Behavioral health plays a pivotal role in individuals’ overall quality of life. Timely identification and intervention of behavioral health concerns are essential for building a supportive community. In case of emergencies, first responders (police, fire, EMT) provide support to the community. The report of the first responders contains valuable cues about individuals struggling with behavioral health concerns. Currently, first responders manually screen these reports to determine individuals with behavioral health concerns. Manual screening of these reports is time-consuming and prone to error. Automated analysis of these reports is essential for providing timely intervention. Natural language processing techniques demonstrated potential for automated analysis of textual data. However, due to the unstructured and complex nature of the data, it’s challenging for traditional deep learning and natural language processing techniques to identify these cases. Moreover, support from subject matter experts is essential for the effective identification of these cases. In this research, we have developed a multi-label behavioral health detection framework that utilizes domain knowledge from subject matter experts for the effective identification of behavioral health cases. In addition, we have expanded the domain knowledge base by recognizing new behavioral health-related terms. We validated the efficacy and effectiveness of our proposed model through a comprehensive evaluation of real-world data. Furthermore, our proposed model demonstrated 6-24% performance improvement over the state-of-the-art models.

Behavioral health, which covers mental health, lifestyle choices, addictions, and crises, poses serious issues in the community. Thus, appropriately analyzing and classifying behavioral health data is crucial for making informed health-care decisions. Traditional deep learning and natural language processing approaches struggle to effectively identify behavioral health issues because the data is unstructured, complex, and lacks sufficient context. Furthermore, subject matter experts must be consulted to ensure effective identification. In this work, we proposed a deep learning-based framework consisting of several modules: A) domain concept encoder converts the keywords and their evidence types to vectors, which were predefined by a subject matter expert; B) the semantic representation encoder (SRE) is trained on the vectors to learn the relationship between them; C) transformed-based feature learner is an advanced learner that extracts feature embeddings from documents and generates attention weights since it has more context given the incorporated relationship weights; D) The behavioral health multilabel classifier utilizes feature embeddings to classify a document into one or more behavioral health classes; and E) The LLM-enabled explainer provides explanations based on attention weights and classifications. Our proposed framework outperformed state-of-the-art models in multilabel behavioral health case classification while also providing explanations for each classification. Which is crucial in behavioral health analysis.

Behavioral health is a broad term which encompasses our overall mental and physical well-being. Behavioral health issues do not just impact an individual’s life, they also impact wider society. First responders are often the first point of contact for individuals facing emergency. Reports from first responders are a great resource to identify potential behavioral health cases. Manually identifying these cases are time consuming and prone to error. Automation in identifying these reports would be a huge time saver for facilities and allow for them to spend more time on addressing these problems or on more pressing matters. Natural language processing techniques are widely used by researchers for analyzing texts. However, due to the linguistic complexity of the first responder reports it is challenging for traditional natural language processing and deep learning technique to identify these reports. To address this issues in this work we proposed a novel correlational network which utilizes domain experts knowledge along with their correlational relationships to appropriately identify behavioral health cases. Our proposed framework uses the correlation between features and combines them which allows for granular and broad scope insights into a report. Moreover, we performed multi-label classification which reflects the real world scenarios where reports can contain multiple classes. We performed extensive evaluation to determine the efficiency and efficacy of our model. Furthermore, our proposed model outperformed the state-of-the-art model with an accuracy of 72%.

Appropriately addressing behavioral health challenges is a problem in the United States. Behavioral health challenges can manifest in the form of crimes committed by individuals with mental illness, social and domestic violence, etc., which may result in emergency calls to first responders. For example, in domestic violence situations, involved parties may need various supports and resources if no criminal charges are filed. In these instances, when a 911 call is made, it mobilizes the entire first responder system (e.g., police, fire, EMT), each of whom write their respective incident reports. A majority of the calls that first responders are addressing are social and behavioral in nature, which they are often ill-equipped to handle. The recurring nature of these calls and limited available resources, first responders encounter challenges in providing appropriate responses to community residents. Therefore, early identification of these cases through incident reports could guide service providers and first responders to take appropriate actions on behalf of the community resident. However, the presence of noise and contextual variability within these unstructured public narrative reports makes it challenging for identifying social and behavioral health based cases. To address this challenge, in this study, we propose a novel model to analyze the first responders’ public narratives: the model learns contextual features automatically from the public narratives by employing self-attention techniques and incorporates automatically extracted domain knowledge by comparing expert-given keywords to represent document features to identify behavioral health cases. Unlike conventional approaches, integrating domain expert knowledge with the contextual representation of the public narrative report enriches the model’s understanding of social and behavioral health cues and enhances its overall performance. Extensive evaluation showcases the efficacy and effectiveness of our proposed novel model. Our evaluation shows that our model identifies behavioral health with an accuracy of 82% and F1-score of 86%, outperforms the baseline models.