
www.github.com/d-pap/NLP-sentiment-analysis
Summary
I analyzed 61,000 tweets about CS:GO and competitor games, built a 94.7% accurate sentiment analysis model, and used the insights to suggest targeted marketing strategies. The goal: help gaming companies attract and retain more players by understanding their frustrations and interests.
Insights
Approaching this as if I worked at Valve (makers of CS:GO), I found three key insights:
-
Cheating frustrates CS:GO players the most
Recommendation: Regularly highlight Valve’s anti-cheat actions publicly. -
Players love Twitch content and cosmetic items ("skins")
Recommendation: Run more Twitch events and giveaways, similar to competitor successes (Apex, Valorant). -
Many competitor players (Call of Duty, PUBG, Fortnite, Rainbow Six) are highly frustrated
Recommendation: Use targeted ads highlighting CS:GO’s smoother updates, strong stability, and availability in markets like India, where PUBG was banned.
To see how I got this, check out my GitHub Repository or read my Medium article
Potential Business Impact
To see the financial benefit of implementing targeted sentiment-driven advertising, consider this scenario:
Audience. Suppose Valve targets social media users expressing frustration with competitors.
- From roughly 32 million monthly active CS:GO users (ActivePlayer.io, 2024), a conservative assumption is that Valve’s marketing efforts could reach 5% (1.6 million) of users who are dissatisfied with competitor games.
Conversion assumption. With targeted and relevant ads, it's reasonable to assume 2% of these users could be converted into new or returning active CS:GO players each month.
- Monthly new players gained: 1.6 million users x 2% conversion = 32,000 new players/month
Revenue impact (reasonable assumption). CS:GO primarily generates revenue through cosmetic items and keys, with estimated average revenue per paying user (ARPPU) around $5 per month (Newzoo Cosmetics Report, 2023).
- Monthly revenue gain: 32,000 players × $5 ARPPU = $160,000 per month
- Annual revenue gain: $160,000 × 12 months = $1.92 million annually
Even with conservative assumptions, leveraging sentiment analysis to better target frustrated competitor audiences can yield a significant revenue uplift.
How to Use It
- Deploy sentiment analysis in daily social listening for real-time targeting.
- Implement targeted ad tests to measure real-world campaign effectiveness (CTR, conversion rates).
- Expand analysis to additional platforms (Reddit, YouTube) for richer audience insights.
Tools
- Python, pandas, scikit-learn (data prep & NLP analysis)
- TensorFlow, Keras (model building)
- Matplotlib, Seaborn (visualizations)
Behind the Scenes



I tested several NLP models (Logistic Regression, Gradient Boosting, MLP Neural Network) and selected an MLP model because it achieved the highest accuracy (94.7%). High accuracy means we reliably understand player sentiment and effectively target ads.
Why sentiment analysis?
- Identifies real-time player frustrations.
- Helps marketing create relevant, timely ads.
- Significantly improves ad effectiveness (higher click rates, conversions).
Future steps could expand the analysis to platforms like Reddit and YouTube, adding richer insights.