The paper focuses on the system’s efficiency in labelling early signs of anorexia.
Concordia graduate students Elham Mohammadi and Hessan Amini developed a research paper explaining an algorithm, using artificial intelligence, to detect signs of anorexia through social media for the Conference and Labs of the Evaluation Forum (CLEF) 2019, this September.
CLEF is a conference that has been running since 2000 in Lugano, Switzerland. It aims to address a wide range of topics, primarily focusing on “the fields of multilingual and multimodal information access evaluation.” Mohammadi and Amini worked under the supervision of Concordia Professor in Computer Science and Software Engineering Leila Kosseim.
Social media platforms are a rich source of information for research studies because people use these outlets to share a large sum of data in relation to their emotions, thoughts and everyday activities.
The research was based on a simulation scenario using past posts from social media. In an interview with The Concordian, Amini explained that there are a few reasons the study was focused on anorexia specifically.
“It wasn’t covered that much in literature,” he said. “Finding out the patterns requires a more complicated source of analyzing information.”
Their focus was on the early detection of the eating disorder.
“We don’t want to detect the risk after it has happened or after it has caused damage to the person,” Amini explained. “We want to detect that the person is showing signs of anorexia.”
The focus of the study was to test the algorithm. Amini clarified that their role is not to diagnose or analyze the data. The study is about the system’s efficiency in labelling these signs. With this, they are able to send this data to an expert to closely evaluate it.
Surfing through over 2,000 social media posts would be tedious and time-consuming, so the researchers used an algorithm called “attention mechanism.” This algorithm systematically filtered through the abundance of posts to detect those that were the most important, using keywords.
They had one data set that was already separated by users that showed signs of anorexia and those that did not, as well as another set of data that was not categorized at all. Mohammadi and Amini analyzed the data to compare the function of the system; however, it must be noted that when dealing with personal data, ethical complications may occur.
Mohammadi explained that when dealing with user’s data, some people might be hesitant to have their personal information analyzed. “People might not be comfortable with it,” he said.
In being able to detect certain patterns of anorexia on social media, more complex research topics arise. Although this is a good start, Amini explains that this research requires many experts sitting together and discussing solutions.
Amini notes that although people think artificial intelligence (AI) systems like this are set up to replace humans, the opposite is true.
“AI is going to be there to help humans,” he said. Amini explains that it will make the lives of psychologists and mental health practitioners easier.
Although this research is not the final solution, it can help bring awareness to those in need of mental attention and create a healthier society.
Graphic by Victoria Blair