Thursday, May 14, 2009

On the structure of event representation

Today I took the day off, and spent the afternoon and early evening in the "Achziv Park" seen above. My third daughter Hadas is in a few days trip to the Western Galilee, and today was a day where the families can join for falafel in the park, so I have travelled there with my fourth daughter Daphna, and spent some quality time with the girls, even played ball.

Earlier in the day, I spent a couple of hours in my favorite coffee shops, starting chapter 5 of the "Event Processing in Action" book. Writing a book is a major commitment, and since I am doing it in my (imaginary) spare time, this is quite a burden on my time. I had a lunch last week with Roy Schulte from Gartner who is also writing a book (together with Mani Chandy), and he also complained that the book writing messes up all his spare time.

I was asked about the scope of the book, well - the book concentrates on pure event processing. There are some are complementary technologies like -- image processing, text analysis, speech recognition, sensor networks, statistical reasoning, machine learning that can be used to automatically generate either events or patterns (in relative small number of applications at this point), as they are really complementary technologies, they are mentioned briefly in the advanced topics chapter, the main stream of the book is about event processing .

After this long introduction , I'll turn to write about some portion of chapter four of the book which we submitted yesterday to the author. Chapter four deals with the representation of events (meta-data). There are currently no standards about representation of events, thus, we have taken ideas from different directions, to form such a model. We are looking at typed events, thus, each event type has some information particular to event type. We distinguish between three types of such information: header, payload, and event to event relations.

  • Header attributes provide information about the event - type, time granularity for the event, and times associated with the event (occurrence time, detection time), event identity and some others.
  • Payload attributes provides information about the event content -- references to entities and other attributes
  • Event to event relation provides information about semantic relations among events.
In subsequent postings I'll discuss each of them in depth. It is late -time to rest.


Arturs said...


I remeber to have read the paper 'Semantic event model and its implication on situation detection', which you are the co-author as well. Have you identified some significant follow-ups to this in terms of semantic event expression, or this is a completely revisited model? And do you plan to include any thoughts from the 'uncertainity' field?

Opher Etzion said...

Hi Artur. The paper you mention is quite an old one, but, it is a derivative of the same model. The follow-ups is about event causality relations that will be described later in the book. About the uncertainty, see some discussion here:
I'll return to discussing uncertainty later in the book.