Sentiment Analysis

So far, you have learned how different algorithms like Naive Bayes, Logistic Regression, and SVM can classify text into categories.

Now we apply all that knowledge to one of the most popular and useful NLP tasks: Sentiment Analysis.

Sentiment Analysis helps machines understand human emotions and opinions from text such as reviews, comments, tweets, and feedback.

What Is Sentiment Analysis?

Sentiment Analysis is the process of identifying emotional tone behind a piece of text.

It answers questions like:

• Is this review positive or negative?
• Is the customer satisfied?
• What is the public opinion about a product?

In most basic form, sentiment analysis is a text classification problem.

Types of Sentiment Analysis

Sentiment analysis can be done at different levels, depending on the problem.

1. Binary Sentiment

Classifies text as:

• Positive
• Negative

Example: “This movie was amazing” → Positive

2. Multi-Class Sentiment

More detailed sentiment categories:

• Positive
• Neutral
• Negative

3. Fine-Grained Sentiment

Even more detailed:

• Very Positive
• Positive
• Neutral
• Negative
• Very Negative

Why Sentiment Analysis Is Important

Sentiment analysis plays a major role in decision-making.

• Companies analyze customer feedback
• Brands monitor social media opinion
• Governments analyze public response
• Investors analyze market sentiment

It converts unstructured opinions into actionable insights.

Challenges in Sentiment Analysis

Human language is complex. Machines struggle with:

• Sarcasm (“Great service… waited 2 hours”)
• Context (“The phone is light” vs “The punishment is light”)
• Negation (“not good”, “not bad”)
• Mixed sentiment in one sentence

This is why preprocessing and model choice matter.

Sentiment Analysis Approaches

1. Rule-Based Approach

Uses predefined sentiment dictionaries (lexicons).

• Positive words → +1
• Negative words → −1

Simple but limited. Does not scale well.

2. Machine Learning Approach

Uses labeled data and ML models:

• Naive Bayes
• Logistic Regression
• SVM

Works well with enough data.

3. Deep Learning Approach

Uses neural networks:

• LSTMs
• Transformers
• BERT / GPT-based models

Best performance but requires more resources.

Classic ML Pipeline for Sentiment Analysis

Most industry pipelines follow these steps:

1. Text cleaning
2. Tokenization
3. Vectorization (TF-IDF)
4. Model training
5. Prediction

You already learned each of these steps separately.

Practical Example: Sentiment Analysis Using Logistic Regression

This example demonstrates a complete sentiment analysis flow.

Where to run this code:

• Google Colab (recommended)
• Jupyter Notebook (Anaconda)

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression

texts = [
    "I love this product",
    "This is the best experience",
    "I hate this service",
    "This is terrible",
    "Not satisfied at all"
]

labels = [1, 1, 0, 0, 0]  # 1 = Positive, 0 = Negative

vectorizer = TfidfVectorizer()
X = vectorizer.fit_transform(texts)

model = LogisticRegression()
model.fit(X, labels)

test_texts = [
    "This product is amazing",
    "I am very disappointed"
]

X_test = vectorizer.transform(test_texts)
predictions = model.predict(X_test)

for text, pred in zip(test_texts, predictions):
    print(text, "->", "Positive" if pred == 1 else "Negative")

Output Explanation:

• TF-IDF captures word importance
• Logistic Regression learns sentiment patterns
• The model predicts sentiment for unseen text

Understanding the Results

The model predicts sentiment based on learned word patterns.

• Words like “love”, “best”, “amazing” → Positive
• Words like “hate”, “terrible”, “disappointed” → Negative

Context and data size strongly affect accuracy.

Real-Life Applications

• Product review analysis
• Social media monitoring
• Customer support automation
• Brand reputation tracking

Assignment / Homework

Theory:

• Explain challenges in sentiment analysis
• Compare rule-based vs ML-based sentiment analysis

Practical:

• Add neutral reviews
• Try SVM instead of Logistic Regression
• Test sarcastic sentences

Practice environment:

• Google Colab
• Jupyter Notebook

Practice Questions

Quick Quiz

Quick Recap

• Sentiment analysis extracts emotions from text
• It is a text classification problem
• ML models work well with TF-IDF
• Used heavily in real-world systems

In the next lesson, we will learn about Topic Modeling and how machines discover hidden themes in documents.