Sentiment Analysis and Customer Satisfaction Prediction on E-Commerce Platforms Based on YouTube Comments Using the XGBoost Algorithm

Imagine you own a giant, noisy marketplace (an e-commerce platform). In the past, if customers were unhappy, they would write a formal letter to the manager. But today, instead of writing letters, they are shouting their opinions on a massive, public video stage called YouTube.

This paper is like a team of detectives trying to figure out exactly how happy or angry these shouting customers are, but there's a catch: they are shouting in a chaotic mix of languages, slang, and even political rants.

Here is the story of how the researchers solved this puzzle, explained simply:

1. The Problem: Too Much Noise

The marketplace is so popular that thousands of people are posting comments on YouTube videos reviewing products. Trying to read every single comment by hand is impossible—it's like trying to drink from a firehose. Plus, the comments are messy. Some people are happy, some are furious, and many are just confused. The researchers needed a robot brain to sort through this chaos and tell the business owners: "Are our customers satisfied or not?"

2. The Tool: The "Smart Tree" (XGBoost)

To solve this, the researchers didn't use a fancy, complex robot brain (like a Deep Learning AI that usually needs millions of examples to learn). Instead, they used a tool called XGBoost.

Think of XGBoost as a team of smart detectives working in a relay race.

• The first detective looks at a comment and makes a guess.
• The second detective looks at where the first one made a mistake and tries to fix it.
• The third detective fixes the second one's mistakes, and so on.
  By the time the team is done, they have built a very sharp, accurate picture of what the customer really feels. They also used a method called TF-IDF, which is like a highlighter pen that ignores common words (like "the" or "and") and only highlights the unique, important words that actually tell you if someone is angry or happy.

3. The Surprise Discovery: Politics in the Product Aisle

When the researchers looked at the comments, they found something weird. They expected to see complaints like "The shipping was slow" or "The shirt is too small."

Instead, they found that the comments were heavily polluted with political arguments.

• The Metaphor: Imagine walking into a bakery to complain about a burnt cookie, but instead of talking about the cookie, the customer starts shouting about foreign spies, political conspiracies, and international figures.
• The data showed that words like "foreign minions" and names of political figures appeared constantly in negative reviews. The researchers found that these political rants were actually the biggest drivers of "negative" feelings, more so than the actual product quality. The customers were using the product review section as a soapbox for political grievances.

4. The Race: Simple Team vs. Complex Robot

The researchers ran a race to see which method worked best:

• Team A (The Smart Tree/XGBoost): The traditional, efficient method described above.
• Team B (The Deep Learning Robot/LSTM): A complex neural network that tries to understand the order of words, like a human reading a story.

The Result: Team A won, but just barely.

• Team A (XGBoost): Got 76% accuracy.
• Team B (LSTM): Got 74% accuracy.

Why did the simple team win?
Think of the dataset as a small library with only a few thousand books. The complex robot (Team B) is like a genius student who needs to read a million books to learn how to write a good essay. If you only give them a few books, they get confused and make mistakes. The simple team (Team A), however, is like a practical worker who can get the job done perfectly well with just a few books. Because the YouTube comment data wasn't huge enough, the complex robot actually performed worse than the efficient, traditional team.

5. The Conclusion

The paper concludes that you don't always need the most expensive, complex AI to solve a problem.

• The Verdict: A well-organized team of "smart detectives" (XGBoost) is actually better at reading these specific YouTube comments than the super-complex robot, especially when the data is messy and unbalanced (mostly angry people).
• The Warning: The researchers also noted that because the comments are so unbalanced (mostly angry) and filled with political noise, future attempts to analyze this data will need to use special techniques to "balance the scales" so the AI doesn't just learn to be angry all the time.

In short: The researchers built a system to read YouTube comments about online shopping. They found that people are more likely to rant about politics than about products, and they proved that a simpler, faster computer program was actually better at understanding this specific chaos than a more complex one.