20140918 CeAdar Case Study Identitymatch_Identifying Persons of Interest in Social Network_Idiro and...
-
Upload
irish-software-innovation-network -
Category
Technology
-
view
83 -
download
1
description
Transcript of 20140918 CeAdar Case Study Identitymatch_Identifying Persons of Interest in Social Network_Idiro and...
Copyright © CeADAR 2014 1 Copyright © CeADAR 2014
IdentityMatch: Identifying Persons of Interest in Social Networks Thursday 18th September 2014, Dublin Tuesday 23rd September 2014, Cork Dr. Oisín Boydell, CeADAR Aidan Connolly, Idiro Kevin Neary, ConnectorsMarketplace
Copyright © CeADAR 2014 2
Research Theme
Social Identity Fingerprinting concerns identifying people of interest across social networks
Visualisation & Analytic Interfaces
• ‘Beyond the desktop’
• Ease of interaction
• Changing user behaviour
• Passive analytics
Data Management for Analytics
• Reduce data management effort for analytics
• Data validation
• Relevance of events to relationships
• Data curation (determining useful data)
• Adaptive ETL (Extract, Transform, Load)
Advanced Analytics
• Causation challenge
• Live topic monitoring
• Social trending and contextualisation
• Continuous analytics
• Social Identity fingerprinting
Copyright © CeADAR 2014 3
Research Theme
This is a common analytics need within different industries and different types of data.
Social networks? – not just Twitter, Facebook, Google+ etc.
– Telecoms and communications data
– Financial transactions
– Other records of interactions between people or entities
Copyright © CeADAR 2014 4
Research Theme
People of interest?
– Wide ranging and dependent on the domain
– Examples: • People with specific skills or expertise
• People exhibiting certain behaviors (fraud, churn, propensity to buy, multi-sim use etc.)
• Influencers in a particular area
• Bots and fake users
Copyright © CeADAR 2014 5
Two project partners
Both from different industries with different sources of data, but both share a similar analytics need.
Project Partners
Copyright © CeADAR 2014 6
Copyright © CeADAR 2014 7
Challenge & Solution
• World class Irish researchers
• Low risk Rapid Prototyping
• Industry-led engagement
model
– easy & convenient
• Analyse Social Signals at Scale
• Convert large amount of raw data
to valuable information
• Integrate data from different
sources
Requirement: - Viable Business Tool
CeADAR Solution: - IdentityMatch Technology
Copyright © CeADAR 2014 8
Aidan Connolly, CEO
Copyright © CeADAR 2014 9
Idiro’s Reach
Idiro have analysed data for 12% of the world’s population!
Copyright © CeADAR 2014 10
Contagion
Copyright © CeADAR 2014 11
Samsung users
Apple users
Copyright © CeADAR 2014 12
0
1000
2000
3000
4000
5000
6000
Operator Idiro
Marketing Campaign
Over 400% improvement
Copyright © CeADAR 2014 13
Dual-SIM Behaviour
● Dual-SIM usage is a major challenge for many mobile operators. ● How can we identify dual-SIM users from non dual-SIM users? ● Enter CeADAR Research Project.
Copyright © CeADAR 2014 14
CeADAR Platform Solution
Key approach: Combine both network based features with content analysis
Example network features: – How people are related in the social graph – Diversity of outgoing/incoming connections – Social influence
Example content analysis – Textual analysis of posts, links, profile description etc – Analysis of actions and patterns e.g. time of day of communication type, mobile
top-up amount and frequency etc.
Copyright © CeADAR 2014 15
CeADAR Solution: ConnectorsMarketplace
ConnectorsMarketplace required an interactive user-facing solution to aid finding good ‘connectors’ with defined areas of expertise
Specific challenges:
– User friendly and intuitive
– Search over live data (Twitter)
– An iterative, supervised machine learning approach whereby selected results refine the system in further search iterations.
Copyright © CeADAR 2014 16
CeADAR Solution: ConnectorsMarketplace
Copyright © CeADAR 2014 17
CeADAR Solution: Idiro
Idiro required a solution that could leverage very faint signals in the data
Specific challenges:
– High volume telecoms data
– Ability to identify and exploit very faint signals and patterns to detect dual-sim users
Copyright © CeADAR 2014 18
CeADAR Solution: Idiro
By applying machine learning techniques over many different features (both network and content based) we were able to detect dual-sim vs single-sim behavior with 63% accuracy.
?
Copyright © CeADAR 2014 19
Conclusion
Common industry requirement: identifying people of interest in social networks
CeADAR solution: Combine network based features with content analysis
Approach applied in two different industries on different data sources
Copyright © CeADAR 2014 20
Conclusion
Thanks to our project partners
CeADAR Advanced Analytics research team at UCD
– Dr Gerard Lynch, Dr Guangyu Wu, Dr Jing Su, Hodei Iraola, Dr Oisín Boydell