Research talk:Evaluating RelatedArticles recommendations

Monthly pageviews

Are you sure a list focused on one month's events will be representative? --Nemo 07:59, 21 March 2017 (UTC)

Pretty sure. The main point of using high-traffic Feb articles was that the raters would be very likely to be familiar with the topic (which wasn't the case for articles in set 1, the low traffic condition). This was meant to reflect actual use of the feature more closely than set 1; presumably someone who makes it all the way to the bottom of the article on mobile is interested in the topic, and has some familiarity—at very least, they've read most of the source article by that point. What the articles are about doesn't matter as much, unless we believe that there are dramatic difference in the quality of the recommendations for some topics compared to others. That may very well be the case, and it would be interesting to find out, but I don't have a hypothesis about why it would be the case. Jmorgan (WMF) (talk) 16:43, 21 March 2017 (UTC)