Skip to end of metadata
Go to start of metadata


  • Larri, Norbert (UPC); Orri (OGL); Peter, Renzo (VUA) ; Alex Averbuch (NEO), Duc (CWI) 


Minutes of the SNA call 13/3/2013

Present: Orri, Larri, Norbert, Duc, Peter, Alex, Renzo

Main topic of discussion where the changes proposed by Norbert/Alex after obtaining feedback from the social media users:

  1. Persons and Users (accounts) are unified in a single entity unless the SIB can generate multiple Accounts for a single Person

            Decision: accepted

      2. There exists a hierarchy of Groups

           Decision: rejected. Until it is clearer that group would play an important role in the workload.

      3. Things do not exist, they are Groups as in Facebook 

           Decision: we can rename Thing to Artist

      4. A Person is a follower of another Person (not necessarily know each other)

           Decision: Orri will come with a proposal for an algorithm extension in S3G2 that would create a followers structure. Peter argues that the followers edges should be power-law distributed, yet the structure be correlated with correlation dimensions. He proposes to create Artist persons, and let people who are fan of the artist follow the artist. Also, follow (a subset of) your friends. So we get extremely connected and sparsely connected followers. Maybe in the middle we should generate “topic authorities”, which are persons associated with cerrain discussion topics. You would follow all authorities on the topics you discuss about.

           Renzo  argues that we should have a power-law for the follower connectivity instead of correlations, but Peter argues that this is independent of each other.

      5. A Post can be "Text" or "Photo" (Photo and Post are merged in a single entity)

           Decision: accepted. We do not need to change S3G2 here, just Discussion post and Photo will now generate the same type of object.

      6. A Post has an optional Location

           Decision: accepted. By point 6, photo Posts will have a location.

      7.  A User likes a Post in some date.

       8.  A Post can depict zero or more Users

       9.  A Post can have a User as recipient (direct message as an email)

       10. A Post can be a copy of another Post (retweet)

       11. A Post can reference another Post

           Decision: no decision yet. Peter argues that if we are going to add twitter-style “following” connections, then we should also generate twitter-style message propagation. These user requests essentially ask for this.

           Norbert tells that he has certain ideas what kind of questions might be asked over the twitter post structure (community discovery). We whould hence create an algorithm that can generate twitter conversations in Posts in such a way that these questions can yield interesting results. So, the decision on this should be tied to proposal 4.

      12. Photo Posts can be grouped in PhotoAlbums

           Decision: accepted. Duc says this is already a feature?

      13. Comments do not belong to Forums, they are replies to Posts or to other Comments

           Decision: accepted.This means forums are going to be removed?


Action list:

  • Orri to propose twitter generation strategies
    • Correlated power-law follower structure
    • (correlated)  post/repost/reference/like message genartion algorithm, in which thought leaders and community structures are hidden.
    • Norbert to provide information on the Twitter queries and analysis conveyed to him by the Social Media users


Norbert input on the use of Social Media

Key factors:

  • Solutions must be transversal, considering content and actions in several contexts. For instance:
    •  Relationships networks: Facebook, Linkedin, email ...
    • Microblogging: Twitter, ...
    • Blogs: Blogspot, Wordpress …
    •  Video: Youtube, ...
    • Wikipedia
    • Online press
    •  Web pages
    • Forums
    • Question & Answering: Quora, Stackoverflow
    • Solutions must create a unique model to integrate information from different sources and different formats
    • Volume is a big issue. Solutions must be scalable.

Query Examples:

  • Data enrichment
    •  Discover a relationship between two users based in their interactions (implicit relationships)
    • Discover which messages are copies from others
    • Identify duplicated users
    • Influence
    • Detection of influence. Some people are influential while others tend to assume some behaviors while under their influence.
    • Measure the distance of influence (the average number of users between them through using inferred relationships).
    • Speed of influence. The time lapse between the influencer shares the document and the user influenced makes an action regarding that document.
    • Roles
      • Discover which users are still active. Compute the churn rate.
      • Categorization of users.
      • Information Propagation
        • Depth of propagation. The number of users sharing the same item of information.
        • Speed of propagation. The average time lapse between users sharing the same item of information.
        • Communities
          • Detection of communities. Groups of people who share similar feelings or tastes.
          • Detection of leaders in communities.



  • No labels