An AI study of the language used on social media platform Reddit has revealed increasing levels of anxiety since the beginning of the pandemic.

Study

The study, by MIT and Harvard University researchers, sought to leverage natural language processing (NLP) algorithms to characterise changes in some of the world’s largest mental health (15 groups) and non-mental health support groups (11), by using AI, machine learning and topic modelling to study, and classify the language used in conversations on Reddit. The study looked for changes in 90 text-derived features of 826,961 unique users in their conversations from 2018 to 2020.

Results and Conclusions

The study revealed that levels of anxiety among those in mental health support groups spiked in posts about COVID-19 in January, two months before other support groups started posting about the pandemic. Machine learning analyses found that health anxiety emerged as a general theme across Reddit and, interestingly, the study found that the concerns of a diverse set of individuals converged i.e. the more users posted about COVID-19, the more linguistically similar the mental health support groups became to the Health/Anxiety groups.

Other Language/Stress Research

A study back in 2017, from Carnegie Mellon University in Pittsburgh, highlighted in the ‘Nature’ science journal, also showed how the subtleties in the language people could reveal physiological stress. The research involved 143 adults wearing audio recorders, which switched on every few minutes for two days.  The researchers then compared the language used by the volunteers with the expression in their white blood cells of 50 genes that are known to be influenced by adversity. The study found that when people are stressed, they talk less but use more “emotional intensifiers” e.g. ‘really’ or ‘incredibly’ and use fewer third-person plural pronouns, such as ‘they’ or ‘their’ because when under threat, people focus less on others and the outside world.

What Does This Mean For Your Business?

The AI study showed that the way certain language is used can be an identifier of those at risk from mental health problems and that textual analysis can uncover mental health complaints as they appear in real-time.  This could, in future, be used to help businesses and organisations to spot those employees who are struggling with stress and anxiety, perhaps due to work or home issues.  This could enable businesses and organisations to provide support and help at the right time, thereby addressing the needs of key employees, preventing absenteeism, uncovering potential work issues e.g. the need for training or suffering from by bullying, and could help with the retention and increased loyalty of key staff. If this kind of technology was applied across a business at this time (during the pandemic), it may prove to be another useful tool in helping to manage the softer factors of a physically divided workforce, where stress may not have been spotted by colleagues due to everyone working from home.

Share This