Dr. Furtado of Univ. of Fortaleza (Brasil) presented work he is doing in crowd-sourcing to raise the awareness of crime; transparency of crime reports is a particular problem in South America. Using data from
wikicrimes.org, he has done work to identify crime hotspots by clustering events by locality and level of participation (e.g., number of events, number of participants, type of crime, ...). This could be extended for a number of useful applications: opinions on places to <add activity here>, infection outbreaks, suicide attacks...
I found many other talks related to recommendation systems, and I extracted some general principles:- Identify items/fields of interest in an event
- Get a feed of events
- Perform a similarity calculation on events
- Display recommended results to the user
The interesting work is in #3. Many talks addressed inferring indirect links -- when a user doesn't directly score an event or express a likeness to some other actor. Talks presented applications to concert suggestions, navigational pattern similarity, forum postings and text mining for word co-occurrence.
No comments:
Post a Comment