Big-eRisk: Big-eRisk: Early Prediction of Personal Risks on Massive Data

Mental health and well-being directly affect how we think, feel, or act. Mental disorders are complex and can manifest in very different ways. Despite the severity of these disorders, in a large number of cases, those affected do not receive any treatment or receive it late. Early detection can drastically reduce the adverse effects of disorders for patients and, additionally, it can substantially reduce costs for the public health and social systems.

Many people use social media as a convenient means to share emotions, feelings, and thoughts. The vast amount of publications that individuals post daily can help us improve our understanding of their mental states. Big-eRisk targets this challenge and aims at developing the first generation of tools that support the social and health systems in the early detection of signs of psychological risk.

Objectives

O1. Develop large collections and resources and methodologies for evaluating information algorithms and models for early prediction of risks of psychological disorders on effectiveness, efficiency and scalability.

O2. Define effective textual and semantic search and filtering methods at scale for locating pieces of texts as candidate evidence of psychological disorders. Define efficient and distributable temporal topic analysis models for investigating the textual pieces of evidence.

O3. Develop domain-related linguistic resources for training massive neural language models and helping in the natural language processing pipeline.

O4. Develop efficient models for the crawling, ingestion, and massive processing of social media data at scale.

O5. Define hybrid intelligence methods for supervised and semi-supervised inclusion of expert knowledge of the mental health professionals, revision and validation.

O6. Develop trustable resource recommendation models for individuals at risk with reactive and adaptable suggestions.