ISDA 2011 Cordoba

  • 436 views
Uploaded on

 

More in: Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads

Views

Total Views
436
On Slideshare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
0
Comments
0
Likes
0

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. INTRO
    • Social networking is generating an incredible amount of information that is sometimes difficult for users to process
    • Social networking is now supported by mobile devices. The analysis of on-the-move interactions in social networks is now gaining further attention from researchers
    • Several links, activities, and recommendations are proposed by networked friends every hour, which together are nearly impossible to manage
    • Therefore there is a need to filter and make accessible such information to users
    • This is the motivation behind developing a mobile recommender that exploits social network information
  • 2. PROPOSAL
    • We propose a Social Mobile Activity Recommender that integrates Facebook social network mobile data and sensor data to propose activities (concerts, ongoings, semiars, etc) to the mobile users
    • The recommendations are completely calculated in the mobile device with an embedded data mining component
    • Endowing mobiles with the capability to generate data mining models it may be possible to create a new smart world!
  • 3. SOcial Mobile Activity Recommender (SOMAR)
    • SOMAR provides recommendations on the basis of the users social relationships
    • Hypothesis: Social networks users tend to have social interactions only with a small group of their social network friends
    • Information sources:
      • mobile data in terms of call history or contacts
      • sensor data in terms of sensor location
      • Facebook data
  • 4. SOMAR (1)
    • SOMAR can be divided into: i) generation of the recommender and ii) application of the recommender.
  • 5. Generation of the recommender
    • The recommendations are based on the user social realtionships, for this reason we present the Social Graph.
    • The aim the Social Graph is:
      • to represent the relationships of a user with the users friends or subgroups of friends
      • show how frequently the user interacts with them
    • The characteristics of the Social graph are:
      • the nodes represent friends or groups of friends
      • the graph is based on mutual friendship and the quantity of relationships among users
  • 6. Computing the Social Graph
    • The graph computation in the mobile goes on in three steps:
    • mutual friend computation
    • user clustering
    • affinity degree calculation
  • 7. Computing the Social Graph (1)
    • Mutual Friend Computation
    • Information required:
      • the list of friends of the user
      • the list of friends of users’s friendes.
    • For each pair of friends, the number of mutual friends is calculated
    • Output: a matrix M (n*n) of mutual friendship, where n is the numer of the user’s friends.
  • 8. Computing the Social Graph (2)
    • User Clustering
    • The aim is to group the users, the measure of affinity is given by the number of mutual friends
    • For this aim we propose to use the hierarchical clustering taking as input the matrix of mutual friendship M
    • The output is a number of clusters which represent the connections among friends, independently from the Root user being analyzed
    • In fact, a greater number of mutual friends shared by two users results in a higher probability that they will be in the same clusters
  • 9. Computing the Social Graph (3)
    • Affinity Degree Calculation
    • In this step we determine the degree of the affinity between the Root and the clustered friends
    • The edges connecting nodes are labeled with a value based on the number of relationships among users, which in turn establishes the degree of affinity between those nodes
    • The relationships are defined by: comments, likes, messages etc.
    • The final output is a weighed graph representing the social relations of a user.
  • 10. Application of the Recommender
  • 11. ACTIVITY RECOMMENDATION
    • Activity recommendation requires the following online tasks:
  • 12. ACTIVITY RECOMMENDATION
    • Activity recommendation requires the following online tasks:
    • Data Integration:
      • We will provide a module that integrates and prepares data from sensors, the phone and Facebook
  • 13. ACTIVITY RECOMMENDATION
    • Activity recommendation requires the following online tasks:
    • Data Integration:
      • We will provide a module that integrates and prepares data from sensors, the phone and Facebook
    • Activity Recognition:
      • Activities are easily detectable because they are published by the event Facebook application
  • 14. ACTIVITY RECOMMENDATION
    • Activity Matching
      • The social graph is used to select activities among those detected by the Activity Recognizer
      • The weights between the nodes in the social graph represents how socially close is the user proposing the activities to the target (Root node)
      • This weight is then compared with a threshold based on the social graphs structure and degree, and it can be modified by the user
      • Finally, activities are flaged considering, for example, location and current user activities
  • 15. CASE STUDY
    • In this paper we only focus in the analysis of the feasibility and the performance of generating the social graph in the mobile, in particular:
      • Complexity of step 1
      • Clustering behaviour depending on the number of friends
      • Affinity degree calculation of the basis of the number of clusters
    • We carried out the experiments in a system with a 2.16 GHz Core 2 processor and 2.5GB 667 MHz DDR2 SDRAM memory
  • 16. CASE STUDY
    • Data about the target user include the friend list, public wall and all information shared by friends in the friend list
    • In particular, the user in this case study has:
      • A friend list that contains 149 friends
      • Data comprise 1379 posts generated over two years
      • Data in post include links, photos, likes,comments and tags that involve two or more users
  • 17. RESULTS (performance analysis)
    • STEP 1 has complexity O(n^m), where n is the number of the Root’s friends and m is the maximum number of friends connected to each of the Root’s friend.
    • STEP 2: CPU time grows up to 13 seconds when the number of friends is maximum
  • 18. RESULTS (performance analysis) (2)
    • STEP 3 has a linear complexity of O(k), where k is the number of identified clusters.
  • 19. CONCLUSIONS
    • SOMAR analyses and compute information locally in the device
    • Local computation provides several advantages:
      • It's guaranteeing the privacy of sensitive information
      • It's providing personalized and context aware recommendations
    • The main drawback is relates to resource comsuption and performance
  • 20. CONCLUSIONS
    • The main cost is related to computation of the social graph
    • It depends on the number of friends of the user
    • Nevertheless, for the user (real) used in the experiments having 149 friends has been shown the feasibility of the method
    • For the rest of the steps of the social graph generation, according to the experiments, they are feasible for the maximum number of friends allowed on Facebook