Introducing
Your new presentation assistant.
Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.
Trending searches
webexpo - predictive models
2
WHY YOU NEED PREDICTIVE MODELS AND RECOMMENDER SYSTEMS?
Pavel Kordík
Computational Intelligence research Group, CTU in Prague
So why you do not use it already?
Is there anything all these successful companies have in common?
Too complex background knowledge needed?
They all use RED color in their logo?
Your programmers are not data scientists?
Start again ...
They increasingly rely on PREDICTIVE MODELS and RECOMMENDER SYSTEMS!
You do not see the benefits?
Tools and solutions are too expensive?
Maintain model
Deploy model
Validate model
Build model
Preprocess data
Aggregate data
Collect data
Estimate profit
Estimate improvement in
customer value
Remember?
Identify a business goal
Association rules!
Conclusion
- You will need these technologies to stay ahead of your competitors.
- Such innovations have potential to improve your services significantly.
- Successful companies already know that data, models and services are their weapons in future battlefields.
How you can use predictive models and recommender systems in your business?
Where are those PREDICTIVE MODELS?
Personalized campaigning
Thanks!
Great offer!
Give me more!
which customers are likely to respond?
predictive model will identify a subset
of your customers with the highest probability to buy your new services or products,
so you do not waste money and patience of your customers.
Score produced by a PREDICTIVE MODEL
In this case a regression forest
- Churn, upsell and downsell prediction
- Personalised campaigning
- E-shop recommendations
- Web customization for user segments
- Intrusion detection
- Anomaly detection
- Fraud detection
- ...
Score:
6.2
5.7
3.6
2.1
2.0
E-shop recommendations
Your customers
what I should offer to each customer?
Churn prediction
recommeder system computes
You might also like ...
which customers are likely to leave?
Model is built from data
predictive model will identify a subset
Stay with us, we have a special offer for you!
best matches for each of your customers and improve their shopping experience
so you will sell more!
of your customers with the highest churn or downsell probability
so you can prevent it before it happens!
Better model = better prediction = better recommendation
Models in Recommender systems
Diversity
Find three nearest neighbors
Combined models achieve best precision and diversity of recommedation
3-NN
:-)
Recommend based on their purchase history
Based on user history
Star Wars 3
Predict movies to recommend him
Mission Impossible
Hitchhikers Guide to the Galaxy
Task: Recommend movies (or items in general)
Precision