        "Return only JSON.\n\nReview: " + review_text
    )
    response = client.messages.create(
        model="claude-haiku-4-5",
        max_tokens=120,
        messages=[{"role": "user", "content": prompt}]
    )
    raw = response.content[0].text
    start = raw.find("{")
    end = raw.rfind("}") + 1
    return json.loads(raw[start:end])

print("ShopMax India Review Classification Report")
print("=" * 50)
for review in REVIEWS:
    result = classify_review(review["text"])
    flag = " *** URGENT ***" if result.get("urgent") else ""
    print(review["id"] + flag)
    print("Text:", review["text"][:60] + "...")
    print("Sentiment:", result["sentiment"],
          "| Confidence:", result["confidence"])
    print("Topics:", ", ".join(result["topics"]))
    print()

It gives the following output,

ShopMax India Review Classification Report
==================================================
R001
Text: Samsung TV delivered to Mumbai on time but remote was missing...
Sentiment: NEUTRAL | Confidence: 0.82
Topics: DELIVERY, PACKAGING

R002
Text: Picture quality is outstanding! Very happy with my purchase...
Sentiment: POSITIVE | Confidence: 0.97
Topics: PRODUCT_QUALITY

R003 *** URGENT ***
Text: Worst experience. AC stopped working after 2 days. Nobody...
Sentiment: NEGATIVE | Confidence: 0.99
Topics: PRODUCT_QUALITY, CUSTOMER_SERVICE

R004
Text: Decent product but delivery took 8 days instead of promised 3...
Sentiment: NEGATIVE | Confidence: 0.88
Topics: DELIVERY

R005
Text: Ekdum best washing machine! Hard water bhi saaf kar deta hai...
Sentiment: POSITIVE | Confidence: 0.95
Topics: PRODUCT_QUALITY

For ShopMax India at scale, batch classify reviews using the Batch API (50% cost savings) and route URGENT flagged reviews to a Slack alert channel for same-day response. Track sentiment trends by product category and city to identify systematic issues - a sudden spike in negative DELIVERY sentiment for Hyderabad may indicate a logistics partner problem. Use confidence scores below 0.7 as a signal to queue reviews for human verification before acting on them, since borderline cases (sarcastic reviews, mixed Hindi-English) have lower reliability.

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