We have a system for allowing users to write their own posts, recommend other people's posts, and theoretically bubble the best ones up to a wider audience. How exactly the system works is not documented, users don't seem to know how it works, and in my opinion, it works poorly. I believe we're missing many of the best posts on dailykos, and that a better system would bring more of them to our attention. Separately, whatever system we have should be documented, so we know how it works.
Below, I propose a recommendation system that I think would best achieve the goal of promoting the best posts for more of us to see, while remaining as simple to use as the current system. And, I call for the recommend system to be documented.
No Documentation
I haven't been able to find any real documentation of how the recommended diaries feature works here. I've asked a few times on open threads, but the few responses I get tend to be other people saying they don't know how it works, either. Some things are common knowledge, such as the fact that you can only recommend a diary in the first 24 hours after it's posted, but they're not written down anywhere that I can see.
Heck, even the basic fact that a "diary" was a post made by a user, isn't documented. It took me over a year of cursory dailykos use before I figured that one out. Don't laugh, it may be obvious to those of you who started with scoop blogs, but many of us have been using other blog software for longer. Diaryland was one of the first blogging hosts on the net (they existed at least as early as 1999), and there the term "diary" is synonymous with "blog" (the term "blog" didn't exist yet, or wasn't in wide usage, when they began). Many people have learned that "blog", "web journal", and "online diary" are synonyms, so our natural assumption is that a user's "diary" is the collection of all of their posts. For at least a year, I thought "recommend diary" was a link that I'd use to recommend a user's personal blog. I didn't know how to recommend an individual post, even though I knew that was possible and searched for the feature. But there was no documentation
It's a little less confusing now, because the recommend button just says "Recommend", and the link to post says "New Diary Entry", but people still talk about individual posts as "diaries" and say "recommend this diary", etc. Over on Blue Mass Group, where I'm a frequent front page poster, we had a discussion about this shortly after moving to SoapBlox (which is very similar to Scoop), and decided to rename "diaries" to "user posts".
Ahem. That was a tangent. My point is, the recommendation system, like much else on dailykos, is not documented. I don't think it works very well, and I'd like to recommend an alternative, but I think it would be very helpful to find out how the current system actually works. For example:
- How many recommendations does it take for a post to become "recommended"? Is it an absolute number or does it relate to how many recommendations other posts are getting?
- Once it is in the recommended box, how long does it stay there? Is there a scheduled time, and if so, is it based on the original posting time, or the time when it first became recommended? Or, is it until some other post bumps it off?
- Does the timing of recommendations have an effect? I think I've seen evidence of that - sometimes 30 recommendations in the first hour or two are enough to recommend a post, but a post with 60 or more recommendations 10 hours after it got posted my not be "recommended". But I don't know if that's an artifact of what was happening with other posts at those particular times.
- How many posts are in the "recommended diaries" box at any given time? Is that something the software decides independently for each post, or does the number of currently recommended posts affect whether new posts join the box, and/or is it affected by new posts joining the box?
It Doesn't Work Well
The goal of the recommendation system is to find the best diaries, the ones most of us would want to see but otherwise wouldn't, and bring them to our attention. The simple idea is, if more readers recommend a diary, it's more likely that other readers will want to see it. I used to think no more of it than that.
But then I wrote this post, which got around 50 recommendations, and 41 comments. An identical copy on MyDD got frontpaged, and Blue Mass Group posted a link to it at the top of the blog as recommended reading, for a few months - even though it's not on a topic either of them usually cover. It is a topic dailykos covers, and it's exactly the kind of post that, if it had been written by someone else, I'd have wanted to see. But it never got to the "recommended diaries" box. And that got me thinking: what else am I missing that's this level of quality?
After that, I thought about the flaws in the recommendation system, and in my opinion, the big ones are:
- 24 hours is too short a time. If a good post doesn't get noticed right away, sometimes it gets noticed through publicity elsewhere, on other blogs or email lists. But it's rare that all of that builds up quickly enough in the first 24 hours. It could take half a day for another blogger to post about it, another half a day for one of their regular readers to see the post and send the link to a relevant email list, another full day for most of the readers of that list to see it. By limiting recommendations to 24 hours, we're measuring speed more than quality, sometimes.
- The circular nature of recommending: Usually, we don't see posts until after they're in the recommended box. Then, if we think they're worthy, we recommend them. If the system uses any sort of comparison between number of recommendations to decide between posts, this opens up an often unbridgeable gap between the post that just barely made "recommended" and the post that has almost as many recommends but didn't quite. The former post suddenly racks up many more recommendations and suddenly they're not close anymore, but the difference has nothing to do with their quality.
- More generally, absolute numbers of recommendations have diluted meaning. They tell us something about whether people think the post is good, mixed in with telling us something about how many people have seen the post, but they make it hard to separate the two factors.
A Better System
Let's base the decision of whether to highlight a post for the community, not on the number of recommendations, but on its recommendation percentage. Further, let's give them a bit more time to percolate through the blogosphere, before cutting off the voting. Let's allow people to view more posts than fit in the box on the sidebar, sorted by quality, if they want to delve further. And let's document clearly how it all works.
Here's what I propose:
- Keep track of how many users have seen a post, as well as which of them have recommended it. Display three pieces of information in the "Recommend" category:
- How many people have seen this post (a simple integer)
- Who has recommended it (a list of usernames, as we have now)
- Recommendation Percentage = number of recommenders / how many people have seen it
- Set a default recommendation period of 72 hours during which a new post can get recommendations.
- Set a baseline: the number of people a post has been seen by, that qualifies it for being on the recommended list. This number should be low, somewhere in the 6-12 range. A post that has been seen by fewer than this number of logged in users is not yet eligible to have a recommendation percentage and for the rest of what I describe below. A post is "qualified" if it has met the baseline, and is still in its recommendation period.
- Provide a new view of user posts, called "recommended diaries", that shows all currently qualified posts, sorted in order by recommendation percentage, from highest to lowest.
- Set a threshold percentage: a fairly high percentage, such as 40% or 50%. Any post that has a recommendation percentage above the threshold, is qualified, and is within the initial 72 hour recommendation period, enters the "Recommended Diaries" box on the front page sidebar.
- Set a minimum size for the front page "Recommended Diaries" box. If there are fewer than this many posts currently above the threshold, fill out the box with the next few posts from the recommended view, in decreasing order by recommendation percentage. However, the box may be longer than the minimum, if there are more posts above the threshold, since they're always in the box.
- Any post that reaches the threshold immediately has its recommendation period reset to 24 hours from that moment. It no longer matters when the post was first made, only when it first reached threshold. That means that if it reaches threshold in the first two days, its recommendation period will actually be reduced. It will spend 24 hours highly visible, then become unqualified, and drop out to make room for new posts. If, on the other hand, a post reaches threshold on the 3rd day of its default period, this will cause its recommendation period to be extended, so that it can still get a full day on the sidebar.
[ Note: the individual variables for baseline, threshold, default recommendation period, etc. can be twiddled, and this system can still work. So there are two aspects to this proposal: the structure, and the numbers to fill it out with. I do think it important that the default recommendation period be significantly longer than the post-threshold recommendation period, to make room for high quality older posts by pushing the ones that have already "made it" off after they've had enough time. ]