 Hello, my name is Ava. I'm a PhD student at Group Lens at the University of Minnesota, and today I'm going to be talking about our projects, policy deliberations in Wikipedia talk pages, and how editors construct knowledge about mental health. So in this presentation, I'll go over a quick overview of our work, the research questions, our data collection process, and I'll also talk about our future methods and plans for analysis. So it's following that mental health is a topic of global concern. And for the last years, society has been pushing efforts to improve access to mental health information, resources and support services. The term mental health literacy describes someone's knowledge about these things, and research has shown that it has a myriad of benefits, such as increasing help seeking behavior and having a lower risk for mental health disorders. So with this in mind, it's really important that mental health knowledge is accessible, accurate, up to date and coming from multiple perspectives. And this is where Wikipedia comes in, because values like easy access, accuracy, and diverse perspectives are central to the Wikipedia community. This is reflected in the Wikipedia community guidelines, policies and principles. A 2020 study also indicated that Wikipedia's health content is the most frequently visited resource for health info online. Wikipedia has over 100 mental disorder articles and many more that are mental health related. All this being said, little is known about how mental health knowledge is constructed on Wikipedia, in particular, who's contributing knowledge, how do they do it, and how is this process facilitated by Wikipedia guidelines and policies. So this study seeks to answer those questions with three specific research questions. The first probes at patterns and policy discussions in the talk pages of mental health articles. Our second research question looks at the impact of policy discussions on the article information and quality. And our third research question seeks to have a larger discussion about how policy discussions and general editing behaviors on mental health articles might align with or do not align with Wikipedia values like pluralism and neutrality. So our methods for doing this start with gathering articles about mental health and will analyze the talk page data in the main article text data of those articles. Now, we're in the early stages of this project, so I'll discuss that first step of gathering articles and then I'll briefly describe some analysis plans for the future. So the first step is to define what mental health articles are and given that mental health is kind of a vague term and pretty general and Wikipedia also lacks a clear demarcation of what mental health articles are. We created our own definition for this study and for the purposes of data collection. So we say that mental health articles are articles about mental disorders, mental disorder, symptomatology, mental disorder treatments and other concepts that are closely related to mental health. And so with that in mind, we to collect our articles about mental health. We start from a core set, a core set that very clearly aligns with our definition. And we start with the list of mental disorders on Wikipedia, which is an article with around, I think, 217 articles about mental health. The idea here is to use this list of articles that clearly align with our definition as a jumping off point to collect more articles about mental health. So here's a screencap of that article and it lists 217 wiki links to mental disorder articles, all of which are included in the DSM or ICD, which are diagnostic and psychiatric manuals. So with our core set, we plan to our goal is to collect candidate articles by extracting all of the wiki links included in those core articles. For example, major depressive disorder, which is a core article in our set has a large list of wiki links. We'll grab all of those and add it to our set of candidate articles. So we do this for 217 articles and it wind up with 9,499 unique candidate articles. The next step is to filter. So to filter our candidate articles, we ask the question of is the candidate article similar enough to our core set of articles. And by similar enough, I mean we use a threshold of cosine similarity to determine how similar the text of a candidate article is to the text of articles in our core set. Next, we want to analyze the talk pages and the main articles of our whole data set. To do this, we have an approach for each research question, which I'll briefly describe. So for our key one, which probes about policy discussions and mental health article talk pages, we obviously first need to identify what policy discussions are. Thankfully, previous work has done this for us already. Next, we want to know how policy is used. So just a very descriptive understanding of like frequency of policy enactment, which policies are enacted. On a deeper dive, we hope to look at the motivations and purposes and the contextual contextual and situational factors for policy usage. And we'll do this with an unsupervised dialogue at detection method used in previous research on Wikipedia. So we'll be looking for dialogue active features like asking questions or object making objections or clarifying. We'll also look at domain specific topics from the context of a conversation or discussion. Some domain specific topics might include symptomatology or talk of controversial treatment options. For RQ2, we want to take these policy discussions and map them to simultaneous edits in the main article space. So we basically want to see if there are patterns in the editing behavior in the main article in the presence or not in the presence of policy discussions. So we can use computational methods inspired by monkey at all to predict editing behavior in or not in the presence of simultaneous policy discussions. And by editing behavior, we mean deletions, insertions, reburts, etc. So for this last RQ, which is more broad and open, bigger, more broad and bigger picture, we aim to connect our whole investigation and mental health articles to an exploration of how editor practices, like policy usage, reinforces or contradicts Wikipedia values. So in our project, policy usage is one way that editors operationalize Wikipedia values, like citing W and POV. But another and more implicit way is through the composition of editors for an article, namely the spread of administrative editors versus content oriented editors. For the cognitive diversity of editors. So some final thoughts. Little is known about how Wikipedia editors leverage guidelines and policies during the editing process, including how this might reflect community values like pluralism. And so given the social importance of mental health info, we feel that mental health articles are a fruitful pathway for this work. Thank you for listening. And I look forward to hearing your feedback.