“Attention Profile” and “Favorite Authors”

In the profile of the most active users of Habr, a block appeared reflecting a person’s interest in certain authors of publications . The more nickname - the more interest (just like in a standard tag cloud).


Simultaneously with this remarkable thing, the “ Profile of Attention ” block has changed. In addition to the shortened name (before it was so long that I could not even remember it), the construction algorithm was improved.

The process of working on these two blocks turned out to be very interesting, and, below, I would like to share some interesting observations with the public.

First of all

I noticed such a tendency that rare users in the attention cloud are missing Apple , Google and Habrahabr are the most common tags on Habré. I wondered: what will happen to the cloud if all the tags are equalized? As a result, I switched from a quantitative characteristic of “interestingness” to a frequency response, adding to the formula a dependence on the tag weight on the site.
')
What did it do? The attention profile has not changed much, but less common tags have appeared in it. For example, I have a collective intelligence in my profile, and Aist has high loads . And it reflects reality.

By the way, the same principle was applied for “ Favorite Authors ”, only as a weight I took the number of publications.

Secondly

I have recorded such cases when certain tags were as loved as they were non-loved. This noise was decided to filter out. Prior to this, only positive user ratings were taken into account, and now negative ones were taken into account.

Thirdly

I had the idea to take into account the tags of posts written by the user. What was my surprise when it turned out that people are not interested in reading about what they themselves write . People write on topics from their area of ​​expertise, but they are not interested in reading about what they already know. Something like this. This is not an absolute law, it is a statistical regularity, from which there are many exceptions. For example, XaocCPS is the only user I found that has both clouds turned out to be almost identical :)

Waiting for feedback !

PS By the way, besides the positive attention, this time it was possible to build a negative attention profile with a sufficiently high quality reflecting what the user does not like. There is a proposal to make this information available, but, for obvious reasons, only in its own profile.