This paper proposes a knowledge-aware influence propagation recommendation model based on graph attention, integrating user–item interactions with semantic information from knowledge graphs (KGs) and neighbor relations from social networks. In semantic propagation, user preferences are decomposed into deep interests captured via multi-step knowledge graph attention and shallow interests from historical interactions. Influence weights between items are derived from co-interaction frequency and semantic similarity, while user weights incorporate social ties. Final user and item embeddings are obtained through multi-layer propagation. Experiments on seven real-world datasets show that the proposed model outperforms strong baselines in recommendation accuracy. Moreover, we introduce a unified metric-based framework for explainability in multi-source recommendation, including three complementary metrics and a composite Total Score for balanced assessment. Results demonstrate that the proposed model achieves superior and consistent performance across all explainability metrics, attaining the best Total Score in all scenarios. To the best of our knowledge, this is the first work showing that jointly leveraging user-item interactions, KGs, and social relations improves both recommendation accuracy and explainability.

The integration of Large Language Models (LLMs) and Generative AI into mental healthcare is reshaping diagnostic, therapeutic, and support systems, marking a transition toward AI-driven mental healthcare. This paper explores various methodologies for incorporating LLMs into mental health applications, categorizing research into key strategies such as prompt engineering, fine-tuning, emotional theory integration, AI-powered chatbots, and long-term memory architectures. Through case studies, we evaluate their effectiveness in areas like depression detection, suicide risk assessment, and AI-assisted psychological counseling, highlighting both their potential benefits and associated challenges. While LLMs enable more empathetic interactions, personalized interventions, and early risk identification, concerns remain regarding bias, misinformation, safety risks, and ethical issues related to privacy, informed consent, and AI’s role in clinical decision-making. We outline critical research gaps, stressing the importance of real-world validation, standardized evaluation methods, and strong risk mitigation frameworks. This paper serves as a guide for the responsible, scalable, and ethical integration of AI in mental healthcare, ensuring these technologies enhance patient well-being while adhering to clinical and ethical standards.

not-yet-known not-yet-known not-yet-known unknown Introduction: Part-of-Speech (POS) tagging, the process of classifying words into their respective parts of speech (e.g., verbs, nouns), is a fundamental task in various Natural Language Processing (NLP) applications. It serves as a crucial preprocessing step for tasks such as machine translation, question answering, and sentiment analysis. However, existing Persian POS corpora predominantly consist of formal texts, such as news articles and academic publications. Consequently, POS tagging tools and models developed on these corpora often struggle to accurately process colloquial language found in social media and informal contexts. Method: This paper introduces the ”Colloquial Persian POS” (CPPOS) corpus, specifically designed to facilitate POS tagging for colloquial Persian text. The corpus encompasses both formal and informal texts collected from diverse domains, including political, social, and commercial content on platforms like Telegram, Twitter, and Instagram, comprising over 520,000 labeled tokens. Following a year-long collection period, we implemented rigorous preprocessing steps, including normalization and both sentence and word tokenization tailored for social media text. The tokens and sentences underwent meticulous manual annotation and verification by a team of linguistic experts. Additionally, this study establishes comprehensive guidelines for POS tagging annotation. Results: To assess the quality of the CPPOS corpus, we trained various deep learning models, including those from the RNN family. A comparative analysis with the well-established ”Bijankhan” corpus and the Hazm POS tool—both trained on Bijankhan—demonstrated that our model utilizing the CPPOS corpus significantly outperforms these benchmarks. Specifically, with the integration of the new corpus and a BiLSTM deep neural model, we achieved a notable 14% improvement over previous datasets.

Background: Freezing of gait (FoG) is a common and debilitating symptom in individuals with advanced Parkinson’s disease (PD), significantly increasing the risk of falls. Wearable devices have facilitated the detection of FoG and falls, but early prediction remains underexplored. This study investigates the use of multimodal sensor fusion and deep learning for the early prediction of FoG events in PD patients. Research Question: Can a multimodal sensor fusion deep learning model accurately predict FoG events well before time in Parkinson’s disease patients, and how robust is the model to noise and inter-subject variability? Methods: The proposed study utilized Inertial Measurement Unit (IMU), Electromyography (EMG), and Electroencephalography (EEG) signals from PD patients to develop and evaluate deep learning models. The CNN+LSTM architecture was employed and compared with other classifiers. Stratified ten-fold cross-validation was used to assess model accuracy. The robustness of IMU+EMG and IMU+EMG+EEG configurations to noise was tested, and inter-subject performance evaluation was conducted. Pre-FOG detection capabilities were also analyzed to emphasize the importance of temporal dynamics in the multimodal approach. Results: The CNN+LSTM model achieved a high accuracy of 94.45% in predicting FoG events. The IMU+EMG and IMU+EMG+EEG configurations demonstrated robust performance across inter-subject evaluations. The models showed resilience to noise, with the CNN+LSTM and IMU+EMG+EEG configurations maintaining high accuracy. Pre-FOG detection achieved 94.20% accuracy, highlighting the model’s effectiveness in capturing temporal dynamics. Significance: The CNN+LSTM model, particularly in the IMU+EMG+EEG configuration, proves to be a robust and accurate predictor of FoG events in PD patients. The study’s findings underscore the potential clinical impact of multimodal sensor fusion and deep learning in reducing false positives and negatives and enhancing precision, sensitivity, and specificity. These insights are crucial for deploying reliable FoG prediction systems in real-world settings and advancing PD management. Future research should explore additional sensor modalities, transferability to different PD cohorts, longitudinal data, and real-time deployment in clinical environments.

A knowledge graph (KG) is a multidirectional labelled graph used for the graphical representation of knowledge where each node represents an entity, and the edge connecting the two nodes represents a relationship. There has been a rise in the popularity of using KG in information retrieval, recommender system, dialogue system, and question-answering system (QAS). The QAS can be either over structured data or unstructured data. This article studies the existing techniques for question-answering systems over KG (KGQA). We collect more than 30 articles on the question-answering system over a knowledge graph. We give a brief introduction to KG and QAS. Further, we discuss challenges, datasets and evaluation metrics used to evaluate techniques. Finally, we conclude the article by discussing some open research aspects, highlighting factors of the low-resource language for KGQA task and remarks on existing systems.

This study presents a framework for detecting and mitigating fake and potentially attacking user communities within 5G social networks. This framework utilizes geo-location information, community trust within the network, and AI community detection algorithms to identify users that can cause harm. The framework incorporates an artificial control model to select appropriate community detection algorithms and employs a trust-based strategy to identify and filter out potential attackers. It adapts its approach by utilizing user and attack requirement data through the artificial conscience control model while considering the dynamics of community trust within the network. What sets this framework apart from other fake user detection mechanisms is its capacity to consider attributes challenging for malicious users to mimic. These attributes include the trust established within the community over time, the geographical location, and the framework's adaptability to different attack scenarios. To validate its efficacy, we apply the framework to synthetic social network data, demonstrating its ability to distinguish potential malicious users from trustworthy ones.

The quality of machine learning (ML) models deployed in dynamic environments tends to decline over time due to disparities between the data used for training and the upcoming data available for prediction, which is commonly known as drift. Therefore, it is important for ML models to be capable of detecting any changes or drift in the data distribution and updating the ML model accordingly. This study presents various drift detection techniques to identify drift in the survival outcomes of patients who underwent cardiac surgery. Additionally, this study proposes several drift adaptation strategies, such as adaptive learning, incremental learning, and ensemble learning. Through a detailed analysis of the results, the study confirms the superior performance of ensemble model, achieving a minimum mean absolute error (MAE) of 10.684 and 2.827 for predicting hospital stay and ICU stay, respectively. Furthermore, the models that incorporate a drift adaptive framework exhibit superior performance compared to the models that do not include such a framework.

Labeling fine-grained objects manually is extremely challenging, as it is not only label-intensive but also requires professional knowledge. Accordingly, robust learning methods for fine-grained recognition with web images collected from Internet of Things have drawn significant attention. However, training deep fine-grained models directly using untrusted web images is confronted by two primary obstacles: 1) label noise in web images and 2) domain variance between the online sources and test datasets. To this end, in this study, we mainly focus on addressing these two pivotal problems associated with untrusted web images. To be specific, we introduce an end-to-end network that collaboratively addresses these concerns in the process of separating trusted data from untrusted web images. To validate the efficacy of our proposed model, untrusted web images are first collected by utilizing the text category labels found within fine-grained datasets. Subsequently, we employ the designed deep model to eliminate label noise and ameliorate domain mismatch. And the chosen trusted web data are utilized for model training. Comprehensive experiments and ablation studies validate that our method consistently surpasses other state-of-the-art approaches for fine-grained recognition task in a real-world scenario. Simultaneously, this introduces a novel pipeline for fine-grained recognition with substantial efficacy in practical applications. The source code and models can be accessed at: https://github.com/NUST-Machine-Intelligence-Laboratory/DDN.