​​Threadlets: Building a Vocabulary of Communication Patterns through Interactive Visual Analysis

​​TODO

​​Phase-1: Requirements Collection
  • ​​Interviews with the Red Sift team and literature review
  • ​​Get email thread IDs and merge with Enron database
​​Phase-2: Analysis & Design
  • ​​Generate statistics from the threads (thread metrics), some ideas:
  • ​​Pace of interaction
  • ​​# of people (active/passive)
  • ​​# of people (same organisation/different/personal)
  • ​​# of inclusion/exclusion 
  • ​​sender diversity (# of unique senders / # of all involved)
  • ​​thread length (in terms of duration (e.g., 5 days?) / in terms of message count)
  • ​​Compile the thread statistics in a csv with tIDs and stats as columns
  • ​​Do preliminary (interactive) visual analysis using R/Tableau/Mondrian etc.
  • ​​Pay special attention to FW emails / check if they are frequent
  • ​​Pay special attention to inclusion/exclusion and their combinations
  • ​​Identify interesting threads/people/time in Enron where stuff happened (literature/articles/etc.)
  • ​​Look for topologies/patterns of communication
  • ​​Broadcasting (announcement)
  • ​​Information
  • ​​Ping-pong
  • ​​Forward-sequences
  • ​​Loop
  • ​​Short discussions
  • ​​Bursty/Long discussions
  • ​​Develop pattern/topology mining scripts to streamline the above
  • ​​Branching
​​Phase-3: Development
  • ​​Try to integrate the above with the existing D3 prototype
  • ​​simplification
  • ​​aggregate individuals if they have identical send/receive patterns
  • ​​aggregate messages
  • ​​over time
  • ​​by sender → is it any useful?
  • ​​combine two aggregations
  • ​​ordering
  • ​​individuals/groups
  • ​​temporal: when they first involved in the thread
  • ​​engaged: how many messages they send, then receive
  • ​​messages
  • ​​are there any other ordering methods besides temporal one that make sense?
  • ​​absolute scale time
  • ​​minor design tweaks
  • ​​relaxed hovering (voronoi-like) to highlight an individual
  • ​​bbc type with dash border
  • ​​in each individual, add subtle lines to connect an exclusion message to the following inclusion message — void space doesn’t a good job in indicating the exclusion/inclusion pattern but if the exclusion period is long, it is difficult to see (see john.shelk — does he get included back?) — but it might make the vis look busy
​​Phase-4: Evaluation
  • ​​Identify/evaluate statistics and patterns that deliver value
​​Phase-5: Integrating with Red Sift’s platform