Public sentiment plays a crucial role in shaping the reputation of public figures, especially in the media industry. Recently, a Facebook post discussing why famous radio hosts failed to expand their reach in another part of the country sparked public debate.
To understand why this happened and what the public thinks, we conducted an in-depth sentiment analysis of the comments on this post. However, not all comments carry the same weight—some receive hundreds of likes, making them more influential in shaping overall sentiment.
This project explores how weighting sentiment based on likes can provide a more accurate representation of public opinion.
- Perform sentiment analysis to classify public opinion into positive, neutral, or negative categories.
- Prioritize high-engagement comments by giving more weight to comments with higher likes.
- Identify trending discussion topics using NLP techniques.
- Explore correlations between comment length, sentiment, and engagement.
- Demonstrate the power of feature weighting in sentiment analysis.
- Brand reputation management (analyzing customer feedback).
- Political sentiment analysis (measuring voter perception).
- Social media monitoring (identifying trends and controversies).
The dataset consists of:
- Comments: User-generated opinions on the Facebook post.
- Likes: The number of likes each comment received (used as a weight for importance).
By analyzing this data, we aim to understand the public mood and identify dominant narratives.
We analyzed how likes were distributed across comments to understand engagement patterns.
- Average number of likes per comment:
57.15 - Maximum number of likes received:
1700
- Some comments received very high engagement, making them key influencers in shaping public perception.
- Many comments had low engagement, which could skew sentiment analysis if not properly weighted.
To understand which opinions resonated the most, we extracted the most and least liked comments.
| Comment | Likes |
|---|---|
| "Can't believe the world we live in when they get p..." | 1700 |
| "they were beaten by ABC local radio..." | 1500 |
| "One of the most embarrassing things about Australi..." | 1400 |
| "I would need to be paid $10 million a year to list..." | 1400 |
| "I don't understand how people in Sydney like them...." | 873 |
| Comment | Likes |
|---|---|
| "I cant stand them Jase and Lauren are so much bet..." | 2 |
| "Certainly aren't worth anything like 10 million. K..." | 2 |
| "Well done Loz and Jase your the best for sure They..." | 2 |
| "I don't find the big deal about them Pro because t..." | 2 |
| "I can do the job for $1 0 0 Paying that salary for..." | 2 |
- The most liked comments strongly criticized the radio hosts, indicating negative sentiment dominates engagement.
- The least liked comments were neutral or positive, showing that critical opinions attract more attention.
We categorized each comment into Positive, Neutral, or Negative sentiment using NLP.
- Negative sentiment was dominant, especially among high-liked comments.
- Neutral and positive comments were significantly lower, indicating widespread dissatisfaction.
We extracted key discussion topics using:
- A word cloud to visualize common terms.
- Topic modeling (LDA) to group discussions into themes.
| Topic | Top Keywords |
|---|---|
| 1 | looks, joke, aren, jackie, choice, public, let, listen, 10, worth |
| 2 | just, 10, don, people, worth, sydney, kyle, melbourne, listen, radio |
| 3 | melburnians, high, course, triple, ridiculous, victorians, talent, little, big, stand |
- The most common words reflect strong opinions about radio hosts.
- The LDA model identified key discussion themes, such as host salaries, listener dissatisfaction, and station changes.
We analyzed whether comment length affects engagement.
- Weak negative correlation (
-0.367), meaning shorter comments tend to receive more likes. - Longer comments provide more detail but are less engaging for casual users.
We examined relationships between sentiment, comment length, and number of likes.
| Metric | Spearman Correlation |
|---|---|
| Sentiment Score vs. Likes | -0.059 |
| Comment Length vs. Likes | -0.367 |
- No strong correlation between sentiment and likes (both positive and negative comments could receive high engagement).
- Shorter comments tend to receive more engagement.
This analysis proves the power of weighting user engagement in sentiment analysis. If we had performed basic sentiment analysis, the overall sentiment would have been misleading.
- Negative sentiment dominates public discourse and drives engagement.
- Feature weighting (likes) alters sentiment perception, revealing more accurate public opinion.
- Shorter, sharp comments attract more likes, while long-form comments go unnoticed.
- Implement deep learning models for advanced sentiment analysis.
- Compare sentiment trends across different social media platforms.
- Analyze time-based engagement trends if timestamp data is available.
📌 This project showcases the impact of weighting features in sentiment analysis and its applications in marketing, politics, and customer feedback analysis. 🚀