 Hello, I'm Patrick Gildersleeve, a post-doctoral fellow in Computational Social Science at the London School of Economics. I'll be presenting a project I'm working on at the moment about depths of Wikipedia, understanding cross-platform online attention, content creation and success. I expect that many of this audience are familiar with depths of Wikipedia already. For those that aren't, depths of Wikipedia is a series of popular social media accounts run by Annie Rowerder that highlights the weird and wonderful of Wikipedia. Beyond the social media accounts, Annie has also worked on various Wikipedia dedicated spin-off works. So, what does a typical depths of Wikipedia post look like? Well, if you'll excuse the dodgy PowerPoint animations, it can simply be posting amusing images found on Wikipedia, or highlight unusual articles, or question check-ins motives for crossing the road. On the whole, depths of Wikipedia posts unusual, typically unpopular content to a wide social media audience. This means that by combining data from different sources, it can be an interesting natural experiment on cross-platform attention. To start, our motivating questions are, can we measure depth of a concept on Wikipedia? What are the effects on Wikipedia of this regular viral social media traffic? And can Wikipedia article properties predict article popularity on social media? The data I collect is from Twitter and Wikipedia. I collect tweets over an 18 month time period from three sources. The depths of Wikipedia account, a typical post looks something like this, with a screenshot and a link in a follow-up tweet. The official Wikipedia account. And then tweets from all users with a Wikipedia link in them. For this version of the study, I only look at the time period above, but I actually have all of this data going back to 2020. In terms of Wikipedia data, I take the articles linked on Twitter and a further 1,000 articles random sample not from Twitter and collect. Article revision history, hourly page views, and then click stream networks of what articles link to each other and how often people use the links. So depth is clearly not as traditional sociological concept. However, we can think of it as an unusualness or novelty of something. I attempt to measure it in four different ways. Firstly, attentional depth. Unusual articles are on concepts that people are not exposed to much. So articles with relatively few page views. I quantify this with the average hourly page views to an article before it is posted on Twitter. I compare the articles posted by depth of Wikipedia against those posted by the official Wikipedia account, the sample of all tweets posted on Twitter, and a set of random articles not from Twitter. Testing the mean scores, we find that depth of Wikipedia articles are indeed more unusual than those posted by Wikipedia and the rest of Twitter, but not the randomly selected articles. So there are lots more unusual articles to explore. Next, I look at the hyperlink network structure. We can quantify how important or embedded in the network of Wikipedia articles that a particular article is using the page rank centrality measure. Unusual articles here are ones that are not central to the network, so lower page rank, and receive relatively few links. Again, testing the mean scores, we find that depth of Wikipedia articles are more unusual than those posted by Wikipedia and the rest of Twitter, but not the randomly selected Wikipedia articles. Similarly, we can consider navigational network structure, not just how many links exist, but how often they have clicks by users, this time taking the personalized page rank centrality measure. Yet again, testing the mean scores, we find that depth of Wikipedia articles are more unusual than those posted by Wikipedia and the rest of Twitter, but not the randomly selected Wikipedia articles. Finally, I consider how unusual the text of the linked articles are. If an article uses many unusual words that are not used often across Wikipedia, its average term frequency inverse document frequency score will be relatively high. Here, we find that depth of Wikipedia articles are more unusual than those posted by Wikipedia, but less unusual than randomly selected Wikipedia articles. Overall, content from depth of Wikipedia is more unusual than what most users are exposed to, but there are many more articles in the depths that have not yet been discovered. Next, I want to consider the effects of these popular social media posts on Wikipedia itself, in terms of additional page view traffic, edits and editors. For page views, we want to find out how much attention is driven towards Wikipedia. I consider the peak page views after an article is featured on Twitter and also the total x-axis views in the following week. I use negative binomial regressions, I consider the page views in the previous day, the page views in the same day, the previous year, and the average page views towards the article, as well as whether this is featured by depth of Wikipedia or the Wikipedia account or another account. Peak views increase by 25 times when an article is posted by depth of Wikipedia compared to other accounts. This is equivalent to around 1,200 extra views. And total access views increase by 9.3 times when an article is featured by depth of Wikipedia compared to other accounts, equivalents to around 11,000 views. The effect is substantially larger for depth of Wikipedia than for the official Wikipedia account. Next, with edits, specifically the number of edits on the day an article is featured. Explanations through variables here are the number of edits in the previous 30 days and then whether the article is posted by depth of Wikipedia or the Wikipedia official account. Daily edits increase by a factor of 2.26 times when an article is posted by depth of Wikipedia compared to other accounts. The measured effect here is not as strong as for page views but it's slightly harder here to model the sparse of data and account for exogenous faxes. Finally, looking at simple accounts for editors, I find that in 493 cases, a new editor's first edit is to an article that has just been posted by depth of Wikipedia. These 493 editors go on to make 15,000 edits across the rest of Wikipedia, not just depth of Wikipedia pages. In this final section, I look to model how properties of Wikipedia articles can predict their popularity on social media or alternatively, naively use computer science to explain popular humor. I'd like to investigate how an article's unusualness or novelty across different measures might determine the number of likes it receives. The regression here is a little longer. I'd like to see how the number of tweet likes depends on the different text, structural, navigation, and attentional depth measures, whether it has been posted by the depth of Wikipedia accounts and then interaction effects. The reason for these being that popularity response may be different if an article is posted by depth of Wikipedia or other accounts. We find that most tweets attract few likes, but articles tweeted by depth of Wikipedia attract many more likes. In addition, higher-page view articles attract more likes and less central articles attract more likes, but there are some co-linearity concerns. And then more central articles attract more likes with depth of Wikipedia, but there is a smaller overall effect size. However, the key point here is that the vast majority of tweets' success is not associated with these depth measures. The account-securation skills are more sophisticated. So, to conclude, I'm looking towards some future work. Content from depth of Wikipedia is measurably more unusual than other Wikipedia content on social media. Depth of Wikipedia drives substantial short-term attention to Wikipedia and is inspired by many new editors, but Wikipedia articles depth is only weekly predictive of success on Twitter. This work expanded in a number of different ways. The Wikipedia page views and the decay count scales could in effect help reverse engineer the Twitter feed algorithm. Spillovers of attention to articles not directly linked by depth of Wikipedia could be interesting. There may be longer-term effects of some of these articles being brought into the public consciousness. I'd also definitely like to look at edit quality after an article is featured, what kinds of edits are made and by what kinds of editors, and there are some methodological refinements that could be made. One study on edit campaigns is a nice match difference and differences approach that could be similarly used here. Thank you very much for your time. Thank you to Annie for running the depth of Wikipedia and helping with various questions, and I'm happy to take various questions as well.