SLIDE 1
Use of Artificial Intelligence in FINN
DND Software 2020 Nicolai Høge
1999 - 2001 2001 - 2006 2006 - 2008
2008
Nicolai Høge
Intelligence in FINN DND Software 2020 Nicolai Hge 2008 Nicolai - - PowerPoint PPT Presentation
Intelligence in FINN DND Software 2020 Nicolai Hge 2008 Nicolai - - PowerPoint PPT Presentation
Use of Artificial Intelligence in FINN DND Software 2020 Nicolai Hge 2008 Nicolai Hge 2001 - 2006 1999 - 2001 2006 - 2008 Hva skal jeg snakke om? Hva er kunstig intelligens og maskinlring? To konkrete eksempler fra FINN:
SLIDE 2
SLIDE 3
Hva skal jeg snakke om?
- Hva er kunstig intelligens og maskinlæring?
- To konkrete eksempler fra FINN:
- Annonsekontroll
- Anbefalinger
- Hva har vi lært etter 5+ år ?
SLIDE 4
What is Artificial Intelligence?
- To get machines to solve tasks, that
humans need intelligence to do
- Most of the work around (AI) is open
and available to everyone. That means that knowledge is not the key, but rather the access to data and competent people
- Many sub categories
○ NLP, ML, Image Recognition etc.
SLIDE 5
The computer learns how to categorize or group input
Why is everybody talking about Machine Learning?
Input Algorithm Output Input Output Model Input Model
Without ML Supervised Unsupervised
SLIDE 6
Ad Control
SLIDE 7
1. 2. 3.
Machine Learning for Ad Control
Ad Control 500 000 changes pr / week Store the results
SLIDE 8
Human rules, manual weighting
SLIDE 9
ML weighting of rules
From manual weighting to ML weighting of rules
SLIDE 10
ML generated rules
From manual rules to ML generated rules
SLIDE 11
Train model with more data
Train the ML model with more data
SLIDE 12
Keep tuning until “good enough”
SLIDE 13
Recommendations
SLIDE 14
One visit, many signals
Typically 100 signals pr. visit
100 different signals pageviews recommendation inscreen contact_action search ...
SLIDE 15
Many visits, A LOT of signals
200-300 million signals pr. day Data lake 9 TB
SLIDE 16
From signals to recommendations
Data lake 9 TB
Machine Learning
- 30 - 40 models
- Experimentation
- Always 3-4 different
algorithms running against each other
- Validation of models in
production
AB testing
SLIDE 17
Recommendations based on users or objects
Recommendation Service
- r
Input Output
SLIDE 18
One engine, many products
Recommendation Service Personalized search BLINK Recommendations Use signals to improve recommendations
SLIDE 19
Results
CTR 30% 1000 clicks/ min
SLIDE 20
What is under the hood?
SLIDE 21
Combining more than one model
- Collaborative Filtering works
with a lot of data
- Confused with little data
- Combine text, image and
collaborative filtering
KDD 2018: Eide, Zhou, Øygard: Five lessons from building a deep neural network recommender
SLIDE 22
More relevant results
SLIDE 23
When things go wrong...
SLIDE 24
SLIDE 25
- Data Scientists are part of the
product development teams
- Tons of experiments
- Our ability to create products
Our success factors... ...and challenges
- Difficult to recruit the right
people
- Hard to gather data
- The models are only as good
as the data they learn from
- Off the shelves models may be
good enough, but you probably need use them in inventive ways