[Data-modeling] Events

[Shawn Simister]

Kirrily Robert wrote:
> Hi all,
>
> A while ago I modeled some "event" types in my user domain and these got promoted into the "Time" domain as /time/event and related types.  Now we're starting to realise that what's there isn't really working for us, so I wanted to open up some discussion about how better to model events and time periods.
>
> As I see it (now), the fundamental building block is just a time period which is simply two date/times, a start and an end.  This can be used for just about anything you might want, eg. the duration of a war, the run of a Broadway play, or the time when a sports event occurs.
>
> Currently the /time/event includes additional properties such as "Location" and "People involved" which I don't think should necessarily be on the event type.  I think that subtypes should do the necessary things there.  For instance, for a war, it might be sensible to have location, but instead of people involved you'd probably just rely on the combatant properties that "military conflict" has.  So I'm thinking that we should probably leave the basic building block (whether we call it time period, event, or whatever) as bare as possible and rely on other types to get specific where necessary.
>
> Further to this, I've got a type called "historic period" (/time/historic_period) which I now think is a big mistake.  There's been some discussion on the boards in the Time domain about instead having something that's more like a generic timeline.  A timeline can contain any kind of events at all, not necessarily related in the way that "historic period" suggests.  For instance you could have a timeline of important events in world history, or book tours, or weather patterns.  The current "historic period" topics would just revert to being more like a "time period" as discussed above.
>
> For added thought, I'd like to point you at an article that Robert sent to me about Events, and about 9/11 and how many events actually occurred when the planes hit the towers.  The link is http://www.npr.org/templates/story/story.php?storyId=15370150
>
> So, thoughts please? 
>
> K.
>   
Kirrily,

Thanks for the link to that article, it was a great read and the book 
also looks promising. I've often thought about how to define events and 
although there are a lot of different scenarios to cover I think it 
should be possible. I like the approach that was taken with Product 
where it almost just acts as a marker type and most of the details are 
delegated to more concrete types.

In my opinion, the primary purpose of the Event type should be to link 
together topics as part of one or more timelines. To do this, all it 
needs are pointers to contributing events and resulting events. The rest 
(times, dates, locations, people) would be left up to other types. There 
is a lot of interesting research[1] in the area of Topic Detection and 
Tracking that looks at how to break down a continuous stream of news 
items into a chain of events. There have also been some attempts to come 
up with very specific types for all major categories of news events[2]. 
Of course, news events aren't the only type of events that are stored in 
Freebase but I think that they are the most commonly understood 
definition of an what constitutes an event. Just my 2 cents.

Shawn

[1] http://portal.acm.org/citation.cfm?id=1321554
[2] http://www.signiform.com/newsextract/newsund.htm