Towards a Retrieval Augmented Generation System for Information on Suicide Prevention*
Pablo Ascorbe, María Campos, César Domínguez, Jónathan Heras, Ana Rosa Terroba-Reinares · 2023
Suicide is a leading cause of mortality worldwide with more than 700 000 per year. Hence, it is instrumental that the general public can access to reliable information that help in suicide prevention. In this work, we have tackled this problem by developing a question-answering system based on retrieval augmented generation — this approach allows the system to generate answers based on a corpus of documents curated by psychologists and psychiatrists. Several alternatives have been tested for the two main components of the system: an embedding model to retrieve relevant contexts from the corpus of documents to answer a given question (the best option was a BERT-based model), and a language model to generate a response from the extracted contexts (the best alternative was the Bertin model). The developed system is a first step towards helping in one of the greatest global public health concerns.Clinical Relevance: The system suggested in this work will offer a reliable and accessible resource that could aid in providing timely information and support to individuals at risk and their families, thereby potentially enhancing clinical interventions and patient outcomes in the realm of mental health.