Sentiment Analysis — Hands-on with Python

Lexicon-Based Sentiment with VADER (Great for Social Media Text)

VADER (Valence Aware Dictionary and sEntiment Reasoner) is a lexicon-and-rule-based sentiment tool tuned specifically for short, informal text (tweets, reviews) — it natively handles negation, punctuation emphasis ("!!!"), capitalization ("GREAT" vs "great"), and degree modifiers ("very good").

import nltk
from nltk.sentiment import SentimentIntensityAnalyzer

nltk.download('vader_lexicon')
sia = SentimentIntensityAnalyzer()

texts = [
    "The movie was absolutely fantastic!",
    "The movie was not good at all.",
    "The movie was okay, nothing special.",
    "The movie was GREAT!!! Loved it :)"
]

for t in texts:
    scores = sia.polarity_scores(t)
    print(t, "->", scores)

# The movie was absolutely fantastic! -> {'neg': 0.0, 'neu': 0.406, 'pos': 0.594, 'compound': 0.6239}
# The movie was not good at all.      -> {'neg': 0.379, 'neu': 0.621, 'pos': 0.0, 'compound': -0.3865}
# The movie was okay, nothing special.-> {'neg': 0.0, 'neu': 1.0, 'pos': 0.0, 'compound': 0.0}
# The movie was GREAT!!! Loved it :)  -> {'neg': 0.0, 'neu': 0.238, 'pos': 0.762, 'compound': 0.8955}

• compound is the single overall normalized score, ranging from -1 (most negative) to +1 (most positive).
• Notice VADER correctly flips "not good" to negative, and boosts "GREAT!!!" with capitalization + exclamation marks.

def classify_sentiment(compound):
    if compound >= 0.05:
        return "Positive"
    elif compound <= -0.05:
        return "Negative"
    return "Neutral"

for t in texts:
    score = sia.polarity_scores(t)['compound']
    print(t, "->", classify_sentiment(score))

Lexicon-Based Sentiment with TextBlob (Simple API)

from textblob import TextBlob

text = "The battery life is excellent but the camera is quite disappointing."
blob = TextBlob(text)
print(blob.sentiment)
# Sentiment(polarity=0.05, subjectivity=0.65)
# polarity: -1 (negative) to +1 (positive)
# subjectivity: 0 (objective/factual) to 1 (subjective/opinion)

ML-Based Sentiment Classifier (Reusing the Naïve Bayes Pipeline)

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.pipeline import Pipeline

reviews = [
    "amazing product loved it", "terrible quality very disappointed",
    "excellent value for money", "worst purchase ever made",
    "great performance highly recommend", "not worth the price at all"
]
sentiments = ["positive", "negative", "positive", "negative", "positive", "negative"]

pipeline = Pipeline([
    ('tfidf', TfidfVectorizer()),
    ('classifier', MultinomialNB())
])
pipeline.fit(reviews, sentiments)

print(pipeline.predict(["great quality highly recommend"]))  # ['positive']
print(pipeline.predict(["very disappointed with this"]))     # ['negative']

Aspect-Based Sentiment (Simplified Rule-Based Approach)

from nltk.sentiment import SentimentIntensityAnalyzer
sia = SentimentIntensityAnalyzer()

review = "The camera is disappointing but the battery life is excellent."

aspects = {"camera": "camera is disappointing",
           "battery": "battery life is excellent"}

for aspect, clause in aspects.items():
    score = sia.polarity_scores(clause)['compound']
    sentiment = "Positive" if score > 0 else "Negative" if score < 0 else "Neutral"
    print(f"{aspect}: {sentiment} ({score})")
# camera: Negative (-0.4939)
# battery: Positive (0.5719)

Real-World Deployment Considerations