Suicide remains a critical public health problem, with increasing cases of Suicide Ideation (SI) necessitating improved identification and intervention strategies. Although generative Large Language Models (LLMs) have demonstrated potential in text analysis for mental health applications, their effectiveness in accurately detecting SI and generating reliable explanations remains underexplored, specially in Brazilian Portuguese (PT-BR) language. This study proposes a new architecture for the Boamente, an AI-based system for suicide prevention. The proposal aims to incorporate into Boamente architecture an explanation generation stage through prompt engineering. The aim is to improve text detection through the classifier model and exploit prompt engineering through an explainer model to generate explanations of why a text in PT-BR does or does not contain SI, increasing the model's interpretability and providing more transparent justifications for its predictions. The proposed structure expands the Boamente architecture by exploring generative LLMs as classifier models (i.e., SI identification) and integrating an explainer model, which generates justifications for predictions. A quantitative evaluation was conducted with different LLMs using classification performance metrics, in which Qwen 2.5 (14B) achieved the highest AUC (0.9898), while the 3B and 7B versions of the same model achieved the best Recall (0.9545). In addition, a qualitative evaluation was conducted with the participation of three professionals (a computer scientist, a linguist, and a psychologist) to analyze the explanations generated by LLMs. The participants considered that LLaMA 3.1 (8B) produced the highest quality justifications. The findings highlight the potential of combining classification and explanation LLMs to enhance explainability and trust in an AI-driven system for suicide prevention.

Data publishing in healthcare raises major concerns due to the sensitive data involved. In this paper, we propose an interactive method for data publishing with privacy preservation between the Manager (data publisher) and the Miner (data requester), in which the Miner specifies the manipulations accepted in the data based on the privacy level defined by the Manager. The method works with structured tabular data and privacy regarding protecting records and attributes. Through a use case, we demonstrate that the use of the proposed method made it possible to prove privacy in data publication, with an impact of less than 3% on the usefulness of the data.

Digital Phenotyping of Mental Health (DPMH) aims to passively collect data from ubiquitous devices to be used as evidence in the process of diagnosis, treatment, and monitoring. Literature presents different sensing mobile applications for digital phenotyping, however they are not extensible and can not be customized for use in other research. In this paper, we propose OpenDPMH, a framework for developing mobile sensing applications able to collect contextual data in order to produce useful user information that represent situations of interest for mental health professionals and researchers, such as human behaviors and habits. Our solution is extensible and reusable, as it allows the inclusion of modules for collecting and processing new raw context data with features for data distribution. By implementing a case study, we demonstrate that OpenDPMH is suitable for the development of DPMH mobile applications. Moreover, we carried out experiments to evaluate the energy consumption on smartphones, which demonstrate a low battery cost to run applications developed using the proposed framework.

Mental health professionals use clinical evidence and information self-reported by patients for diagnosing mental disorders. However, self-reports and questionnaires are approaches affected by cognitive biases due to imprecision in the report. Machine learning has been used along with sensor data embedded in mobile devices (e.g., smartphones) to identify patterns correlated to mental disorders. This study proposes a solution that uses a clustering algorithm to analyze group behaviors over time. We believe that our solution can help mental health professionals to continuously monitor patients, identify similar behaviors, and changes in behaviors. We present an analysis of different clustering algorithms and show group dynamics. We conclude that the Birch algorithm has the best performance for grouping behaviors in our experiments.

Traditionally, the process of monitoring and evaluating social behavior related to mental health has based on self-reported information, which is limited by the subjective character of responses and by various cognitive biases. Today, however, computational methods can use ubiquitous devices to monitor social behaviors related to mental health rather than relying on self-reports. Therefore, these technologies can be used to identify the routine of social activities, which enables the recognition of abnormal behaviors that may be indicative of mental disorders. In this paper, we present a solution for detecting context-enriched sociability patterns. Specifically, we introduced an algorithm capable of recognizing the social routine of monitored people. To implement the proposed algorithm, it was used a set of Complex Event Processing (CEP) rules, which allow the continuous processing of the social data stream derived from ubiquitous devices. The experiments performed indicated that the proposed solution is capable of detecting sociability patterns similar to a batch algorithm and demonstrated that context-based recognition provides a better understanding of social routine.

Depression is a mental disorder characterized by persistent sadness, loss of interest, and a set of behavioral changes. The high prevalence of depression imposes a significant burden on the world population, demanding methods capable of monitoring and treating this mental disorder. Currently, a large number of mobile applications have been designed to provide support to depressive people. This paper aims to identify, analyze and characterize the current state of mobile applications focused on depression. To do so, we conducted a systematic review of applications for depression assistance. The two most popular mobile app stores (Google Play Store and Apple App Store) have been explored to find the most relevant apps. After applying the inclusion and exclusion criteria and performing the quality assessment of the results, 216 applications were selected for the data extraction phase, where we summarized their benefits and limitations and identified gaps and trends. The results of this review evidenced that there is a growth in the diversity of apps' purposes such as chatbot, online therapy, educational tools, mood tracker, testing, and self-help.

Clinical scores are a widely discussed topic in health as part of modern clinical practice. In general, these tools predict clinical outcomes, perform risk stratification, aid in clinical decision making, assess disease severity or assist diagnosis. However, the problem is that clinical scores data are traditionally obtained manually, which can lead to incorrect data and result. In addition, by collecting biological/health data in real time from humans, the current mobile health (mHealth) solutions that computationally solve that problem are limited because those systems are developed considering the specificities of a single clinical score. This work is part of the MDD4ClinicalScores project that addresses the productivity in developing mHealth solutions for clinical scores through the use of Model Driven Development concepts. This paper focus in describing DSML4ClinicalScore, a high-level domain-specific modeling language that uses the Ecore metamodel to describe a clinical score specification. To propose the DSML4ClinicalScore we analysed 89 clinical scores to define the artifacts of this proposed Metamodel. In the end, a practical case study using this DSML is provided to validate the DSML4ClinicalScore Metamodel, and to show how to use the proposal in a clinical situation scenario.

Fighting games are complex environments where challenging action-selection problems arise, mainly due to a diversity of opponents and possible actions. In this paper, we present the design and evaluation of a lighting player on top of the FightingICE platform that is used in the Fighting Game Artilicial Intelligence (FTGAI) competition. Our proposal is based on hierarchical reinforcement learning (HRL) in combination with Monte Carlo tree search (MCTS) designed as options. By using the FightingICE framework, we evaluate our player against state-of-the-art FTGAIs. We train our player against the current FTGAI champion (GigaThunder). The resulting learned policy is comparable with the champion in direct confront in regard to the number of victories, with the advantage of having less need for expert knowledge. We also evaluate the proposed player against the runners-up and show that adaptation to the strategies of each opponent is necessary for building stronger lighting players.

This paper describes a multi-agent system (MAS) acting on an arterial road network, where each intersection is controlled by an agent. The agents are concerned with the efficient control of their intersection. In order to improve the overall performance of the system we propose a model of explicit coordination that directs the behavior of agents for the formation of green waves on the artery, in addition to maintaining the autonomy of each agent. The model is tested in simulation and compared with the traditional approach of synchronization of traffic lights. The results obtained in simulation overcome the traditional model by representing a more realistic model of traffic.