Natural Language Processing Helps Shed Light on Youth Mental Health Crisis

American youth have been battling high anxiety and grief amid the COVID-19 pandemic, according to a new report.

The Crisis Text Line, a nonprofit organization that provides free, 24/7 text-based mental health support, published its third annual national data report, ‘United in Empathy: What 1.3 Million Conversations Say About The State of Mental Health In America’ earlier this week. The report used natural language processing (NLP) to generate insights gleaned from conversations between texters and counselors in 2021.

Over the course of these conversations, 40 million text messages were exchanged regarding mental health struggles such as anxiety, depression, and suicide. Seventy percent of conversations were held with users under the age of 25, highlighting the severity of the US youth mental health crisis. The Centers for Disease Control and Prevention (CDC) report that suicide is the second leading cause of death among youth ages 10 to 24, only behind deaths caused by unintentional injuries.

In the Crisis Text Line report, among other sources, it has been established that the COVID-19 pandemic significantly exacerbated already widespread mental health issues in America, particularly among young people. In particular, the report found that multiple mental health crisis trends either continued, returned, or worsened in 2021 compared to 2020.

High anxiety continued to be present in users’ conversations. Reports of anxiety as a main topic of conversation peaked in 2020 at over 35 percent, dropping only slightly in 2021. But overall, anxiety levels have yet to return to those in 2019.

Users also experienced rising rates of grief related to the pandemic, with texters being 13 percent more likely to bring up grief and bereavement in their conversations in 2021 compared to 2019. Over the past two years, reports of grief as a main topic of conversation have increased from 4.3 to 4.8 percent.

Further, texters experienced a sharp rise in eating and body image issues, with users being 17 percent more likely to bring up topics related to these issues in 2021 versus before the pandemic began. Overall, rates of eating and body image issues coming up in conversations rose from 3.1 to 3.7 percent.

As schools reopened, young people also began reporting school-related stressors such as bullying and self-harm that decreased during remote learning. Remote learning was correlated with a 40 percent drop in conversations about bullying, and conversations about self-harm lessened as well. In 2021, however, conversations about bullying increased by 20 percent, while those about self-harm jumped by 10 percent.

NLP is a type of deep learning. Deep learning, a subset of machine learning (ML), is designed to imitate the way neurons connect to process information in human brains. It relies on algorithmic architecture layers, known as artificial neural networks (ANNs), to create these connections and analyze data. By analyzing how the data is filtered through the ANN’s layers and how those layers interact, a deep learning algorithm can ‘learn’ to see connections and correlations in a dataset.

Specifically, NLP algorithms are based on recurrent neural networks (RNNs). These types of ANNs use temporal or sequential data, such as text or speech to generate insights. NLPs also provide additional benefits in that they can be used to analyze and classify large amounts of unstructured data.

These benefits are highlighted in the Crisis Text Line Report, which processed over 1.3 million conversations to generate its findings. De-identified and anonymized data collected for the report includes text conversations between users and counselors, which highlighted texters' language usage and coping mechanisms. It also includes data generated when the counselor tags the conversation for key topics discussed and screens the texter for suicide risk and optional post-conversation data, which is gathered when users complete a survey on demographic information, satisfaction, and other metrics related to service quality.

As part of its analysis, the report utilized NLP to track which words were most prominent in texters’ conversations to see what other trends and emotions played a role in users’ mental health. Compared to 2020, 2021 saw the words “trauma,” “triggered,” “therapist,” “plans,” and “exhausting” increase the most in prominence. Across both years, however, the words “thinking,” “help,” “Mom,” “depressed,” and “time” were among the most prominent.

At the end of each conversation, Crisis Text Line’s volunteer crisis counselors work with users to find healthy coping mechanisms that can help them return to a state of calm. Of those who used the service in 2021, over 379,000 texters reported using these coping mechanisms, which involved online and offline activities.

The coping mechanisms that increased in popularity the most were entertainment platforms like TikTok and streaming services, where users reported watching anime and documentaries; written media, such as manga; gaming platforms, particularly Twitch, Minecraft, and Roblox; physical self-care, such as getting a massage or going to the gym; spiritual rituals, like burning incense or using crystals and affirmations; activities centered around food, like going out for lunch or dinner, or eating favorite foods such as chocolate or pizza; and smoking marijuana or vaping.

Despite the increase in healthy coping mechanisms like these, mental health advocates state that swift, collective action is needed to support these youth.

The US Surgeon General published a 53-page advisory report entitled ‘Protecting Youth Mental Health’ in December 2021. The report outlines ways that Americans can take collective action to support youth mental health, whether they are family members, educators, healthcare workers, or organizations and businesses.