🧠 AI with Python – 🏷️ Text Classification using Transformers

Every day, we generate enormous amounts of text—emails, customer reviews, news articles, support tickets, social media posts, and product feedback. Manually organizing all this information is time-consuming and often impractical.

This is where Text Classification becomes valuable.

Using Large Language Models and transformer-based architectures, we can automatically assign meaningful categories to text, enabling faster decision-making and intelligent automation.

In this project, we’ll build a Text Classification application using the Hugging Face Transformers pipeline.

---

What Is Text Classification?

Text Classification is a Natural Language Processing (NLP) task where a model assigns one or more predefined labels to a piece of text.

For example:

Input

The customer support team resolved my issue quickly.

Prediction

POSITIVE

Another example:

Input

The product stopped working after two days.

Prediction

NEGATIVE

The model analyzes the text and predicts the category that best represents its meaning.

---

How Transformers Classify Text

Transformer models learn language patterns from massive datasets.

When presented with new text, they:

• understand the context
• identify important words and phrases
• capture relationships between words
• predict the most appropriate label

Unlike traditional machine learning models that rely heavily on handcrafted features, transformers learn rich language representations automatically.

---

Creating a Text Classification Pipeline

Hugging Face makes text classification incredibly simple.

from transformers import pipeline

classifier = pipeline(
    "text-classification"
)

When executed for the first time, the required pretrained model and tokenizer are automatically downloaded.

After that, they are cached locally for future use.

---

Classifying a Single Sentence

Once the pipeline is ready, classifying text requires only a single function call.

result = classifier(
    "Python is an amazing programming language."
)

The model returns the predicted label along with a confidence score.

Typical output:

Label : POSITIVE
Score : 0.9998

This confidence score indicates how certain the model is about its prediction.

---

Processing Multiple Documents

One of the advantages of the Transformers pipeline is that it can process multiple texts in a single call.

documents = [
    "The service was excellent.",
    "The delivery was disappointing.",
    "The product quality exceeded expectations."
]

results = classifier(documents)

This makes it suitable for analyzing customer reviews, survey responses, or large collections of documents efficiently.

---

Understanding the Output

Each prediction contains two important pieces of information:

Label

The predicted category.

For example:

• POSITIVE
• NEGATIVE

Confidence Score

A probability-like value indicating how confident the model is in its prediction.

Higher scores generally indicate greater confidence.

---

How the Pipeline Works

Behind the scenes, the workflow follows these steps:

Input Text
      │
      ▼

Tokenizer
      │
      ▼

Transformer Model
      │
      ▼

Predicted Label
      │
      ▼

Confidence Score

The tokenizer converts text into tokens, the transformer processes these tokens, and the model predicts the most appropriate category.

---

Where Text Classification Is Used

Text classification powers many AI applications, including:

• sentiment analysis
• spam email detection
• customer feedback analysis
• support ticket routing
• news categorization
• document classification
• content moderation

It is one of the most widely deployed NLP techniques in production systems.

---

Why Use Transformers Instead of Traditional ML?

Compared to traditional approaches such as TF-IDF with Logistic Regression or Naive Bayes, transformer models offer several advantages:

• better understanding of context
• improved performance on complex language tasks
• no manual feature engineering
• support for many languages
• access to powerful pretrained models

This allows developers to build accurate NLP applications with very little code.

---

Key Takeaways

---

1. Text classification automatically assigns labels to text.
2. Hugging Face Transformers provides pretrained models that require minimal setup.
3. Each prediction includes both a label and a confidence score.
4. The same pipeline can classify individual texts or batches of documents.
5. Text classification is one of the most widely used applications of Large Language Models.

---

Conclusion

Text classification is a fundamental building block of modern Natural Language Processing. By combining pretrained transformer models with Hugging Face’s simple pipeline API, developers can quickly build intelligent systems that automatically organize and analyze text. Whether you’re filtering emails, analyzing customer reviews, or categorizing documents, transformer-based text classification provides a powerful and scalable solution.

This continues the LLM Foundations track in the AI with Python series. Next, we’ll explore Sentence Embeddings, where we’ll learn how AI converts text into numerical vector representations that power semantic search, document similarity, recommendation systems, and Retrieval-Augmented Generation (RAG).

---

# =========================================================
# 📦 Install Required Libraries
# =========================================================

# Run this in terminal if not installed:
# pip install transformers torch


# =========================================================
# 📦 Import Required Libraries
# =========================================================

from transformers import pipeline


# =========================================================
# 🤖 Create Text Classification Pipeline
# =========================================================

classifier = pipeline(
    task="text-classification"
)


# =========================================================
# 📝 Classify a Single Sentence
# =========================================================

text = (
    "Python is an amazing programming language for AI development."
)

result = classifier(text)


# =========================================================
# 📊 Display Prediction
# =========================================================

print("=== Single Text Classification ===\n")

print("Input Text:")
print(text)

print("\nPredicted Label:")
print(result[0]["label"])

print("Confidence:")
print(
    round(
        result[0]["score"],
        4
    )
)


# =========================================================
# 📄 Classify Multiple Documents
# =========================================================

documents = [

    "The customer service was excellent.",

    "The delivery was delayed and disappointing.",

    "The product quality exceeded my expectations.",

    "I will never buy this product again.",

    "The support team resolved my issue quickly.",

    "The application crashes every time I open it."

]

results = classifier(documents)


# =========================================================
# 📋 Display Batch Predictions
# =========================================================

print("\n\n=== Multiple Document Classification ===")

for text, prediction in zip(
    documents,
    results
):

    print("\n" + "=" * 60)

    print("Text:")
    print(text)

    print("\nPredicted Label:")
    print(prediction["label"])

    print("Confidence:")
    print(
        round(
            prediction["score"],
            4
        )
    )


# =========================================================
# 📊 Create Prediction Summary
# =========================================================

summary = []

for text, prediction in zip(
    documents,
    results
):

    summary.append({

        "Text": text,

        "Label": prediction["label"],

        "Confidence": round(
            prediction["score"],
            4
        )

    })

print("\n\n=== Prediction Summary ===\n")

for item in summary:

    print(item)


# =========================================================
# 📈 Count Labels
# =========================================================

positive_count = sum(
    1
    for item in summary
    if item["Label"] == "POSITIVE"
)

negative_count = sum(
    1
    for item in summary
    if item["Label"] == "NEGATIVE"
)

print("\n=== Classification Statistics ===")

print(
    "Positive Predictions:",
    positive_count
)

print(
    "Negative Predictions:",
    negative_count
)


# =========================================================
# 🧪 Try Custom Examples
# =========================================================

examples = [

    "The movie was absolutely fantastic.",

    "The laptop stopped working after one day.",

    "I highly recommend this course.",

    "The hotel room was dirty and noisy."

]

example_results = classifier(
    examples
)

print("\n\n=== Custom Examples ===")

for text, prediction in zip(
    examples,
    example_results
):

    print("\n" + "=" * 60)

    print("Input:")
    print(text)

    print("\nPrediction:")
    print(prediction["label"])

    print("Confidence:")
    print(
        round(
            prediction["score"],
            4
        )
    )


# =========================================================
# 💾 Save Results
# =========================================================

import pandas as pd

results_df = pd.DataFrame(summary)

results_df.to_csv(
    "text_classification_results.csv",
    index=False
)

print(
    "\nResults saved to text_classification_results.csv"
)


# =========================================================
# 📂 Load Saved Results
# =========================================================

loaded_results = pd.read_csv(
    "text_classification_results.csv"
)

print("\n=== Loaded Results ===\n")

print(loaded_results)


# =========================================================
# ✅ Final Note
# =========================================================

print("\nText Classification demo completed successfully.")