Munmun De Choudhury: Mental health is one part of medical sciences where we have not seen as much progress.…And that’s where the research that I do finds its motivation. Can we find other ways of assessment that can improve the status quo in the way we both diagnose people with mental health risks, and also the way we treat them?

Noshir Contractor: Today, our guest is Munmun De Choudhury, who is a professor of interactive computing at the Georgia Institute of Technology, where she leads the Social Dynamics and Wellbeing Lab. You just heard her talking about what motivates her innovative research in Web Science, which uses social media in order to understand and improve mental health. She adopts a highly interdisciplinary approach, combining social computing, machine learning and natural language analysis with insights and theories from the social, behavioral and health sciences. She has been recognized with the 2021 ACM-W, or the Association for Computing Machinery’s Women, Rising Star Award, the 2019 Complex Systems Society Junior Scientific Award, and over a dozen best paper and honorable mention awards from the ACM and the Association for the Advancement of Artificial Intelligence. Her work has also received extensive coverage in popular press including the New York Times, NPR and the BBC. 

Welcome, Munmun. 

Munmun De Choudhury: Thank you very much for having me here.

Noshir Contractor:  This is a pleasure to get a chance to talk with you. Your work focuses on how web science and the web more generally can help us to detect mental well-being issues, to mitigate those mental well-being issues, as well as to facilitate the treatment of these issues. I’m curious what got you interested in looking and applying these computational approaches to study wellbeing?

Munmun De Choudhury: I loved math and science. But I also loved all of the other coursework that I did as a kid in social science and social studies and humanities. Until I was, late in my college years, I didn’t know if there could be a possible way to connect and bridge the two, like, how do you do stem work,that is connected to people in some way or the other. So but thankfully, you know, I found myself around people who have been thinking about intersections of different disciplines for many years. And I think that reignited my passion to connect what I do as a computer scientist with something about people. 

I think the work that I do right now, that kind of started about a decade back when I joined Microsoft Research as a postdoc. It was also around the time when I had lost my father to cancer and that was a moment of an introspection for me in my life about what does research mean for me? What is it that I can do? And how can I kind take these ideas about connecting computer science with social science in a direction that would help me find meaning in that personal loss? That’s how I started to connect my work with the health field, with the wellbeing field. And over a course of time, I found my home in mental health broadly speaking. 

Noshir Contractor:  It’s always curious how certain personal events in one’s life can explain where we pivot in terms of our professional goals and aspirations. One of the things obviously, that has motivated your work is the very high prevalence of mental health issues. The National Institute of Mental Health has identified that one in four adults, or about 61 million Americans, report to experiences that are challenges in a given year. What do you think about the ways in which we are currently handling these issues, and how the web and social media approaches that you’re taking might be able to address some of the obstacles we face?

Munmun De Choudhury: So you know, the last 100 years or so have been incredible for medical science, we have made a lot of progress when it comes to illnesses such a infectious diseases, I mean, we are living through a pandemic right now, and if you just look at the pace of progress that we have done around it, it’s incredible. But mental health is one part of medical sciences where we have not seen as much progress. The methods that we currently use are pretty much still the methods that were prominent about the time of the First World War, which is when some of the earliest recognition was given to mental illness as illness. 

We saw some developments in the 60s and 70s, with the invention of antidepressants and other drugs. But in terms of assessment and diagnosis, we are kind of still about 100 years old, a primary paradigm is self-reports from individuals. And unlike other illnesses, where we have objective tests for diagnosis or, or to treat people across the course of their journeys, we don’t have it for mental health. And that’s where the research that I do finds its motivation. Can we find other ways of assessment that can improve the status quo in the way we both diagnose people with mental health risks, and also the way we treat them?

Noshir Contractor: And then one of the things that has also I think, contributed to some of the challenges is a general stigmatization of mental health issues, at least until relatively recently in society. And that brings me to a bit of a conundrum. If people are not in general willing to talk about these issues in public because of fear of being stigmatized, how does looking at social media help?

Munmun De Choudhury: So the beauty of our social media is that we can use it the way we want to. One of the interesting developments that we have seen as social media platforms have become a part of our lives more and more, is that people are finding people with the same lived experience, who are probably going to understand the struggles that they’re going with who are probably not going to be judgmental of the experiences that they have faced in life. And hopefully, it would be less stigmatizing. 

So as much as there is the concern that social media platforms are performative, right? At the same time, we do see other users where people are being candid. And this provides a window of opportunity to look at what struggles they’re going through in terms of their mental health. 

Noshir Contractor: And I imagine that it with some irony, while you might be less inclined to talk about some of these issues with your close friends, you might actually be more comfortable talking to strangers.

Munmun De Choudhury: Sometimes we don’t feel comfortable talking about our deepest struggles with people we know in the offline world, because they might be our coworkers, they might be our bosses. And we don’t want to disclose something that we feel we could be judged on. 

Noshir Contractor: A lot of your work has also looked at how you could look at the passively shared data on social media to proactively detect one’s risk of mental wellbeing challenges. What you are using as a detection strategy has to be somewhat more nuanced than just literally filtering social media postings for those who say they are depressed. Tell us a little bit more about how you go about getting that kind of information about individuals. 

Munmun De Choudhury: You’re absolutely right, that we are looking at more subtle signals, nuances and the writings of people: what type of words that they’re using. So I’ll give you an example. When we use a lot of first person pronouns, such as I, and me and myself, literature and psychology says that it shows an inward focus in terms of our attention, I’m talking mostly about myself. Sometimes, experiences of mental health can be detached from the external surroundings, from the social contexts that people live in. And that can manifest in this inward focused attention. 

But on the other hand, when we use words, such as we and us, it shows that we find ourselves as part of a larger collective, or when we use second person pronouns, it shows that we are interacting with another person. And these are really valuable cues when it comes to somebody’s mental health. 

We also find that social interactions are very, very valuable signals. Am I having a lot of interactions with other people? Am I getting the support that I think I should be getting? So these kinds of signals that are less consciously regulated by people, those are the types of signals that we look for in our work. 

Noshir Contractor: Now, the signals that you’ve talked about in my mind fall into two categories. One is looking at the content and doing a sentiment analysis if you may, or parsing the words to decide what kinds of pronouns people are using. But the other sort of signals that you refer to, are things like just the amount of activity you have, the amount of friends or the amount of responses that either you give to others or others give to what you’re doing.

The latter gets sometimes referred to as metadata, which is not looking at the content, but data about the interactions. Do you see any differences in the utility of and the efficacy of looking at the words versus the data about the data? 

Munmun De Choudhury: Yeah, so that varies by the platform. So for instance, if you’re looking at Twitter, normally, the content words carry a lot more signal.And that might be because people are relatively more candid on Twitter, compared to, let’s say, a platform like Facebook. But on Facebook, we found that some of these metadata or some of these social interaction attributes on these network aspects are more valuable, because for a lot of people, their presence on Facebook is also closely tied with their presence in the offline world.

Noshir Contractor: How do you reconcile that the signal that you get about a person from any one platform may be incomplete, inadequate? And are there ways of being able to cut across different platforms to be able to get a richer picture of someone’s well being? 

Munmun De Choudhury: So I would answer that in two ways. The first part is whether any one platform or a couple of platforms, is that sufficient for us to get a more comprehensive understanding of their mental health?

The reality is that right now, if you look at the state of the art in mental health diagnosis, or treatment, none of those signals are being factored in. So now, even signal from one or two platforms can be additional knowledge to the person themselves, to their caregivers, to their family, or clinicians, whoever might be able to take actionable steps and use that information in helping the person. It’s some data, it’s not all data, but I think it’s still valuable data.

Still, there is the question of, we have our identities that are fragmented across different platforms. And that is more and more the case. So a lot of the work that I have been doing has been to go across platforms and think of these data in terms of their multimodal natures. So I absolutely agree with you that as much as information from a couple of platforms are valuable, nevertheless, there is still value to considering the fragmented nature of our identities.

Noshir Contractor: I know that this is initial work that you’re doing, but are there any examples of insights that were different or modified? Because you were looking at multiple digital services providers?

Munmun De Choudhury: What we have definitely seen is not maybe as much of a contrast, but having data from one platform giving us context about data that we see on another platform. Some of our work recently was looking at physical activity data that is collected through smartphone use. And then we also had, for these individuals, we had their Facebook data. So lining those two up was really insightful for us. So when we saw that the person’s, let’s say, heart rate increase at a certain point in time, we can go back and see what might have been going on on their Facebook, maybe they reported a major life event, they reported something difficult that they were going through. So I think those are definitely some of the strengths of an approach that cross cuts across different sources. 

Noshir Contractor: Technically, how difficult is it today to be able to connect what someone said on Facebook with, with some of what they report, say on a fitness device?

Munmun De Choudhury: From a technical perspective, it’s quite difficult, because you need appropriate infrastructures that can collect that data. Social media data is longitudinal, it is fairly sparse, it is largely text. Fitness data, we are talking about, you know, a very high sampling rate, dense data. And again, I mean, these are largely time series data, for instance, they’re not text. 

There is the question of feasibility, right, like finding enough people who are willing to share not just their data stream on one platform, but across multiple platforms, there is a question there as well.

Noshir Contractor: In terms of ethical issues, are you able to get information about individuals if they consent to sharing that information across platforms? Or do you have to deal with some of the providers themselves?

Munmun De Choudhury: It is a question that is getting harder by the day. There is some good reason why it is getting harder, because I think discussions of privacy and ethics are finally getting the momentum that they deserve, in the field of web science, but at the same time, they’re these questions of multiple stakeholders, who have an interest in a data stream and have different policies, have different value systems around protecting or sharing the data. 

I think at the center of it are the creators of the data, they’re the people who believe they would benefit from, you know, this research. And so we have been doing a lot of work with mental health patients, where they have been voluntarily sharing that data with us. And I feel that is probably the path forward for this kind of research. 

Noshir Contractor: You mentioned that a lot of the data on the social media platforms is largely text, for example. But I’m also thinking of some of the more recent platforms that have got a lot of activity like Tiktok, like Clubhouse. TikTok is. you know, video based for the most part and then Clubhouse is audio based for the most part, have you had much success in being able to parse through video and audio as a way of detecting wellbeing?

Munmun De Choudhury: That’s I think a direction where there could be a lot of research that happens going forward, or we have done work in the image space, particularly on Instagram, which is an image heavy platform, we have done some work on Tumblr, which is also kind of multimodal with images and videos on text. The next frontier are platforms like Tiktok, like Snapchat, where a lot of the young people are going and hanging out. 

Noshir Contractor: A lot of young people now today rely on their social media presence for their own self esteem and for their mental wellbeing, and there have been some efforts recently by some of the platforms to actually not publicize the number of likes a particular post gets or the number of shares a particular post gets, so that people are not overly focused on building their self esteem on the basis of that, do you think that those approaches will work?

Munmun De Choudhury: I think there is definitely something that needs to happen there. Social comparison theory has often been put as sort of the causes of the negative impacts of these platforms on people’s mental health. I think the jury is still out whether these platforms are good or bad for people.

At least the current understanding in the scholarly community is that it just depends upon how somebody uses the platform. Whether it’s for good or for bad, I think the platforms do have a responsibility to consider that their platform(s) are being used by people, sometimes we’re for improving mental health, sometimes not for improving mental health, and how can we change the design of the platforms or the features.

Noshir Contractor: Have you looked at ways in which the research that you’re doing and the tools that you’re developing might be used not just by the person who is in need of help with mental wellbeing issues, but also tools that might be used by family members, or a clinician. Can you talk a little about how the work you’re doing, might have audiences beyond the person, him or herself? 

Munmun De Choudhury: My view of mental health is that it is not a solitary experience. It is an experience that is shaped and it impacts people around you. And therefore, if you’re thinking about solutions that are grounded in these approaches in web science, it’s important to also think about what would that technology look like for these other people in somebody’s social ecology. 

So there are two stakeholders that we have engaged with quite extensively in the past few years. The first is, like you mentioned, it’s the mental health practitioners, Wehave been working with a Northwell Health, they’re are big health system based out of New York State. And we have been, you know, recruiting and working with mental health patients, but also their clinicians, they are part of our research teams. We are kind of adopting a participatory approach there, in building both the algorithms that use patient’s data, but also what kind of technologies could be built on top of those algorithms that could help these clinicians in the treatment that they provide. And so the clinicians form a very important stakeholder in there who can benefit from these algorithms, because they can get a fuller understanding of what might be happening to the individual.

The other stakeholder kind at a very different scale, are public health stakeholders. And for the last two years, we have been working pretty closely with the Centers for Disease Control and Prevention or CDC, in helping with their public health efforts on suicide prevention. Organizations like CDC are realizing that, you know, a lot of these conversations on mental health, on suicide are happening online. That is an entire piece that is missing from their public health work. 

So the work that we have been doing together is to extract meaningful information about how people are talking about suicide, what kind of stressors are being expressed by people. And that knowledge could provide evidence to organizations such as the CDC, to figure out which communities might be in need of greater health than others, how do we allocate budget,to assign mental health resources? And how can we do that in real time fashion.

Noshir Contractor: And based on your encounters working with these different stakeholders, could you comment on what you see as their receptivity? First of all, is a patient typically enthusiastic or concerned about sharing that this kind of information, giving consent to share this information with physicians? Are physicians excited about using these kinds of tools? And are the CDC policy makers enthusiastic about it? What’s been your experience?

Munmun De Choudhury:  In our interactions, what we have seen is clinicians are curious about how this will impact their decision making, questions of these power imbalances in therapeutic relationships? How will that impact their own relationship with the patient? Right, that’s an important one. And also, there are questions of liability. I mean, when you have an algorithm that looks at social media data and makes an assessment of risk, what happens if it’s correct, what happens if it’s wrong? I mean, who takes the responsibility for that? So there are definitely the legal questions, the questions of infrastructural, institutional support. So, as a clinician, they might not feel comfortable to use such technology, if there is no support from their whole institution and allowing them to do that.

Actually, from the patients, the people with the lived experience, we have seen the least skepticism among all stakeholders. And I think the reason is, they see the value that this can bring maybe directly to their own lives, or maybe the lives of other people like them.

The question of stakeholders like CDC is a very interesting one,  I have been pleasantly surprised how open-minded they have been to technology. In my interactions with the researchers at CDC, I think there’s a great deal of interest in taking some of these algorithms that glean insights from the, from the web, and somehow making them a part of their public health efforts. 

Noshir Contractor: Well, we spend a lot of time talking about social media and websites that are there to be able to help individuals who are having these challenges. But there is also an undercurrent, a set of websites, that are actually there to facilitate people engaging in behavior, say eating disorders like anorexia and bulimia. What are your thoughts about those sites?

Munmun De Choudhury: We had so many aspirations from the web, in the 90s, about how it’s going to be liberating, and it could democratize our freedoms, in many ways that ave have been lacking until that point. and a lot of that has been true. But at the same time, it will be foolish for us to not recognize all of the things that are terrible on these platforms and about these technologies. And the example that you cited is a great one in how these platforms, while they can be used for good purposes, they can also be used for harmful reasons. 

And we see health misinformation is a huge problem. And that we see in the context of mental health, we see in the context of substance misuse, there is a lot of misinformation that goes on around. That is an aspect we desperately need to attend to, when it comes to health more broadly, and also more particularly mental and psychological wellbeing. 

Noshir Contractor: As we begin to wrap up here, Munmun, can you talk a little bit about how the general work that you’re doing on wellbeing applies also in the context of workplace experiences, and of course, including now remote work, as well as perhaps hybrid work moving into the future?

Munmun De Choudhury: I mean, this was true even before the pandemic, that  personal wellbeing and workplace wellbeing is deeply intertwined. But I think this blurring has only been intensified. What constitutes work, what constitutes not-work, those lines, we are not able to manage them very well. If we think about the future of work, there are also a lot of questions in that space. What does it mean to be able to understand workplace wellbeing now, and what is the role of technology because ?

Noshir Contractor: The way we’ve been working in the pandemic, a lot of our work even within the organization is using what is called enterprise social media, things like Microsoft Teams, Slack, Zoom and so on. And that of course means that the same kind of information that you’ve been studying in general social media platforms is now amenable as data to help detect issues within the workplace itself. Now, you have the tools and the data potentially to be able to look at interactions that are happening within the workplace.

Munmun De Choudhury: Absolutely, and workplace harassment needs more attention. I think there is tremendous opportunity to look at some of these workplace behaviors, but at the same time, how far should we go, so that it’s still justified and at one point does it become like, “Big Brother,” right? Remote work opens up these possibilities of using these technologies to both get an understanding of our struggles and difficulties, and at the same time, it can be deeply compromising to one’s privacy.

Noshir Contractor: Absolutely. Again, I want to thank you, Munmun, for taking time to talk about this really exciting area of research that you’ve been at the frontier of pushing, in terms of seeing how web science can help us with the general area of wellbeing, and mental wellbeing in particular. Your approaches and techniques are truly interdisciplinary, and the results and insights you’ve shared with us today, and the concerns that you’ve shared with us, are very reflective of the eclectic approaches you use, in terms of theories and models from a variety of social science and computer science disciplines. So thank you again for taking time to talk with us, Munmun. 

Munmun De Choudhury: Thank you so much for having me, and I enjoyed all the questions and conversations.