Sentiment Analysis Demo: Comparing OpenAI LLMs and TextBlob

Introduction

Sentiment analysis is a classic Natural Language Processing (NLP) task that aims to determine whether a piece of text expresses a positive, negative, or neutral sentiment.

In this blog post, we’ll walk through a simple but practical demo that compares:

• A Large Language Model (LLM) from OpenAI
• A traditional rule/statistics-based NLP library, TextBlob

By the end, you’ll clearly see the strengths and limitations of each approach.

Tools Used

1. OpenAI LLM (GPT-style)

• Context-aware
• Handles negation and nuanced language well
• Can provide explanations in addition to labels

2. TextBlob

• Lightweight and fast
• Polarity-based scoring (-1.0 to 1.0)
• Limited understanding of complex phrasing

Step-by-Step Implementation

Step 1: Import Required Libraries

import openai
import time
from textblob import TextBlob

• openai: interacts with OpenAI language models
• time: handles timeout-related logic
• TextBlob: performs traditional sentiment analysis

Step 2: Define the Sentiment Analysis Function

We define a function that:

1. Sends text to an OpenAI model for sentiment analysis
2. Extracts a sentiment label from the LLM output
3. Computes a polarity score using TextBlob
4. Prints and compares the results

def print_sentiment_score(text):
    # --- OpenAI Sentiment Analysis ---
    result = openai.Completion.create(
        engine="gpt-3.5-turbo-instruct",
        prompt=f"Please analyze and give the detailed sentiment of the following text: {text}",
        temperature=0,
        max_tokens=128,
        timeout=10
    )

    openai_sentiment = (
        result.choices[0].text
        .strip()
        .replace("The sentiment of the text is ", "")
        .rstrip(".")
    )

    if "positive" in openai_sentiment.lower():
        openai_sentiment_score = "positive"
    elif "negative" in openai_sentiment.lower():
        openai_sentiment_score = "negative"
    else:
        openai_sentiment_score = "neutral"

    # --- TextBlob Sentiment Analysis ---
    blob = TextBlob(text)
    polarity = blob.sentiment.polarity

    if polarity > 0:
        textblob_sentiment_score = "positive"
    elif polarity < 0:
        textblob_sentiment_score = "negative"
    else:
        textblob_sentiment_score = "neutral"

    # --- Output ---
    print(f"Text: {text}")
    print(f"OpenAI Sentiment Analysis: {openai_sentiment_score} ({openai_sentiment})")
    print(f"TextBlob Sentiment Analysis: {textblob_sentiment_score}")

Step 3: Run the Demo

text = "I absolutely do not feel like waking up early in the morning"
print_sentiment_score(text)

Sample Output

Text: I absolutely do not feel like waking up early in the morning
OpenAI Sentiment Analysis: negative (strong dislike and reluctance toward waking up early)
TextBlob Sentiment Analysis: neutral

Why the Results Differ

TextBlob Limitation

TextBlob mainly relies on individual word polarity.
Phrases like “absolutely not” and emotional context are poorly captured.

OpenAI LLM Advantage

The LLM understands:

• Negation
• Emphasis words
• Emotional intent
• Sentence-level meaning

This makes it far more reliable for real-world sentiment analysis.

When to Use Which?

Conclusion

This demo highlights a key takeaway:

Traditional NLP tools are fast, but LLMs are far more accurate and expressive.

If accuracy, context, and explainability matter — LLM-based sentiment analysis is the clear winner.

Next Steps

• Add confidence scores
• Batch process CSV data
• Store results in a database
• Replace legacy sentiment pipelines with LLM-based analysis

Happy building 🚀

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License .

• ← Previous Post