Multi-Label Behavioral Health Classification from Police Narrative Report

Understanding the behavior of a community is crucial in building a strong society. To achieve this, we need to offer aid to individuals in need of behavioral health support. In recent years, the term “behavioral health” has emerged for defining a wide range of societal issues, including mental health, substance abuse, and so on. During emergencies, the first responders ensure the safety and well-being of the community. At present, first responders rely on manual iterating through public narrative reports at the end of each month to pinpoint cases linked to individuals grappling with behavioral health issues. This method proves arduous and time-intensive. Moreover, there is limited availability of expert resources, which leads to delays in identifying and addressing potential cases. Natural language processing and deep learning techniques have shown promising results in analyzing textual data. However, it's challenging for traditional models to identify the cases since the reports are unstructured in nature and contain noise. To overcome this challenge, in this research, we developed a novel attention-enabled multi-label behavioral health detection framework to identify different behavioral health cases. Our proposed model utilizes attention mechanisms and considers information from different levels(word, sentence, and document) to gain a comprehensive understanding of the police narrative report and effectively identify different types of behavioral health cases. Furthermore, our proposed model outperformed the state-of-the-art models with an accuracy of 71%.