ubuntu-phone team mailing list archive

[Rodney Dawes <rodney.dawes@xxxxxxxxxxxxx>]

On Wed, 2015-04-08 at 22:17 +0200, Niklas Wenzel wrote:
> Hi Rodney,
> 
> 
> Thank you very much for taking the time to reply to my post.
> I think I should have explained better what I wanted.
> 
> 
> My point isn't to display a second rating like the unofficial app
> store currently does. I think it is fine to display the arithmetical
> average in the search results.
> 
> However, I'd vote for not using that as the only parameter for
> sorting. In my opinion, the store should also consider the number of
> reviews an application has to calculate its position in the search
> results.
> 
> 
> The heart rating is described the following way in the FAQ of Brian's
> app store:
> 
>         The heart rating is similar to the star rating, but it rewards
>         apps for having more good reivews. An app with more 5 star
>         ratings will have a higher heart rating than an app with only
>         a few 5 star ratings.
>         Heart ratings are calculated from the star rating where a 5
>         star review is 1 heart point, 4 stars = 0.5 hears, 3 stars = 0
>         hears, 2 stars = -0.5 hears, and 1 star = -1 hears.
>         It is possible to have a negative heart rating.
> 
> 
> I mentioned it to give an example of a value that could be used for
> sorting the applications.

OK. I don't think we should add more ways of calculating things to solve
these problems though. I think there are a few distinct problems here to
solve, and sorting is only one of the issues. There's also the social
problem of people trying to use reviews as ransom for features or bug
fixes. That is not how reviews should be, and I think the solution to
that is mostly one of policy, and not technical implementation.

For this I think we need a way for developers to report/flag
inappropriate reviews/ratings on their apps.

> Indeed, it wouldn't change much for apps without any reviews besides
> the fact that they will rank higher than apps with primarily negative
> reviews.
> 
> 
> To give another example of a value that could be used for sorting, we
> could have a look at the way Brian's app store calculates the average
> star ratings:
> 
>         The star rating is the Bayesian average of the star ratings
>         that user give to an app when they leave a review. The
>         Bayesian average allows apps with more reviews to rank better
>         than apps with less reviews. The star rating can be between 1
>         and 5, where a 0 means that there have not been any reviews
>         yet. For more information about the Bayesian average
>         calculation, check out this article.
> 
> 
> As pointed out in the article the description refers to, it makes
> sense to use something else than the arithmetical average.
> 
> The store doesn't have to display that calculated value. It can still
> display the arithmetical average. However, the search algorithm could
> be smarter than it is today.

I think we should probably use the Bayesian method for calculating the
average, always. The mathematical average is not all that useful to know
anyway. If the result of the Bayesian calculation is always an integer
between 0-5, and nothing like 3.27, then it also makes it much easier to
have a nicer display where we just show 5 stars, N of which are filled.
So that would help developers in this situation have their apps more
visible and easy to find, and also would result in a bit better
experience for people who use the phone. We would just need to also do
something for apps with no ratings at all, so they will get weighted
better than apps that have all negative ratings.

Does that make sense?

Attachment: signature.asc
Description: This is a digitally signed message part