What is the best algorithm for combining (gluing) notifications of the same type in social services?

Question

Notifications

This refers to notifications about the actions of other users related to the activity previously shown by this user. Notifications have the following attributes: action subject, action object, and action type. Some notifications may not have a subject.

We need to solve the problem that when there are too many notifications, they are difficult to read. And there is a natural desire to group notifications in some understandable and logical way. As you can see, it is possible to combine both by user and by the object of the action, and by the type of activity.

For example, when a user writes a response to my comment, we show it in my last notifications: "mr.X responds to your comment in the publication 'Invites to Habr' on 12/22/2011"

And if we decide to group by the subject of the action, then in addition we show in collapsed “More notifications about mr.X activities”, expanding, for example, into the following list:
“mr.X likes your comment in the post 'Social networks' 12/21/2011”
“mr.X does not like your comment in the post 'Facebook: Open Graph' 12/20/2011"
"mr.X adds you to his contacts on 12/19/2011"

And grouping by action object is possible. Then we show in a collapsed form “More notifications about other comments” and the list will be something like this:
"mr.Y responds to your comment in the 'Social Networks' post 12/21/2011"
"mr.Z responds to your comment in the 'Facebook: Open Graph' post 12/20/2011

" In the form we will show “More notices about the publication 'Invites to Habr'”:
“mr.Y responds to your comment in the publication 'Invites to Habr' 12/21/2011”
“mr.Z likes your comment in the publication 'Invites to Habr' 12/20/2011 »

More complex groupings can involve grouping by several attributes at once, such as subject and action type. Actually a question in what type of grouping to prefer? Are several types of grouping possible at the same time? Are there any good implementation examples?

Answer 1

Clustering algorithms can be used, something like en.wikipedia.org/wiki/K-means. In my opinion, the question is incorrectly posed, not “gluing”, but filtering or combining on some basis.
If you can describe an example of use, then it will be easier to suggest.