MIDAS Data Science Fellow Elyse Thulin Awarded Best Poster by a Trainee at the UCSF Promoting Research in Social Media and Health Symposium 

By | Feature, News, Research, research papers
Elyse Thulin

At the 2022 UCSF Promoting Research in Social Media and Health Symposium, Elyse Thulin, a Postdoctoral Fellow at MIDAS and at the Addiction Center in the department of Psychiatry at Michigan Medicine, was awarded Best Poster by a Trainee.

Using traditional epidemiologic, mixed qualitative and quantitative, and computational machine learning methods, Elyse’s broad program of research focuses on how people use online and virtual spaces to interact in ways that both hinder and support wellbeing, mental health, and changes in substance use behaviors. More specifically, Elyse’s areas of research include cyber dating violence, online substance use recovery support groups, and online support groups for traumatic change/loss. Computational skills greatly enhance her work as it enables her to scrape data from online sources, utilize natural language processing to identify top terms, themes, and sentiment from text, and efficiently expand traditional qualitative methods to efficiently code thousands of posts. Elyse’s long term goal is to become a faculty member who teaches, mentors students, and conducts research around expanded applications of computational social science for health and wellbeing.

Online public support group for recovery from problematic cannabis use: trends of use and topics of discussion Elyse J. Thulin, PhD, Anne Fernandez, PhD, Erin Bonar, PhD, Maureen Walton, PhD

Download a PDF version of the poster here.

Elyse provided the following statement about her research:

“Over the past two decades, there have been significant increases in cannabis consumption in the U.S., tied to greater state legalization of recreational (21 states) and medical (37 states) cannabis use, new routes of administration (e.g., vaping, dabbing, edibles), and increased potency of THC. This is worrying given increases in emergency-room injuries related to cannabis use and the increased prevalence of cannabis use disorders (CUD). Despite increased risk of injury related to cannabis and growing prevalence of cannabis use disorder, admission rates for clinical treatment are down, and more than 85% of who would qualify as having CUD do not receive clinical forms of treatment. In contrast, in recent years there has been an uptick in the use of online nonclinical services for those looking to change their cannabis use behaviors. Despite this uptick, very little is known at this time about the functionality, content and interactions occurring within non-clinical, online spaces. In this poster presentation, I aimed to begin to fill this gap by identifying the major themes of conversation, contextualizing information of those themes, and overlap in the present themes with 4 domains of recovery proposed by the US Substance Abuse and Mental Health Services Administration (SAMHSA) in a publicly available online community of individuals who are aiming to cease using cannabis.

I used a data-driven approach to inform the methods of this study. I scraped data from 10 years from a popular Reddit forum on cannabis cessation. I then evaluated the growth of the community across the 10 year period. I next used pre-processing NLP methods (e.g., case uniformity, stemming, etc.) to ready the data for analysis, then identified the top words and terms present in posts. Finally, I extracted a subset of posts to analyze by hand using qualitative methods, to determine the context around top words and phrases. The growth of the community and top words can be found on the poster. Most importantly, we found five major themes in the present study present in posts to the online cannabis cessation community: 1.) individual identify & cannabis use; consequences of cannabis use; reasons for change; cessation strategies; and consequences of change. While examples within these five themes overlapped with the three SAMHSA domains of health, community and purpose, the domain of home was less common and may be less pertinent to this community. Simultaneously, many posts referenced individual identity and cannabis use in posts. Examples were “I smoked daily for ten years” and “I took my first tok at 14, and by 16 I was using in the morning, afternoon and night”. In the context of a common (but incorrect) public narrative that cannabis is not harmful or addictive, individuals in this community may find it important to share the frequency or longevity of their experiences to help emphasize the significant role that cannabis had in their day to day lives. It may be that increased public awareness of that cannabis can be addictive and harmful, particularly when use begins in adolescence or early adulthood or is heavy and frequent, would create greater opportunities for individuals who have experienced dependence or are wanting to change their cannabis use behaviors.”

MIDAS Challenge Award lead Dr. Stephanie Teasley and collaborator Dr. Kelly discuss learning analytics, higher education and employment

By | News, Research

Based on an NSF-funded workshop, Drs. Teasley (School of Information) and Kelly discuss learning analytics goals and research priorities for the coming decade.  They report on a research agenda that would strengthen the connection between learning analytics and recognition of learner competencies. This could have a transformative impact on the relationship between higher education and employment.  The agenda will also generate significant new research questions leading to insights about learning and the data science techniques for analyzing learning.

Read more:

A MIDAS Challenge Award team, led by Stuart Soroka (Communication and Media & Political Science), works with CNN on a Sentiment Analysis about the 2020 election candidates

By | News, Research

A MIDAS Challenge Award team, led by Stuart Soroka, professor of Communication and Media & Political Science, works with CNN on a Sentiment Analysis of recalled news about the candidates for the U.S. 2020 election. The analysis shows Net Sentiment for all respondents, then each of the Democratic, Independent and Republican respondents.

Data scientists and artists team up to demonstrate “fair representation”

By | News, Research

Depictions of people in museum visual arts collections often reflect the histories of inclusion and exclusion. For example, museum collections in the 1920s likely reflect much less the lives of African Americans than those in the 1960s. U-M Museum of Art’s (UMMA) entire collection of ~24,000 pieces of artwork, spanning 150 years, has been digitized.  A group of data scientists and artists will apply algorithms to recognize faces and to examine how art collections can (mis)represent humanity and how that representation changes with time.  

The project team includes: 

This project, funded by the U-M Arts Initiative, is the result of months of discussion about how data science and arts can give each other a stronger voice to promote social justice.

The issue of fair and unbiased representation looms large as Big Data and Artificial Intelligence (AI) are impacting many aspects of our society, from political messaging to targeted marketing, from parole granting to hiring practices.   With biased data, algorithms will reach superficially compelling but ultimately false conclusions and amplify the biases and the injustices that exist in the data.  This project will illustrate this point by generating a collection of composite representations of human faces from the UMMA dataset, and presenting these representations through an art installation, interactive digital displays and narratives.  Through this new way to communicate research that is understandable, meaningful, and impactful to the public, this project highlights the caveats that lie beneath Big Data and AI applications and inspire the public to participate in shaping our transition into a data-driven society.

Read more:

MIDAS Challenge Award team collaborates with CNN and provides polling expertise

By | News, Research

Drs. Josh Pasek (LSA Communication and Media), Michael Traugott (LSA Political Science), Ceren Budak (School of Information) and Stuart Soroka (LSA Communication and Media), along with their Georgetown colleagues, have been collaborating with CNN to improve survey questions and carry out data analysis.  This work is partly an extension of the Challenge Award project that MIDAS has funded.  Read more:

MIDAS collaborates with the EPA to reduce the environmental impact of vehicle emissions

By | News, Research

In collaboration with the US Environmental Protection Agency (EPA) National Vehicle and Fuel Emissions Laboratory (NVFEL), MIDAS will develop and administer a comprehensive educational program during the 2020-2021 academic year. The overall goal of the program is to support the EPA in its mission of improving air quality through applied data science, while providing experiential learning opportunities for students. In this program, senior-level undergraduate and graduate students will apply advanced data science techniques to real-world problems related to reducing the environmental impact of personal and freight mobility systems, design systems to reduce overall vehicle emissions using connected and autonomous vehicles, and engage in environmental policy research by analyzing the potential environmental benefits shared and/or automated vehicles.

Learn more about the courses.

EECS 498-009 will be on Reducing Emissions through Applied Data Science, the course syllabus is attached.