Tekstschrijven voor tekstdata analyse

Text data help detect mental problems due to corona

Science Works 3 min Femke Trommels

The ongoing corona crisis is causing mental problems, but what about when measures are lifted? Surveys show that people experience fewer negative emotions when there are fewer measures. When you dig deeper, however, it turns out that many people do not shake off negative emotions so easily. This is shown by a text analysis by researcher Bennett Kleinberg: "About half of the people still felt angry or defeated, even after measures were lifted. The surveys alone gave a distorted picture. Further analysis of text data can help identify potential at-risk groups earlier and provide more targeted help for mental health problems."

Researcher Bennett Kleinberg (Methodology & Statistics) is involved in Natural Language Processing. He develops models to statistically analyze unstructured text data. "Since the rise of social media like Twitter, text data has become ubiquitous. We have now developed good working models with which you can study this qualitative data statistically among large groups of people. In this way, you can reveal changes in society and heterogeneity within a group, where questionnaires give a more general picture."

Distorted picture

Together with fellow researchers Isabelle van der Vegt and Maximilian Mozes (University College London), Kleinberg asked 2,500 respondents between the ages of 18 and 83 from the United Kingdom about their feelings toward the pandemic in April 2020 and again in April 2021, when there was a well-established vaccination campaign and measures were scaled down. "The second time around, some 1,700 people participated. This gave us a unique dataset that reflected the changing feelings in individuals. In a survey, people indicated the extent to which they experienced emotions such as fear, joy, anger or sadness. In addition, they wrote a short text to describe their feelings. In general, we found that on average respondents were doing better - positive emotions increased, and negative emotions decreased. Furthermore, we found that texts that dealt with a back to normal and hope for an improvement increased, while texts about panic buying and worries about family and friends decreased. Most strikingly, through more nuanced analysis of the emotion data, we discerned two subgroups, which we called the well-coping and the resignators. The first group felt substantially better in April 2021 than a year earlier. The second group did not see things as rosy. They were less hopeful. While they experienced less worry, they felt more anger, sadness, and disgust."

On average people felt better, but a closer look at written texts revealed two subgroups with very different emotions

Bennett Kleinberg


Bennett Kleinberg:

"Text data help look deeper at differences between individuals within a group. The words people use indicate where their attention is at that moment. That again is related to some extent to their emotions. People who often referred to the vaccine, for example, were less sad and suffered less from anxiety. Words, by the way, are not predictors of emotions. We are therefore now looking at factors that might be. For example, whether someone lost their job during the pandemic or works in a sector that is under pressure."

Acknowledging feelings

The study by Bennett Kleinberg and his colleagues took place among residents of the United Kingdom. What can we learn from this for the Dutch situation? "Generalizing the results to other countries is not possible; the situation is different everywhere. The bigger picture is the opportunity that text analysis offers to find out what is going on with people at an individual level. For example, you can find out more quickly which groups are concerned. What is keeping them busy? And how are they processing the situation?

Text analysis offers the opportunity to find out what is going on with people at an individual level

Mental problems will probably continue to play a role long after the pandemic has passed. This has individual and social consequences. In order to offer targeted help, you would ideally speak to everyone personally, but that is of course impossible. Text analysis can help identify risk groups. By taking their concerns and how they deal with such a situation seriously, you can help these people in a more targeted way."

Date of publication: 7 December 2021