QtMobility contacts scoring

[Ivan Čukić]

So, there are a few problems regarding scoring contacts from QtMobility

1. How to assign them to activities

As you already know, scoring is saved for a triple (application,
activity, resource).
This is done via desktop events, that are being registered as they
happen. Thanks to that, we know which application did access the
resource in question and in which activity the user was.

For the contacts, since we don't have the ability to alter the
messaging applications of the device, we can only generate the scores
from existing sent and received messages. So we *can't* have the info
about the application nor the current activity.

Solution 1:
- Score the contacts globally, and do the connections between a
contact and an activity manually. Forget about the application.
  This is (for me) bad for one big reason - the whole point of the
scoring mechanism is so that the user doesn't need to do things
manually most of the time. And while the user would get contacts
sorted by 'how much I was in contact with' it wouldn't be per-activity
unless the contact was manually assigned to an activity.

Solution 2:
- Store the events about when an activity was turned on and off. This
would work similarly to the desktop events where we track when a
document was opened/closed.
- Forget about the received mail (the user doesn't choose when
somebody else sends an email to her), and score only based on the sent
ones. We know the timestamps for the sent mail, and we'll know which
activity was active at that moment.

Solution 3: (aka 2++ :) )
- Store the events about when an activity was turned on and off. This
would work similarly to the desktop events where we track when a
document was opened/closed.
- Calculate the main score only based on the sent messages like in S2.
This score will be tied to an activity.
- Calculate a secondary score based on received mail. This score will be global.
- Make a combination of these two scores when sorting the
recommendations. The way we combine the scores would need to be
thoroughly thought about.

----

Lets discuss.

For me, S1 is bad and shouldn't be considered at all.

S2 and S3 have an interesting side-effect - we'll be able to predict /
score the activities (something we talked about before), and we'll be
able to link them to location, time-of-day, etc.

S3 adds a layer of complexity for which I'm not sure that it gets
scores better by an order of magnitude to justify the complexity.

While S3 looks like it would provide better results (more statistical
data used) it might not be the case

Just imagine the user Alice which dumped Bob - Bob sends a lot of mail
to get back together. S3 would put Bob on the top of the list, while
Alice didn't send him a single e-mail since the break-up. She probably
doesn't want Bob to get a high score. :)

 So, for me S2 is the best solution.

Any thoughts from the rest of the team. (we can discuss this in a
telco if needed)

-- 
Cheerio,
Ivan

--
While you were hanging yourself on someone else's words
Dying to believe in what you heard
I was staring straight into the shining sun