-15

In response to the recent problems with the HNQ, I think we need to rethink how the Hot Network Questions are ranked. As @E.P. pointed out, in their answer, one of the biggest, if not the biggest, problem with the list is that it is filled with clickbait.

Featuring clickbait on the HNQ just attracts new users that enjoy clickbait. These shouldn't be the people we are trying to attract. We want users that actually care about the site and will uphold the quality standards. This goes along with what @NicolBolas said in their answer. Showing clickbait just makes other users think using clickbait to get upvotes and views is OK.

What we really need to do is sort out which questions are clickbait and which aren't. That way we get good questions that encourage people to join the site and leave an insightful, helpful answer, instead of getting people who take a look at it and then leave or join just to post a spam answer. The kinds of questions that should be on the HNQ should be in need of answers, with a sufficient number of answers having a high score.

Unfortunately, click-bait-y answers get a lot of upvotes by virtue of having more views from more people clicking it, so votes is not an accurate indicator of whether the question is a good one (non-clickbait). The best way to tell whether the answer is clickbait might be based on the proportion of upvotes to views.

To combat clickbait, we should explore the possibility of a views:upvotes ratio to determine if a question is a good one. It would be simple enough to implement because it uses metrics we already have without needing checking for clickbait keywords.

Question:

What is the ratio of views:upvotes in popular clickbait questions compared to the ratio for a non-clickbait popular questions? Popular meaning these questions meet the objectives for a HNQ. These proportions will be important in determining the relationship between the two metrics, and will tell us if it is possible to determine what is and isn't clickbait.

This question asks for a ratio backed up by hard evidence. Please do not provide "Well I guess X because Y" answers. I know the answer will be a little subjective based on how you define clickbait, and it may require some digging and searching, but I think the results will still be accurate enough to be of use.

If you don't agree with this question

A lot if users have pointed out that, in their opinion, cickbait is not an issue with the HNQ, or that the views:upvotes ratio won't be helpful. Thank you for the feedback, but I am not asking about feasibility here.

For the purposes of this question, I do not want to argue about whether this metric will be helpful/accurate. If you think there is a problem with gathering the data I described, I am open to that. If you think the HNQ has a different problem, please discuss this on the HNQ post I linked to. Please do not complain about this question because you think the metric will be unhelpful. Also try to think about that if you are planning on downvoting or VTC-ing. All I am asking for is this ratio, my analysis of the HNQ issues is only provided as a background for the question.

If you are downvoting, it would help if you could please leave an explanation of why in the comments so I can improve this post. I have tried to edit this post multiple times, but it seems that there is still a problem.

  • 3
    Why is this even a problem? If the title is clickbait then it should probably be edited. – Laurel Oct 19 '18 at 2:48
  • 1
    I can't say I agree with your last sentence. What constitutes "clickbait" is hardly "a little subjective". And yet, it's the very basis of the premise of your query. The person who decides what is a "clickbait" title is the one who will decide what the "hard evidence" looks like. And therefore, you will end up proving nothing. And personally, I disagree with EP's post; it's not "clickbait" that is selected for; it's controversy. These are not the same thing. – Nicol Bolas Oct 19 '18 at 2:50
  • 3
    The upvotes:views ratio isn't indicative of clickbait. You should be interested in the displays:clicks ratio instead. Clickbait ends up on the side of webpage windows, and the title makes people choose to look at what's inside. When the viewers are only the active site members who know the subject, a high upvotes:views ratio is frequently a symbol of quality content and active voters, both of which are characteristics of a healthy site. – user392547 Oct 19 '18 at 4:32
  • 1
    Besides, most good questions has a lot more views than upvotes because most vewers don't have SE account. – user202729 Oct 19 '18 at 5:44
  • @user202729 Those questions are not necessarily good: it could just be that they're very common stuff asked by the kind of people who usually don't have SE accounts with voting privileges. We discussed this on Physics Meta recently (here) – user392547 Oct 19 '18 at 8:30
  • @Chair "Good enough". (to not be closed) – user202729 Oct 19 '18 at 8:45
  • Clickbait is click-bait. Period. It grabs clicks. If it's controversial content behind the trash title, it may get more votes. If it's neutral/boring content behind the title, there may be fewer votes. There's no statistical characterization possible. Even using ratios for displays-clicks isn't reliable due to the variables, like the nature of other content on the page. – user392547 Oct 19 '18 at 10:50
  • 1
    Updated the question, thank you everyone for the feedback. – John Locke Oct 19 '18 at 14:55
  • Trying to dictate how users should vote isn't going to help you. All it's going to do is gather a more negative reception. – fbueckert Oct 19 '18 at 16:01
  • @fbueckert I updated the question so that it's not asking for votes, just pointing out that the aim of the question is not to get an answer about feasibility – John Locke Oct 19 '18 at 22:46
  • For anyone downvoting, can you leave a comment as to why so I can improve this post? – John Locke Oct 20 '18 at 12:34
  • @JohnLocke My downvote is because I disagree with the assumptions that (1) There's a solid quantification for click-bait. It's more like I know it when I see it, and stats aren't reliable here (2) Views:upvotes ratios are related to clickbait. I already expressed that I strongly believe that the votes don't mean much, and it's all about displays:clicks, hence the name 'clickbait'. – user392547 Oct 22 '18 at 12:37
  • @Chair Unfortunatley, I don't see a way to make clickbait an objective term, and so I have to go off of what the answerer thinks is clickbait. As for the ratio, I don't think there is a way to tell how many times a question title is seen. Even if there is a way to tell, what about when a user refreshes the page but isn't looking at the question list? What about good questions that just attract a lot of clicks even though they have a title that is on-target? – John Locke Oct 22 '18 at 16:09
3

I have no idea how you would quantify click-bait. I can query for questions that have a lot of votes in the early days of their existence. Those questions can then serve as data points for a views/upvotes ratio.

The following SEDE query does that for IPS.se. I checked with some mods/regulars of IPS and they confirm the questions in that result set have been hot and some have click-baity titles.

select top 100 
       postid as [Post Link]
     , datediff(d, convert(date,p.creationdate), v.creationdate)  days
     , count(*) [up_down]
     , (datediff(d, convert(date,p.creationdate), v.creationdate) + 1.0) -- just a day
       / 
       count(*) [days/votes ratio]
     , sum(case when votetypeid = 2 then 1 else 0 end) upvotes
     , sum(case when votetypeid = 3 then 1 else 0 end) downvotes
     , p.viewcount
     , p.viewcount / sum(case when votetypeid = 2 then 1 else 0 end) [views/upvotes]
from votes v
inner join posts p on p.id = v.postid 
where votetypeid in (2,3)
and p.posttypeid = 1 -- Q
group by postid
        , datediff(d, convert(date,p.creationdate), v.creationdate)
        , p.viewcount
having  datediff(d, convert(date,p.creationdate), v.creationdate) < 2
order by (datediff(d, convert(date,p.creationdate), v.creationdate) + 1.0) / 
       count(*) asc

When run today the query returns this result for IPS.se:

IPS.se hot questions and the viewcount / upvote ratio

I leave it to you to analyse this further as it is not clear to me what you want with this ratio's. Nor is it clear if you expect network-wide results, which would require major additions to this query.

Keep in mind SEDE is only updated once a week, on Sunday 03:00 UTC. Take a look at the awesome tutorial and/or say Hi in chat

  • Thanks, I'll try this out and try to figure out what the ratio looks like for the two categories. From your screenshot, the days/votes ratio seems to be about 0.02, while the views/upvoted varies a bit. – John Locke Oct 21 '18 at 16:53
4

According to this monstrous SEDE query (details below), based on all questions (and the HNQ list) between August 4th and October 13th this year, the Views:Upvotes ratio for HNQs is higher than for the average question on that site. This is not surprising; many people browsing the HNQ don't have an account on that site (or no association bonus) and can't vote; non-hot questions get relatively more traffic from users with a dedicated interest in that community and having the privilege to vote. There are a couple of exceptions (Poker, Monero, maybe a few others). Feel free to fork the query to analyze other statistics you can obtain from the data.

These are the sites with the highest (average) Views:Upvotes ratio for Hot Network Questions:

enter image description here

The list of hot questions (the IDs in the top half of the query) were obtained by scraping the Hot Network Questions list every 10 minutes, as described here.

  • Thanks for the answer, that query must have taken forever to write! Is the V:D in the last two columns for votes:displays? – John Locke Oct 23 '18 at 11:09
  • It's for downvotes. So those ratios are much higher. – Glorfindel Oct 23 '18 at 11:13
  • So in the ones where the last colum was blank had no downvotes on HNQ questions? – John Locke Oct 23 '18 at 11:16
  • Yes, that's correct. Often, those sites only had a single HNQ in the two months I analyzed. – Glorfindel Oct 23 '18 at 11:17

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .