Thank you for you answer! I am sorry, it seems that my feature request was not really clear. So I will subdivide it into several parts.
-
Yes, I understand that “Summarize” button is only valid for rather long topics. The question, however, was about the procedure which is used to select the posts for a summarized view. Simply speaking: How posts are selected to be shown in summarized view of the topic? The most relevant post for this question I was able to found is here Feedback: Improved topic summary mode , however, it mostly describes how to use summarize feature instead of explaining what is under the hood. Digging the source is probably an overkill to get an answer
-
My actual request was to add one more button “Show most liked posts first” along with “Summarize button”. This new button should provide a new view of the topic, exactly as it is promoted on the button: after I press the button I will see the most liked post in the topic right after the topic starter message. It is something similar to StackOverflow manner of sorting the answers, so the most useful answer is shown the first, some valid alternatives can be shown afterward. Such an “SO”-mode can be extremely useful for user self-support section of the forum. Probably, a topic starter should be able to switch on this mode by default for other users in the topic.
-
One more way to provide an automated summary of the topic is to use “user normalized likes” sorting (probably I will describe how it is already implemented, sorry). Let me use Post Importance (PI) metric for each post to select posts for such topic view. It can be evaluated with some simple formula, e.g.
PI = (1 + {number of current mesage likes})*sqrt(Tune+{total likes of the author messages}).
The exact equation can be discussed later, here I assume that the PI is proportional to the number of likes for the post and the reputation of the post author. However, author reputation is less important by default (so I use square root operator) and there is some Tune value which will tune the possibly of a new author giving a better post compared to an old resident. This Tune value can be used to add some artificial intelligence to the forum in the following way. There should be some default Tune value, however, for each display of this “user normailzed likes” topic view of topic the engine should use slightly derivated value, e.g.
{Tune used} = {Tune default} + random(+/- 10% of Tune default)
The real tuning starts than a user likes some post, then such a Tune used value is pushed to a small array (lets call it Tunes best). We should remove the oldest value from Tunes best before adding new value, so Tunes best is simply several (e.g. ten) values of Tune used with a user like. And a Tune default is a simply average value of Tune best. That`s all! The forum “user normalized likes” view of the topic will automatically adapt to the typical user activity of the forum, moreover it will change over the time as the forum community changes (e.g. follow the change of share newcommers\experts and so on). As we have a PI value for each post in the topic we can sort them (and show best posts first) or use 10-30 posts with best PI values in the same manner as current implementation of “Summarize” button.
P.S. Adaptive feature is inspired by my recent experience with adaptive stochastic optimization algorithms, you can find much more information about them in Google.