This document summarizes research on detecting contexts from behavioral data collected via smartphones and examining how context affects application usage. It describes collecting data from handsets on application usage along with location via cell IDs. Context detection algorithms infer user contexts like home, work, abroad from this data. Analysis shows how application usage varies by detected context, with usage patterns differing at home versus elsewhere. Ongoing work aims to better understand how context influences behaviors like communication channel selection and mobile social networks.
1. Context detection and effects
on behavior
Elisa workshop on “Lifestyle Sensing and Behavioral
Analytics”, June 29th, 2012
Dr. Timo Smura, Dep. of Communications and Networking
(presented work by T. Soikkeli, J. Karikoski, H.-H. Jo, M. Karsai, et al.)
timo.smura@aalto.fi
2. Outline
• Behavioral data collection in Aalto / Comnet
– Multi-point measurements
– Examples of data sets
– Holistic view of service usage
• Ongoing work related to contexts and behavior
– Handset based measurements
– Location detection
– Context detection algorithms
– Context dependence of application and service usage
4. Multi-point measurements
Potential sources of digital behavioral data
Our data sources:
• Handset monitoring
panels + questionnaires
• IP traffic measurements
• Web analytics systems
• Mobile operator
accounting systems
5. Holistic view of service usage
Measurement points vs. service components
Modified from: Smura, Kivi, Töyli 2009
7. Handset-based measurements
Research process and data
• Based on a software client installed to a panel of smartphones
• Collects rich data about handset usage:
– What: Application, bearer
– Where: Base station cell IDs (hashed), WLAN SSID
– When: Time stamps
– How much: Time stamps, amount of generated traffic
• Gives a detailed view of the usage patterns and behavior of panelists
– All applications, also offline and WLAN usage
– Location / context detection
Source: Karikoski 2012
8. Handset-based measurements
Current focus areas
Shares of time and
1. Multi-channel communications usage per context
services 17
24
– Diversification of communications 29
7
channels (phone calls, SMS, 8 8
email, social media services) 9
12 Elsewhere
– Effect of relationship type on 12 Other meaningful
channel selection Office
Home
– Mobile social phonebooks 66 Abroad
53
2. Location and context detection 47
– Context detection algorithms
– Human behavior and time use in 2 3 3
different locations and contexts Share of Share of Share of
total time sessions interaction
– Effects on usage: e.g., sessions, spent (%) (%) time (%)
applications / services
Sources: Karikoski & Soikkeli
10. Context detection algorithms
Simplified version, not utilizing WLAN SSID data
A) Temporal boundaries for user’s trajectory in cells: E) Criteria for assigning other contexts:
B) Duration, i.e., time spent by user in cell c:
C) ”Abroad” context determined by Mobile Network
Code (MNC)
D) For the cells in Finland, more detailed durations:
Sources: Soikkeli 2011, Jo et al. 2012
11. Application usage by context
Exemplary data from a single user during two days
Source: Jo et al. 2012
12. Context dependence of service usage
Fractions and intensities of service usage by context
Source: Jo et al. 2012
13. Conclusions (1/2)
• Aalto / Comnet collects rich data on mobile usage
– Continues a series of measurements since 2005
– Holistic view of mobile devices and services in Finland
• Each measurement methods has its pros and cons
– Level of: Granularity, Coverage, Representativity
– In terms of: Devices, Applications, Networks, Content
• Actors have different views to mobile usage and users
– E.g., Device vendors vs. Operators vs. Content providers
– Increasing value of user data induces competition
• May lead to, e.g., traffic encryption, routing via own gateways
14. Conclusions (2/2)
• Data collected by current smartphones can be used to infer
the context of people
– Then use it as a variable to explain behavior
• By far, research has focused on developing and testing the
technical algorithms for detecting the contexts
– Demonstration of value with descriptive analysis of usage data
• Examples of statistical analyses on the effect of context on
behavior are still rare
– Typically based on survey-based studies and self-reported context
and usage information
– Ongoing / future work: combine existing theories and hypothesis-
based statistical methods to the data collected in smartphone
monitoring panels
15. References
• Soikkeli, T. (2011). The effect of context on smartphone
usage sessions. M.Sc. Thesis.
• Karikoski, J., & Soikkeli, T. (In Press) Contextual usage
patterns in smartphone communication
services, Personal and Ubiquitous Computing.
• H.-H. Jo, M. Karsai, J. Karikoski, and K. Kaski,
Spatiotemporal correlations of handset-based service
usages, arXiv:1204.2169 (2012)
• Smura, T., Kivi, A., & Töyli, J. (2009). A Framework for
Analysing the Usage of Mobile Services, info, vol. 11,
no. 4, pp. 53-67.
16. Useful contacts in Aalto / Comnet
• Project management:
– Prof. Heikki Hämmäinen, Timo Smura
• Researchers:
– Handset-based measurements
• Juuso Karikoski, Tapio Soikkeli
– Mobile network traffic measurements
• Antti Riikonen
– Handset features and evolution
• Timo Smura, Antti Riikonen
– Web analytics –based research
• Timo Smura
– Bayesian Belief Networks –based analytics
• Pekka Kekolahti
• firstname.lastname@aalto.fi
• http://momie.comnet.aalto.fi