Determinants of health, dimensions of health, positive health and spectrum of...
Activity Profiles in Online Social Media
1. Activity Profiles in Online Social Media
Mohamed Faouzi Atig, Sofia Cassel, Lisa Kaati
Amendra Shrestha
2. Overview
• Trends in social media
• What is an activity profile?
• Experiments
• Future work
3. Online Social Networks
• 1.6 billion social network users
• 64% of Internet users
– spend time, stay in contact, catching up news
• Most popular social network sites
– Facebook, LinkedIn, Twitter
• Examples of user generated content
– text
– images
– multimedia
Statistics from: http://www.statista.com/markets/424/topic/540/social-media-user-generated-content/
4.
5. Behavior in Social Media
• People make use of multiple accounts
• Free and easy to change identity
• Personal information is available on the Internet
– communication pattern
– writing style
– geographical tags
– IP addresses
– web browser history
– search patterns
6. Objectives
• To what extent can we group people with
similar communication patterns?
• Experiment on boards.ie discussion forum
7. Activity Profile
• Users most and least active period
• Chronotypes
– morning type, evening type or neutral
– stable over time
• Use of temporal information from posts
• Divided each day into 6 periods
– information gain
• Activity profiles consist of:
– first activity peak
– sleeping peak
– second activity peak
10. Experiment
• Activity Profile
– create an activity profile for each user
– similar activity pattern
• at least two same activity period
• at least one inactivity period
• Multiple Aliases
– activity profiles
– stylometry and post time analysis
11. Data
• Irish discussion forum boards.ie data
• Available data
– 11 years data
– 50 GB of disk space
– 9 million documents
• Used data
– 2003 year data for activity profile
– 2008 year data for multiple aliases
14. Summary
• Capture activity profiles of Internet users
• Group Internet users with similar behavior
• Activity profiles can be used to detect multiple
aliases
• Minimize analysis time
15. Future Work
• Dynamic methods for activity peaks
• Cross-platform experiments
• Large-scale experiments
Hello everybody, I am Amendra Shrestha, I am here from Uppsala University, Sweden. The title for my today’s talk is ….. This is a quick overview of my today’s talk. I will start my presentation by presenting some statistics about social media and ongoing trends in online community. I will introduce activity profile and will explain it in details. I will explain our experiment set up and data used for experiment. I will present the result of the experiments and at the end I will conclude my presentation with summary and the list of future work.
we have been looking at the possibility to detect users that make use of multiple alias
Easiest way to change identity in online community is to create new account with different information
User can be track down by looking his series of behavior on the Internet
Activity profile is based on the theory of chronotypes which
Chronotypes is the attribute of human being reflecting at what time of day their physical functions is active. Here physical functions mean hormones level, body temperature, eating and sleeping habits. This phenomena is commonly reduced to sleeping habits only.
Activity profile is created by splitting 24 hours of a day into 6 periods.
Dividing hours of day into 4 hour block is more likely to capture the chronotypes of individual rather than considering the activity of user for each hour. We have compared the usefulness of features between 24 hour of day and 6 period of day using information gain which is the entropy based feature selection method. And the periods of day seems to be more useful than hour of day.
Depending upon at what time user is active or inactive in online communication we create activity profile for each individual.
We will take a period
To illustrate that activity profile can be used to capture the behavior of a user we experimented with splitting one user into two separate users. The idea was to check if both users would end up in the same group i.e. with similar activity profile. The two users have similar activity profile if
they have at least two same activity peak and
at least one inactivity peak
We created activity profiles of users and after that did stylometry and time analysis to detect users with multiple aliases.
Stylometry: study of authorship of document purely based on linguistic style exhibits by author
Used baseline features like punctuations, word count, letters, digits and function words
Looking at the point in time when user has created their posts can give us important clue to whether two different aliases refers to one or not.
distribution of users behavior into 150 groups
differences in group
Exemplify with one of the groups
We tried to find out if Activity profile increases the accuracy. Once the experiment was done without activity profile and using activity profile. Comparing users that had same activity profiles gives better result than comparing without same activity profiles.
In this paper we created the notion of activity profile for individual users and a way of to group users with similar activity pattern in social media.
Our experiment shows that activity profile increases the accuracy of detecting users with multiple accounts
Activity Profiling could be the first step for further analysis which will minimize the analysis time
We have used static time range for creating groups, It would be interesting to elaborate time periods and use dynamic methods to define activity and inactivity periods
It would also be interesting to see how activity profiles on different social media platforms relate to each other