We use the topics assigned to each of the posts to cluster posts into pages.
Options:
Top-down clustering (e.g. k-means)
Bottom-up clustering
There are methods to estimate a good number of clusters. Since we want to use the clusters to make pages from them we could set a range (5-20 pages) and then find a good number of clusters for the data we have.
Once clusters have been established, we sort posts in each cluster by a combination of the post score (from upvotes&time) and the posts' proximity to the cluster center
Since this leaves the possibility open, that posts with a high score are not displayed if they are very far off from the cluster center we should have in mind that we may need a 'breaking news' section. This would contain interesting (high score) posts that don't fit any of the clusters.
Future: Let users assign a tag to all posts in a cluster/page. The tags would then form additional dimensions to consider in clustering.
We use the topics assigned to each of the posts to cluster posts into pages.
Options: