I don’t think there needs to be one from each category to meet the need here. We can keep the same number of posts and topics in the digest.
Let’s say I have 7 categories, and in the past week, each has the following number of topics:
| category | topics |
|---|---|
| 1 | 15 |
| 2 | 8 |
| 3 | 3 |
| 4 | 1 |
| 5 | 0 |
| 6 | 0 |
| 7 | 0 |
Depending on how likes are distributed, currently, there’s a high chance that all of the popular topics in the digest are from category 1, from what I’ve seen.
I’d like to see the algorithm pick some topics and posts from categories 2-4 even if they are less liked than those from category 1. Maybe it is changed to something like:
Popular topics: the X most popular topics in this past period. Topics will be from at least Y different categories, if possible.
For example the popular topic algorithm may work as follows:
- pick the top X (default: 5) topics from across all categories.
- count the number of different categories (N). If N is less than Y (default: 3), throw away the bottom Y-N topics.
- pick the top Y-N topics across all categories except those already represented in the list of popular topics.
- repeat until there are Y categories represented in the list of X topics.
Then, even if the top 5 topics are from category 1, I’d only get 3 topics from category 1. The other 2 would come from categories 2-4 (and not from the same one).
This would still work if I have 100 categories.
We already try to select what is relevant to the user in the “new for you” section of the summary/digest email, I think. But I think the digest can help reach those users who are less active on the site that may not be following many categories or topics. I’d like to favor some diversity in the topics shown to these users when we know less about their interests.