3. Data Science
Yet another probably useless definition:
Interdisciplinary technically-oriented field focusing on acquisition of
innovative business-relevant intelligence from very heterogeneous
data sources.
The key word in “Data Science” is not Data, it is Science.
(Jeff Leek, 2013)
5. Three big-data loving modes of telco (1)
Mode 1
Selling Data
• many subjects are interested in
insights hidden in telco data
• anonymized telco data are sold
externally
• telco company exploits only
negligible potential of the data
• what about differential privacy?
$
6. Three big-data loving modes of telco (2)
Mode 2
Data-Driven Enhancements
• data products enhance
internal processes and current
business
• utilization of Big Data platform
for BI and product/process
improvement
• “data science“ can be partially
automated out-of-the-box
Actions
Enhancements
Optimization
7. Three big-data loving modes of telco (3)
Mode 3
Monetizing Data-Driven
Intelligence
• new revenue streams based on
data products
• data-driven intelligence, insights
and products are sold instead of
raw data
• Big Data platform and full-blown
Data Science team – disruptive
potential for new business
development
Actions
Enhancements
Optimization
9. O2 Big Data Platform
Teradata
EDW
HDP
Hadoop
Teradata
Aster
Local R
and
Python
10. O2 Big Data Streams
Customer
EDW data
Webtraffic
DPI
SS7
network
stream
O2TV
STB event
stream
11. Some of O2‘s Data Products
OneTable
Customer
Geoprofiling /
TVprofiling
Customer
Webprofiling
Look Alike
Targeting
Models
Outdoor
Advertisement
Measuring
Credit Scoring
Mobility
Modeling and
Analysis
Next Best Offer
Cross-Media
Analysis and
Targeting
O2
Media
Liberty
API
13. Customer Profiling
• continuous modelling of attributes and segmentations
of resident customers
• used for marketing segmentation, targeting, BI
analyses, etc., but also as inputs (predictors) to most
of other data products
• done from all data sources
• wide range of mutually interconnected models and
filters of various types
• fast/slow attributes
• interpretable/latent attributes
• automatization and integration of all outputs in our
OneTable
14. Next Best Offer
• O2 product recommendation system
• developed in-house – suits all O2’s specificities
• own algorithm – combination of unsupervised
clustering a supervised predictive models
• utilization of all predictors from OneTable
• large matrix operations
• ARPU increased at all channels using NBO
18. Smart Targeting
• supervised and unsupervised machine learning for
targeting of B2B2C campaigns (SMS, online)
• click prediction from all predictors in OneTable
• feedback-based learning
19. SNA and Community Detection
• predictive models for detection of communities
(families, interest groups, etc.) and Social Network
Analysis from CDR
• marketing utilization
• input to other data products
• utilization of NLP methods modified for graph
analytics
• utilization of neural networks and other graph
analytics tools
20. Web-Interests
• segmentation of customers’ interests based on web
browsing behavior
• automatic webscraping and machine classification of
web pages to interest categories based on text
processing
• good old bag-of-words & SVM approach with n-grams
and TF/IDF – still suitable for this needs
• statistical modelling of customers’ interests based on
interest categories of visited webs