Saturday, September 15, 2012

On Google Intelligence Events


I am using Google Analytics (quite infrequently) to view the activity on this Blog, but for those using websites for commercial purposes, or as a social media vehicle, tracking activity on website is a very valuable tools, especially in our metrics-driven universe. The illustration above which seems as a typical sense and respond cycle is taken from an article in seemingly odd location;   the social media sun recently reported on a feature in Google Analytics called "Google Intelligence Events" (although I think this feature is not new).    The article claimed that this is "a simple feature called complex event processing", falling into the "complex" word trap,  and in other cases it asserts that "Google's intelligence events is only a basic use of event processing engine".
In fact, what this tool is capable of doing is issuing alerts of two types:

Automatic alerts:  indicating significant changes of traffic to the website
Custom alerts:     threshold crossing of a certain indication (e.g. the traffic from Singapore was down 20% from previous day).  This is a threshold over all tracked variables (source, demographics, bounce rate etc..). 

As far as event processing is going -- it is indeed limited capabilities, mainly threshold oriented comparing two set of events (based on time).   However, it shows that event processing has got into the web analytics world and there is a potential of doing more in this space.  More on this -later.    

Sunday, September 9, 2012

Event is a relative term

Following recent discussion with Jeff Adkins about the semantics of event,  one of the observations has been that event is a relative term.  First - different observers can describe the same event in different terms, and indeed different observers may not agree about - what really happened, or where it happened, or when it happened, or all the above, kind of "Rashhomon effect".    It may also be relative in the sense that different views may look at different properties of the event,  for example one view will only look at a customer enters the store, another view will also look at demographic properties of the customer -- age group, gender.
The relativism effect can mean that:  event meta-data may have different views with different structure, and event instance may have several corresponding instances according to the observers.  
The question is -- how to consolidate different observers?  this may involve modal  semantics.  I'll write about it at later phase.  

Jeff also recommended me to read a philosophy book called :"The ontology of mind: events, processes and states" by Helen Steward.  I have ordered this book and will write a review about it upon reading it.