1,720,968 research outputs found
GPR-63 Adaptive Attention Aware Fusion for Human-in-Loop Behavioral Health Detection
Get PDFIdentifying behavioral health is paramount for law enforcement officers to provide appropriate follow-up community care. In the current practice, law enforcement offices manually identify these behavioral health cases to allow the designation of the relevant follow-up resources. In this work, we develop a tool to automatically detect behavioral health cases from police public narrative reports by identifying behavioral health indicator signals. We propose a novel adaptive attention-aware fusion model for detecting behavioral health signals in sensitive police reports. Our model leverages contextual and semantic information from the reports and relevant behavioral health cues as keywords from a pre-trained attention-weighted keyword-based model. Our model also employs label self-attention mechanisms to correlate label embeddings with the report and keyword representations. Furthermore, we propose a novel clustering-based uncertainty-enabled informative sampling query strategy to integrate humans-in-the-loop in the active learning framework to reduce required annotation from experts. This querying strategy selects the most informative and diverse samples for expert annotation. Our experimental results showed that the proposed model outperforms state-of-the-art classifiers on a dataset of 300 manually annotated ground truth police reports, achieving an accuracy of 87.58% and an F1-score of 85.67%. Applying our querying strategy to our proposed model increased the detection of behavioral health, achieving an accuracy of 92% and an F1- score of 91.1%. Also, our proposed model achieves an accuracy score of 93.75% and an F1-score of 93.61% on unseen samples. Lastly, our proposed model demonstrates its interpretability by extracting the keywords associated with each behavioral health category
GR-288 Comparative performance analysis of hybrid quantum machine learning algorithm to assess Post stroke rehabilitation exercises
Get PDFDue to the advancements in technology, data is growing exponentially. With this increased dataset size, the computation to process the generated information is rising sequentially. And the currently available classical computational tools and learning algorithms will not work due to the limitations of Moore\u27s law. To overcome the computational issues, we have to switch to Quantum Computing which works based on the laws of Quantum Mechanics. Quantum Machine Learning (QML), a subset of Quantum Computing, is faster and more capable of doing complex calculations that a classical computer can\u27t. Classical Computers work on bits - 0 or 1, whereas a Quantum Bit (also known as a qubit) works on the superposition principle and can be 0 and 1 at the same time before it is measured. Other properties known as Quantum Entanglement, Quantum Parallelism, etc., also will help in understanding the other qubit state and parallel processing the data. In this paper, we introduce hybrid quantum and convolutional models built using PennyLane on the UI-PRMD dataset for the Kinect sensor. By involving quantum layers in a traditional network, a better performance can be achieved compared with the traditional neural network performance
GPR-18 Case Exploration: Automatic Keyword Matching Framework for Behavioral Health
Get PDFIn this demonstration, we propose a framework for exploring, identifying, and matching repeated behavioral health keywords in first-responder reports to the current set of Behavioral Health Index Terms provided by subject matter experts (SMEs). The tool incorporates behavioral health-related keywords and has a Graphical User Interface (GUI) that allows non-technical users to explore and analyze 911 first-responder reports. We utilized an inverted index, best-matching (BM25), and plain-text searching algorithms to match keywords in first-responder reports. This tool provides a comprehensive approach to report analysis by identifying indicators of mental health disorders and taking into account the assessments of humanities and social science professional
GPR-2212 Explainable Multi-Label Classification Framework for Behavioral Health Based on Domain Concepts
Get PDFBehavioral 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 healthcare 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
Engineering prompts while navigating the tradeoffs of LLM integration in First Responder Software Application
Get PDFThis paper explores the implications and complications of incorporating LLM for software engineering or LLM4SE in conjunction with prompt engineering and vibe coding, a term coined by Andrej Karpathy, using tools such as Cursor and GitHub Copilot into the development lifecycle of an end-to-end software application. We present a case here in which we use these tools to implement a modular MVC architectural system with a SQLite-SQLAlchemy backend to support the import process, classification, and triage of large-scale case data. To support the large-volume data ingestion without compromising system stability, we implemented multithreaded chunked imports, which in turn enhanced UI responsiveness, minimizing memory overhead and reducing crashing risk. In this paper, we argue that although LLMs supported certain key logic components, in order to build a robust engineering pipeline, traditional software engineering practices had to co-exist, highlighting the importance of having an expert in the loop. We also examine the extent to which domain researchers, i.e., those without experience in software development, can leverage vibe coding and LLM-assisted workflows to build complete, production-level end-to-end software systems
Classification and Comparative Analysis of In-Hospital Suicidal Behaviors of Patients Using Neural Networks And NLP Techniques
No full textSuicide, an alarming public health, is one of the top 20 problems in the United States that leaves a lasting impact on families and communities. After a two years decline, a total of 47,000 people committed suicide last year. According to the US CDC report, a person expires every 11 minutes due to attempting suicide. Suicidal behavior information in the health records will help to understand the mental health situation of a patient. By identifying the patients who are ideating and are anticipating attempting suicide using the growing technology, physicians can help the patients\u27 lives by keeping close monitoring. As part of the Phase-I of this project, we built a simple LSTM model on the extracted ScAN [1] (Suicide Attempt and Ideation Events Dataset) data, achieving 94.83% accuracy in predicting the Suicide Attempt (SA) and Suicide Ideation (SI) classes. In the next phase(s), we will apply complex and state-of-the-art model architectures, such as Capsule Neural Networks and EXAM-a three-layer architecture model, for classifying the suicide annotated data
GPR-233 Human-Assisted AI for Detecting Mental Health Indicators in Social Media
Get PDFMental health is essential to overall well-being, and mental illness includes conditions that affect a person’s psychological health, causing significant distress and limiting daily functioning. With advancements in technology, social media has become a platform where individuals openly share their emotions and thoughts, offering a unique window into their psychological states. However, traditional machine learning models struggle to interpret social media data\u27s wide range of linguistic nuances. To analyze this data effectively, collaboration with human experts is crucial. This study proposes an innovative human-AI teaming framework that integrates human expertise with artificial intelligence (AI) to address these challenges. Our framework leverages multi-dimensional data along with expert feedback to identify factors contributing to mental illness. Through extensive testing on Reddit data, our model demonstrates a 9% improvement in performance over the state-of-the-art model, underscoring its efficacy and impact
GPR-16 Attention Driven Framework for Detecting Mental Illness Causes from Social Media
Get PDFMental health is a critical aspect of our overall well-being. Mental illness refers to conditions that impact an individual\u27s psychological state, resulting in considerable distress, and limitations in functioning day-to-day tasks. Due to the progress of technology, social media has merged as the platform, for individuals to share their thoughts and emotions. The psychological state of individuals can be accessed with the help of data from these platforms. However, it is challenging for conventional machine learning models to analyze the diverse linguistic contexts of social media data. In this work, we propose a novel attention-driven deep framework to overcome these challenges. Our proposed framework utilizes multi-level (word, sentence, and document) data to identify the causes behind mental illness. The efficacy and effectiveness of our proposed model are shown by extensive evaluation on Reddit data. The insights from this research deepen our understanding of different factors behind mental illness and would aid mental health professionals in formulating effective interventions
GMR-208 Automatic Categorization of Behavioral Health Issues in Police reports
Get PDF911 is often the first place contacted for dealing with behavioral health related (BHR) issues. Its estimated at least a fifth of all calls are related to behavioral health, and with BHR affected convicts having a recidivism rate of around 30%, its not hard to see how straining these issues can become on systems already stretched thin, where chronic understaffing is often a reality. A great solution would be if we could intervene as soon as possible to get people the treatment they need, police reports would be excellent for identifying and treating these individuals, but annotation is a long tedious task only certain people have security clearance to do and as mentioned earlier departments are often understaffed. That is why with the help of keywords given to us by behavioral health professionals, we have developed a model for automatic categorization of police reports that can classify police reports into several categories of class type (Situation, Situation Mental Health, Child, Disposition, Disposition Mental Health, Drugs, Medication, Medication Mental Health) by learning the correlation between co-occurrences of class types given keywords, evidence type given keywords, and class type given keywords and then combining those with the embeddings of a Feed Forward Network that analyzed relevant sentences from reports. With this model we were able to achieve an accuracy rate of 72% which was significantly higher than other state of the art methods typically used
- …
