CLOSE
Tags Posts tagged with "Andrew Schwartz"

Andrew Schwartz

by -
0 1182
COVID-era Human Language Analysis Lab Meeting in July, top row from left, MZ Zamani, Matthew Matero, Nikita Soni ; bottom row from left, Adithya Ganesan, Oscjar Kjell, Linh Pham and H. Andrew Schwartz in the middle. Photos taken July, 2020

By Daniel Dunaief

Computers might not be able to tell you how they are doing, unless they run a diagnostic test, but they might be able to tell you how you are doing.

Using artificial intelligence, a team of scientists at Stony Brook University recently received a $2.5 million grant from the National Institutes of Health to study how social media posts and mobile phone data may be able to predict excess drinking among restaurant workers.

By using data from texting, social media and mobile phone apps, these researchers, led by Andrew Schwartz, an Assistant Professor in the Department of Computer Science, are hoping to use artificial intelligence to predict excessive drinking.

According to the National Institute of Alcohol Abuse and Alcoholism, unhealthy drinking involves seven drinks a week for women and 14 for men.

Schwartz said the study hopes to be able to address whether the researchers, including Richard Rosenthal, the Director of Addiction Psychiatry at Stony Brook Medicine and Christine DeLorenzo, Associate Professor in the Departments of Biomedical Engineering and Psychiatry, could “say what the mood predicts how much participants will be drinking in the future.”

By analyzing the content of texts and social media posts, Schwartz and a team that also involves scientists from the University of Pennsylvania will explore whether an increase in stress is more likely to happen before an increase in drinking.

The researchers will study the effect of empathy, which can be health promoting and health threatening. “We believe AI-behavior-based measures will work better than questionnaires for detecting an unhealthy style of empathy,” Schwartz explained in an email. The AI will search for non-obvious patterns of social media posts and texts to determine which type of empathy a person might demonstrate and whether that empathy could lead to a drinking spiral.

Empathy theoretically may add to stress for bartenders and restaurant workers as they often listen to customers who share their tale of woe with food service professionals and are also in a social job.

Indeed, amid the pandemic, where levels of stress are higher during periods of uncertainty about public health and in which restaurant workers might be more likely concerned about their employment, this study could provide a way to understand how increases in alcohol consumption develop potentially to inform new ways to interrupt a negative spiral. “The extra stress of job security is heightened right now” for restaurant workers, among others, Schwartz said.

By validating AI against accepted tools, the researchers hope to gauge the AI-decoded link between emotion and unhealthy drinking behavior by aligning what an individual is expressing in social media with indicators of their emotional state and drinking.

Participants in the study are filling out brief surveys several times a day.

In the long run, the scientists hope this kind of understanding will allow future public health professionals to offer support services to people without the cost of having to administer numerous questionnaires.

The researchers had received word that their proposal had received the kind of score from the NIH that suggested they would likely get funded last July. They could have received positive funding news any time from November through May, which was when they learned that they had secured the financial support to pursue their research.

The topic of study is “extremely relevant,” he added, amid the current uncertainty and the potential for a second wave in the fall or winter.

“We’re interested in studying how unhealthy drinking develops and how it plays out in people’s daily lives,” Schwartz said.

Social media provides a window into the emotional state of the participants in the research.

To be sure, the researchers aren’t looking at how people post about drinking, per se, online. Instead, the scientists are looking at how people in the study answer questions about their drinking in the regular questionnaires.

The researchers came together for this effort through the World Well Being Project, which is a research consortium in collaboration with scientists at the University of Pennsylvania, Stony Brook University and Stanford.

The project involves groups of computer scientists, psychologists and statisticians to develop new ways to measure psychological and medical well-being based on language in social media, according to the group’s web site, which describes “Authentic Happiness.”

In a recent study, 75,000 people voluntarily completed a personality questionnaire through Facebook and made their status updates available. Using these posts, the researchers were able to predict a user’s gender 92 percent of the time just by studying the language of their status updates.

Researchers in substance use approached the World Well Being Project, which Schwartz is a part of, about the topic of unhealthy alcohol use.

The Artificial Intelligence methods Schwartz is developing and that the scientists are testing through this grant are aimed at understanding how a person is changing their language over time through their digital footprint.

In the future, Schwartz believes this approach could contribute to personalized medicine.

“When someone is most at risk, apps that are validated [may be able to] detect these sorts of patterns,” he said. While this study doesn’t provide a personalized patient app, it should provide the tools for it, he explained.

Optimizing this work for false positive and false negatives is a part of this study. The researchers need to create the tools that can make predictions with minimal false positives and false negatives first and then hope it will be used to interact with patients.

In this type of artificial-intelligence driven work, researchers typically need about 500 words to come up with a conclusion about a person’s emotional state. A goal of this work is to get that number even lower.

Fotis Sotiropoulos, the Dean of the College of Engineering and Applied Sciences, offered his enthusiastic support for this effort.

Schwartz is blazing a trail in advancing AI tools for tackling major health challenges,” Sotiropoulos said in a statement. “His work is an ingenious approach using data-science tools, smart-phones and social media postings to identify early signs of alcohol abuse and alcoholism and guide interventions.”

by -
0 1774
Andrew Schwartz. Photo courtesy of Stony Brook University

By Daniel Dunaief

In the era of social media, people reveal a great deal about themselves, from the food they eat, to the people they see on a subway, to the places they’ve visited. Through their own postings, however, people can also share elements of their mental health.

In a recent study published in the journal Proceedings of the National Academy of Sciences, Andrew Schwartz, an assistant professor in the Department of Computer Science at Stony Brook University, teamed up with scientists at the University of Pennsylvania to describe how the words volunteers wrote in Facebook postings helped provide a preclinical indication of depression prior to a documentation of the diagnosis in the medical record.

Using his background in computational linguistics and computational psychology, Schwartz helped analyze the frequency of particular words and the specific word choices to link any potential indicators from these posts with later diagnoses of depression.

Combining an analysis of the small cues could provide some leading indicators for future diagnoses.

“When we put [the cues] all together, we get predictions slightly better than standard screening questionnaires,” Schwartz explained in an email. “We suggest language on Facebook is not only predictive, but predictive at a level that bears clinical consideration as a potential screening tool.”

Specifically, the researchers found that posts that used words like “feelings” and “tears” or the use of more first-person pronounces like “I” and “me,” along with descriptions of hostility and loneliness, served as potential indicators of depression.

By studying posts from consenting adults who shared their Facebook statuses and electronic medical record information, the scientists used machine learning in a secure data environment to identify those with a future diagnosis of depression.

The population involved in this study was restricted to the Philadelphia urban population, which is the location of the World Well-Being Project. When he was at the University of Pennsylvania prior to joining Stony Brook, Schwartz joined a group of other scientists to form the WWBP.

While people of a wide range of mental health status use the words “I” and “me” when posting anecdotes about their lives or sharing personal responses to events, the use of these words has potential clinical value when people use them more than average.

That alone, however, is predictive, but not enough to be meaningful. It suggests the person has a small percentage increase in being depressed but not enough to worry about on its own. Combining all the cues, the likelihood increases for having depression.

Schwartz acknowledged that some of the terms that contribute to these diagnoses are logical. Words like “crying,” for example, are also predictive of being depressed, he said.

The process of tracking the frequency and use of specific words to link to depression through Facebook posts bears some overlap with the guide psychiatrists and psychologists use when they’re assessing their patients.

The “Diagnostic and Statistical Manual of Mental Disorders” typically lays out a list of symptoms associated with conditions such as schizophrenia, bipolar disorder or depression, just to name a few.

“The analogy to the DSM and how it works that way is kind of similar to how these algorithms will work,” Schwartz said. “We look at signals across a wide spectrum of features. The output of the algorithm is a probability that someone is depressed.”

The linguistic analysis is based on quantified evidence rather than subjective judgments. That doesn’t make it better than an evaluation by mental health professional. The algorithm would need more development to reach the accuracy of a trained psychologist to assess symptoms through a structured interview, Schwartz explained.

At this point, using such an algorithm to diagnose mental health better than trained professionals is a “long shot” and not possible with today’s techniques, Schwartz added.

Schwartz considers himself part computer scientist, part computational psychologist. He is focused on the intersection of algorithms that analyze language and apply psychology to that approach.

A person who is in therapy might offer an update through his or her writing on a monthly basis that could then offer a probability score about a depression diagnosis.

Linguistic tools might help determine the best course of treatment for people who have depression as well. In consultation with their clinician, people with depression have choices, including types of medications they can take.

While they don’t have the data for it yet, Schwartz said he hopes an algorithmic assessment of linguistic cues ahead of time may guide decisions about the most effective treatment.

Schwartz, who has been at SBU for over three years, cautions people against making their own mental health judgments based on an impromptu algorithm. “I’ve had some questions about trying to diagnose friends by their posts on social media,” he said. “I wouldn’t advocate that. Even someone like me, who has studied how words relate to mental health, has a hard time” coming up with a valid analysis, he said.

A resident of Sound Beach, Schwartz lives with his wife Becky, who is a music instructor at Laurel Hill Middle School in Setauket, and their pre-school-aged son. A trombone player and past  member of a drum and bugle corps, he met his wife through college band.

Schwartz grew up in Orlando, where he met numerous Long Islanders who had moved to the area after they retired. When he was younger, he used to read magazines that had 50 lines of computer code at the back of them that created computer games.

He started out by tweaking the code on his own, which drove him toward programming and computers.

As for his recent work, Schwartz suggested that the analysis is “often misunderstood when people first hear about these techniques. It’s not just people announcing to the world that they have a condition. It’s a combination of other signals, none of which, by themselves, are predictive.”

TBR News Media is your local news and entertainment website.
We provide you with the latest breaking news and information for your community.
Contact us: [email protected]

Most Viewed

©