In today’s fast-paced digital world, an increasing number of individuals expressed their thoughts, feelings, and experiences through social media and other online platforms. These digital footprints offered invaluable insights into their mental well-being. While automatic diagnostic technology wasn’t the goal of the Big-eRisk project, we pushed the boundaries of what technology could do to identify early signs of psychological risks, understand the evolution of these risks, and support mental health professionals in their crucial work.
At the core of the Big-eRisk project was the ambition to develop innovative methodologies for detecting the initial signs of mental health risks, such as mood disorders, and understanding their progression from early indicators (e.g., mood changes, insomnia) to severe symptoms (e.g., suicidal ideation). For instance, analyzing the gradual shift in a person’s language usage as they developed an eating disorder provided critical early warning signs—information that was often hard for health professionals to track manually.
Many individuals use online medical questionnaires for self-assessment of their mental health. Though these tools, developed by respected medical institutions, served an important role, they had limitations, particularly around user engagement and the temporal analysis of mental health evolution. We proposed complementing these tools with automatic text analysis technologies that could extract psychological traits from users’ digital content. This approach not only made mental health screening more engaging but also allowed for the tracking of how a user’s condition evolved over time.
Our project covered a range of applications with significant potential value:
1. Enhanced Screening Tools: By analyzing users’ digital content instead of relying solely on questionnaires, we engaged a broader population and improved early detection rates.
2. Temporal Analysis: Sophisticated text analytics uncovered the origins and progression of mental disorders by examining the chronological sequence of user writings.
3. Support for Mental Health Professionals: These technologies assisted psychologists and educators by providing insights into the temporal patterns of psychological risks within specific populations or individual cases.
4. Massive Data Analysis: Harnessing the power of big data, we conducted comprehensive analyses of psychological traits at a scale never before possible.
The Big-eRisk project was inherently multidisciplinary, blending expertise in text processing, computational linguistics, and high-performance computing with deep psychological insights. Our team included experts in psychology who ensured that our models incorporated expert knowledge, were validated rigorously, and could be seamlessly integrated into day-to-day professional practice. Our fundamental hypothesis was that natural language use could reveal signs of various psychological disorders. We aimed to develop early detection technologies by analyzing the textual content published by individuals. Previous studies had shown a strong correlation between language use and psychological conditions, but these studies were often limited in scope and scale. We intended to expand this research to a larger scale and incorporate temporal analysis.
Our primary scientific and technological objectives:
1. Develop Large Collections and Resources: Created extensive collections of labeled data for various psychological disorders to evaluate the algorithms and models we were developing.
2. Effective Textual and Semantic Search: Created scalable methods for locating text evidence of psychological disorders and developed temporal topic analysis models.
3. Domain-specific Linguistic Resources: Built resources for training neural language models to support the natural language processing pipeline.
4. Massive Data Processing Models: Developed efficient models for the collection, ingestion, and processing of social media data on a large scale.
5. Hybrid Intelligence Methods: Combined expert knowledge with advanced algorithms for supervised and semi-supervised learning.
6. Trustable Resource Recommendation Models: Developed models for recommending reliable and personalized resources to individuals at risk.
The Big-eRisk project stood at the forefront of technological innovation in mental health. By leveraging the power of natural language processing, big data, and machine learning, we aimed to develop tools that could make a meaningful impact on early detection and intervention for psychological disorders. Through our efforts, we hoped to pave the way for a future where mental health professionals were better equipped to understand and address mental health challenges, ultimately improving the well-being of individuals and communities.
This study has been led under the project “Big-eRisk: Early prediction of personal risks on massive data” (PLEC2021-007662) funded by The Ministry of Science (MCIN/AEI/10.13039/501100011033) and the European Union (NextGenerationEU/PRTR).
