Social Impact

(Update, September 2017) Two things I'm really thrilled about. One is the response I've been getting personally in response to this page, especially from prospective students, who have been letting me know they care about impact, too. Yes, for many people on the "methods" side of computational linguistics, the interest is in improving the technical state of the art and the specific application is secondary as long as it's interesting and produces publishable results. But more and more I've been hearing from people who would love to connect their technical prowess to work that can affect the world in a positive way. The second thing I'm really happy about is the way the broader community has been thinking more and more about research impact, something which existed before but feels like it has picked up over the past several years. I'm seeing more critical mass as people look at applications of data science for social good, and I'm also seeing increased attention to the complex ethical questions that our work can give rise to.

(Update, September 2019) I continue to be really happy to see that people in the community are thinking more about social impact, as well as research ethics. I've just done an update on this page reflecting the continued evolution of my activities andi nterests.

Here are some updated thoughts on particular areas where impact is something I care about a lot.

• Language data and mental health. The facts and figures regarding the mental health crisis in this country are staggering: mental health disorders affect approximately 450 million people globally (World Health Organization, 2017), costing $2.5 trillion, with economic output loss due to mental disorders anticipated at $16.3 trillion worldwide be- tween 2011 and 2030 (Trautmann, Rehm, and Wittchen, 2016). Even in high income countries, 35-50% of those affected by mental disorders don’t receive treatment, and in middle- and low- income countries that number skyrockets to 76-85% (Saxena, Thornicroft, Knapp, and Whiteford, 2007). In the U.S., some 25 million adults will have an episode of major depression this year, suicide is the third leading cause of death for people aged 10-14 and second among people between people age 15-34 (Pal 2015, Kliff 2012, NAMI 2014, CDC 2015), and some 120 million Americans live in federally-designated Mental Health Professional Shortage Areas. Quite simply, the importance of mental health as a topic of research cannot be overstated.

  There is an extremely promising line of research that is on the upswing that may help address these issues: using people's language, e.g. what they say on social media, as a source of evidence for early detection and/or monitoring. For a great recent paper, see Coppersmith et al. (2018), particularly their discussion of the way that social media can offer a window into what's happening with people in the "clinical whitespace" between healthcare encounters. Recognizing this as an area where the R&D community was starting to see serious activity, I (working with Rebecca Resnik, a clinical psychologist, and Microsoft researcher Meg Mitchell) instigated the first-ever Workshop on Computational Linguistics and Clinical Psychology in Baltimore in 2014. That has since become an annual event, with the help of the energetic organizational skills of others, including my former postdoc Kristy Hollingshead. I am currently working on a collaborative project with Prof. Deanna Kelly of UMD's medical school looking at social media language analysis in connection with schizophrenia and depression, and I am also collaborating with colleagues and students looking at ways to identify risk of suicide based on social media postings. I connect closely with members of the suicide prevention community as a member of the Technology & Innovation Committee of the American Association of Suicidology (AAS).

  Mental health data can be very sensitive, which prevents a lot of people from working on it. Some of my recent work has been focused on addressing the crisis-level lack of data for research. This includes the creation of the University of Maryland Reddit Suicidality Dataset (see papers cited on that page) and, over my 2018-19 sabbatical, creation of The Mental Health Data Enclave, a joint project of with NORC at the University of Chicago, which is designed to provide secure, ethical access to mental health datasets for qualified researchers.

• Computational political science. In their 2009 Science article, “Computational Social Science”, David Lazer and colleagues wrote that our online activity “leaves digital traces that can be compiled into comprehensive pictures of both individual and group behavior, with the potential to transform our understanding of our lives, organizations, and societies.” This is an area of great interest to me, particularly because many of those digital traces involve language behavior (social media being the most obvious example) and much of our individual and group behavior is strongly influenced by the language we consume, e.g. political and media framing. There's been a great deal of recent attention on how people make decisions under the general umbrella of "behavioral economics", but far less attention on the connections between language and decision making -- however, Congressional hostility to social science research (e.g. http://www.washingtonpost.com/blogs/monkey-cage/wp/2014/03/13/house-bill-would-cut-social-science-funding-by-42-percent/) threatens to undermine social science research just as it's building momentum toward evidence based approaches driven by large-scale data. The situation is complicated further by private companies stepping into the research vacuum; I organized a panel to draw attention to the issues at South by Southwest Interactive (SXSW) in 2015.

  I've also begun engaging quite a bit with the public opinion research community, particularly at the American Association for Public Opinion Research (AAPOR) conference, where in May 2017 I co-taught a course on text analysis with qualitative research guru Andrew Stavisky, An Introduction to Practical Text Analytics for Qualitative Research. I'm on the founding board of the Text as Data Association, which is broadly concerned with "the development and application of computational, quantitative methods for social scientific and humanistic inquiry into human behavior" (with, at present, a strong leaning toward political science).

• Academia and entrepreneurship. While in academia I've also been an entrepreneur, with experience that includes being technical co-founder of CodeRyte (clinical natural language processing, acquired in 2012 by 3M), lead scientist for Converseon (spearheading development of their sentiment analysis platform), an advisor to FiscalNote (tracking, analysis, and forecasting of legislative and regulatory information), and founder of React Labs, which commercialized my research on scalable real-time response measurement and engagement using mobile devices; in addition, I'm active in consulting. As a result, the interaction between academia and entrepreneurship is something I'm quite interested in, not least because my own primary area of academic research, natural language processing, has recently been identified as an up-and-coming industry that will be worth $13.4B by 2020 with a growth rate of more than 18% over the next five years (http://www.prnewswire.com/news-releases/natural-language-processing-market-worth-134-billion-by-2020-507411291.html)! I enjoy helping to guide people from industry in the right directions with respect to language technology in order to solve real world problems, and helping students and others in academia understand and navigate the landscape where the startup world and industry are concerned.