|
Recently TL introduced a 'featured blogs' section in an attempt to highlight the better blogs produced by the community. This is a really good idea, with the implicit idea that some blogs are better than others, and that ranking by this idea of value was possible and desirable. I study Information Retrieval (specifically recommendation, which ranks things without a specific query) so this new feature was interesting to me, not just from the stand-point of wanting to see good blogs, but also wanting to find ways to rank blogs well. I voiced some concerns in the initial announcement thread, since then I've loved reading the suggested blogs, and also did some tinkering and came up with something that might be of interest. I call it 'Team Liquid Blogs of the Moment'. You can subscribe to both of the lists I generate with RSS if you are into that sort of thing 
![[image loading]](http://i48.tinypic.com/25hobjm.png)
Ranking items based on quality is really hard. It's quite complex, whenever anyone talks about ranking by quality or relevance, the factor traditionally used to rank, both are relative measures. What is of interest to me might be of no interest to you. We (in theory) have a straight-forward mechanism for indicating how interesting a blog is: the ratings. Since the assumption is the quality of the blog post should dictate its place on the list this seems reasonable. But based on my experience blogging (perhaps not an indicative sample) rating is rare in comparison to views, and this creates a problem of sparse data to base ranking on (It also creates a problem of malicious users deciding to one-star having a greater impact than they should). Like voting, it's not nearly as good if only a few participate. This is compounded if ratings aren't used well, which Evan Miller talks about in an article called How Not to Sort by Average Rank. This was why I recommended that future algorithm improvements should take into account factors not limited to ratings, as things like views and comments give more implicit data to work with.
In its current state featured blogs look like they are ranked based on average rating of the author. This may be (in fact I'm willing to bet it is) a simplification, but the end result features blogs based on some shaky principles. A small group will be auto-featured based on prior performance, regardless of what that blog says, and users who write interesting content will have to spend a long time doing it before they get into that group. This, at least for me, is similar to the RSS set-up, I am subscribed to the highly-rated bloggers I want to read.While the blogs that have been featured are great this rating-focused approach could be seen as bad (it's not necessarily representative of the current state of blogs, it funnels views into blogs by users who are already highly praised, people who write good content and aren't already well known won't get featured). It is open to abuse too, as users high on the rankings have a 'veto blog', i.e. absolutely anything Trozz blogs about right now will be in the featured blogs section, and will knock out the lowest ranking entry.
So, while I believe it is not the intent of the section, the question becomes "Should we have a special group of bloggers who are 'featured'?". This could be an entirely valid way to form a featured blog list, as could manual curation, and it warrants discussion. Either way a list of great blogs formed based on shaky principles sort of grinds my gears, which i suppose is why I research this stuff!
Another way to highlight specific blogs is to potentially weight for content, so for example we hide girl blogs and promote terms we think will be associated with good blogs, both of which could create problems. We could miss "[Girl Blog] Video of me being shot down by every single Iron Lady contender" or some similar interesting blog, but more concretely this can be gamed, so we'd end up with fake blogs like:
[Random malicious blogger]: I beat Day[9] in Starcraft, here are the games, commentated by Klazart
(I would love to see that blog)
As you can see it's not a trivial task to try and pinpoint good blogs, but I came up with a couple of ranking methods that I've been playing with to find good blogs. Note that for 'Blogs of the Moment' there are two methods; one that favours blogs that encourage discussion (similar to a good ratio of views to comments, allowing for interesting blogs with a low number of views to appear) and one that tries to offer a balanced list of interesting content. I only scrape the blog front-page for this so can only use what you see there, views and comments. I don't have access to ratings so I don't use them, but I think without them there are some pretty interesting results.
Pros
Highlights the blogs that currently seem to be generating the most interest, regardless of author current views
Offers a very straightforward mechanism for rewarding a blog; comment!
Reflects the current state of blogs
Very hard to game
Cons
Only works on first page
Doesn't use ratings
I'm sure it could be refined further
So I've rambled and made assumptions (if they are wrong then we have two additional ways to find blogs to read at worst), I hope I've been clear on my concerns despite the interesting blogs I've seen through the featured blog section.
Edit: added image
|
ultimatehurl for president yo
|
On April 27 2012 04:15 7mk wrote:
ultimatehurl for president yo
of the world.
Nice write-up, i've been thinking of the featured blog area for a bit, although I don't feel personally afflicted probably because I don't write many blogs.
|
I hide all the featured blogs. If someone thinks they're so special they deserve to go in a special blog section, then they're probably not a very good person, and probably write very bad blogs.
Ah it's based on previous blog rating. And not one is interesting. I take that back. Haji's and Sirjolt's are interesting.
Learning what makes a blog featured changes things.
|
I like your thoughts on the subject, fitting for someone interested in IR.
|
Estonia4644 Posts
|
Thanks guys, I rarely write blogs myself, but I read a lot and it irks me to think that I might be missing good ones because the author wrote something else that wasn't liked.
I love your work fusefuse, SirJolt showed it off with some pride prior to posting his blog
|
In its current state featured blogs look like they are ranked based on average rating of the author. This may be (in fact I'm willing to bet it is) a simplification, but the end result features blogs based on some shaky principles.
It's not a simplification, it's just wrong. Compare the featured blog list (http://www.teamliquid.net/blogs/?featured=1) with the old one (http://www.teamliquid.net/blogs/?featured=0). Whoever came up with this system just went through the blogs and picked a number of them to be featured.
The fact that they are sorted by rating is just incidental. You can sort them by date if you wanted to (http://www.teamliquid.net/blogs/index.php?sort=lastpost&featured=0).
The system is basically "this is a list of good blogs," not "these are the best blogs at this current moment."
Edit: actually you might be talking about the sidebar thing, which i don't understand. nevermind me
|
On April 27 2012 08:34 Gheed wrote:Show nested quote +In its current state featured blogs look like they are ranked based on average rating of the author. This may be (in fact I'm willing to bet it is) a simplification, but the end result features blogs based on some shaky principles. It's not a simplification, it's just wrong. Compare the featured blog list (http://www.teamliquid.net/blogs/?featured=1) with the old one (http://www.teamliquid.net/blogs/?featured=0). Whoever came up with this system just went through the blogs and picked a number of them to be featured. The fact that they are sorted by rating is just incidental. You can sort them by date if you wanted to (http://www.teamliquid.net/blogs/index.php?sort=lastpost&featured=0). The system is basically "this is a list of good blogs," not "these are the best blogs at this current moment."
The clear difference in the two lists is that people high on the old list who haven't blogged in a while have been removed. As noted I've no access to the backend, only the observed result. If it is a hand-picked group the 'potentially good blogger vs. actually good blog post' ambiguity is is still there, with little room for new blood to get in, which seems a shame. This tool is more to highlight good content in blogposts, not good content leading up to those blogposts.
Edit: The sidebar looks like it's the first page of featured (which you have to be highly-rated/hand-picked to get onto) ordered by date. I had only been looking at the sidebar, hadn't clicked on 'Featured blogs' more than once.
|
Estonia4644 Posts
On April 27 2012 08:27 UltimateHurl wrote:I love your work fusefuse, SirJolt showed it off with some pride prior to posting his blog
I feel like such an ass for flaunting here then ;_;
MEH STILL RELEVANT
|
On April 27 2012 09:55 fusefuse wrote:Show nested quote +On April 27 2012 08:27 UltimateHurl wrote:I love your work fusefuse, SirJolt showed it off with some pride prior to posting his blog I feel like such an ass for flaunting here then ;_; MEH STILL RELEVANT
It's totally worth flaunting man, good job
|
brilliant write up of my exact thoughts on this subject.
|
Can you explain in detail on how the logarithms themselves work? View-to-comment ratio seems fairly straight forward, but balance-of-topic doesn't. Do you rely on keywords for balance-of-topic? How do you generate categories for individual blogs to ensure a proper balance?
|
Starparty: Thanks man,hope the tool helps you see blogs of interest
Primadog: Both algorithms are a little more complicated than that, but while the first revolves around comment-to-vote ratio the second takes into account the volume of views and comments.
The first algorithm works on the assumption that a good blogpost generates comments from a good number of the people who see it.
The second algorithm assumes that a good blogpost gets a lot of comments but might have a lot of views too, so if a blog has been interesting enough to get a lot of views compared to comments, but still generates a conversation it is worthwhile. Evan Miller's article there explains why just sorting by rating (in the two obvious ways) is bad and explains an algorithm I did some heavy adaptation to for my tool.
|
I'm glad they are highlighting the higher achievers among us, now I always know where I can find Gheed's blogs
|
It certainly is great to have a way to highlight good blogs, and you'll be happy to hear every Gheed blogpost has been on my Blogs of the Moment site, but currently it's almost impossible for new people to be featured is all.
|
|
|
|
|
|
EPT![[image loading]](http://dl.dropbox.com/u/42425448/TL/TLGBL_crop_400.png)
