application statistics

[Bryan Clark]

Nicu Buculei wrote:
> Bryan Clark wrote:
>>
>> So we've been collecting some application usage statistics [1] on 
>> mugshot for a little while now and it's starting to reveal some 
>> interesting (and obvious) stuff.  You could look at evolution vs. 
>> thunderbird and firefox vs. epiphany or gossip vs. gaim.  It's a bit 
>> hard to pull enough context into those comparisons to really get down 
>> to the reasons why some are used more than others but it's a good 
>> start so far.
>
> As we are here on the *Desktop* list, I think it telling that the 
> second most used application is the terminal, very close to the first 
> place (indeed, as you said, not enough context to get to the reasons, 
> but it seems one can't effectively use a Fedora desktop without the 
> terminal).
Yep, it's difficult to know if that's just the mugshot audience or a 
standard distribution among everyone.
> I think one data point would be very useful to determine the context: 
> the total number of counted users, for example right now Firefox has 
> 818 users and the terminal 764 users, but without knowing hte total 
> number of users, you have no idea if this is 90% of users or only 40%. 
> And you can't just add the number of, for example, Firefox and 
> Epiphany users and declare this the total number of users.
We could provide that number, but I'm not sure it really explains this 
situation better.  The usage numbers show that among all people sharing 
their application stats firefox is the most used application.  The usage 
count is only taken once a day for a 30 day period such that this number 
won't continue to raise everyday after the 30 day period, it will likely 
level out and only increase slowly as new people share their 
statistics.  Since the count is only looked at once a day the large 
count for firefox users doesn't mean that they are using firefox a lot 
during the day, but that a lot of people are using firefox everyday.

What I think you're getting at is the need to provide better context for 
the ranking system.  Such that if we knew that both firefox and epiphany 
were web browsers (and the system doesn't actually know this right now) 
we could say among "web browsers" this one gets the most usage or 40% 
more usage than the next one.

>
>> So now we're moving this application statistics idea on to a new 
>> phase and are looking for ideas.  During our talk at FUDCon we showed 
>> a couple of mockups [4] of things we were possibly looking at doing.  
>> While we're still touching on most of the different areas shown there 
>> we now have a decent prototype for the statistical application usage 
>> information and it would be great to drive in that direction for a 
>> little while.
>
> Here is one more idea: provide a widget (javascript, please no flash) 
> which I, as an end user, can include in my web page/blog and proudly 
> show to the world *my* application usage.
Interesting idea, it wouldn't be too hard to include that as one of our 
badges.
>
>> We're looking into, as it was suggested on the blog, that we might 
>> provide correlations between usage such that you could see xterm 
>> users are more likely to run xmms.  However there might be other 
>> correlations that would be good to show as well.
>
> Unfortunately the first 3 applications on the list (Firefox, Terminal, 
> Nautilus) are so disproportionately used, that they inevitably appear 
> on the correlation list for each other application,
Nautilus might be a bit of a misrepresentation as we haven't been 
filtering out "the desktop" yet which is nautilus but isn't showing a 
different wm-class.  This should be fixed eventually.  Well it's sign of 
the data, the correlation we show right now simply looks at the current 
application and asks for the other popular applications that someone 
running that also runs.  Since pretty much _everyone_ seems to run 
firefox, terminal, and nautilus that's the correlation we're getting. 

That is slightly different from asking the system, which applications 
are application $X more likely to run.
>
>> Also we're trying to figure out how we can determine related 
>> applications.  Mime types are a bit of a mess to try linking similar 
>> applications together so we might have to ask people to help edit the 
>> information wiki style.  The application categorires are problematic 
>> for this as well.  Right now there doesn't seem to be any existing 
>> information on how thunderbird, evolution, and balsa are all email 
>> clients.
>
> Asking people to manually edit the "related" info is the best 
> solution, otherwise it would be hard to correlate apps like Gimp and 
> Inkscape or Gimp and Fyre.
> Could this application correlation gathered from users be used in some 
> way in a distant future for a better layout of the GNOME applications 
> menu?
We could certainly use the information to better organize applications 
on the desktop.  It might be difficult use dynamically created data in 
the current applications menu which wasn't really designed for 
additional data beyond categories.  However a new type of menu might 
take advantage or in terms of application search via a desktop search 
system the application linking can be very useful.  Other types of data 
like tags or categories on applications might be another way to help 
people find applications.

~ Bryan