With the continuing active research on deep learning, research on stock price prediction using deep learning has been actively conducted in the financial industry. This paper proposes a method for predicting stock price movement using stock and news data. The stock market is affected by many variables; thus, market volatility should be considered for predicting stock price movement. Because stock markets are efficient, all kinds of information are quickly reflected in stock prices. We create a new fusion mix by combining price and text data features and propose a hybrid information mixing module designed using two map blocks for effective interaction between the two features. We extract the multimodal interaction between the time-series features of the price data and the semantic features of the text data. In this paper, a multilayer perceptron-based model, the hybrid information mixing module, is applied to the stock price movement prediction to conduct a price fluctuation prediction experiment in a stock market with high volatility. In addition, the accuracy, Matthews correlation coefficient (MCC) and F1 score for the stock price movement prediction were used to verify the performance of the hybrid information mixing

Chronic kidney disease (CKD) constitutes a critical global public health challenge. Its early-stage symptoms are subtle and easily overlooked, frequently resulting in delayed diagnosis and escalated treatment expenditures. To enable timely early intervention, this study proposes a deep learning-based multimodal CKD stage prediction model, which integrates Western medical laboratory data and traditional Chinese medicine (TCM) symptom descriptions, thereby overcoming the inherent limitations of unimodal prediction approaches. For Western medical data, the synthetic minority oversampling technique (SMOTE) was employed to address class imbalance. Subsequently, a least absolute shrinkage and selection operator (LASSO) feature selection model was utilized to identify key biomarkers, including serum creatinine and serum chloride. A backpropagation neural network, enhanced with Adam optimization and regularization mechanisms, was constructed for predictive modeling. For TCM symptom texts (e.g., tongue manifestations and pulse conditions), the Google-pretrained Bidirectional Encoder Representations from Transformers (BERT) model was leveraged to learn latent semantic patterns in the textual data. Finally, a multimodal decision fusion strategy based on the attention mechanism was adopted to dynamically learn the relative importance of Western medical and TCM features in CKD staging prediction, culminating in the development of the proposed deep learning-driven multimodal CKD staging model. Validation results indicate that the proposed model outperforms classical machine learning models and unimodal models across multiple metrics, including accuracy, F1-score, area under the curve, and convergence efficiency. These findings confirm its clinical feasibility and effectiveness, providing an innovative multimodal data-driven prediction framework that synergizes the strengths of Western medical quantitative testing and TCM qualitative diagnosis.

Despite the observable benefit of Natural Language Processing (NLP) in processing a large amount of textual medical data within a limited time for information retrieval, a handful of research efforts have been devoted to uncovering novel data-cleaning methods. Data cleaning in NLP is at the centre point for extracting validated information. Another observed limitation in the NLP domain is having limited medical corpora that provide answers to a given medical question. Realising the limitations and challenges from two perspectives, this research aims to clean a medical dataset using ensemble techniques and to develop a corpus. The corpora expect that it will answer the question based on the semantic relationship of corpus sequences. However, the data cleaning method in this research suggests that the ensemble technique provides the highest accuracy (94%) compared to the single process, which includes vectorisation, exploratory data analysis, and feeding the vectorised data. The second aim of having an adequate corpus was realised by extracting answers from the dataset. This research is significant in machine learning, specifically data cleaning and the medical sector, but it also underscores the importance of NLP in the medical field, where accurate and timely information extraction can be a matter of life and death. It establishes text data processing using NLP as a powerful tool for extracting valuable information, like image data.

Precise survival risk stratification for bladder urothelial carcinoma (BUC) remains a clinical challenge. We developed and validated a multimodal AI agent that integrates textual, radiographic, and pathological data from 1185 patients across four medical centers to predict survival risk. The agent employs LLMs to standardize pathology reports, interactive deep learning networks for precise CT image segmentation, and extracts features from CT scans and whole slide images using CTVisionNet and MacroVisionNet. The multimodal fusion framework, MATCH-Net, integrates these features with microscopic pathology information and clinical text embeddings using a multi-head attention mechanism to generate a comprehensive prognostic score. In multi-center validation, MATCH-Net demonstrated robust performance (C-index ranging from 0.836 to 0.874) and effectively stratified patients into high- and low-risk groups, identifying potential candidates responsive to adjuvant chemotherapy. Furthermore, the framework enabled the quantification of novel, interpretable prognostic biomarkers and provides a reliable and clinically applicable solution for personalized BUC prognosis.

The rapid growth of digital communication platforms has generated vast volumes of user-generated textual data and digital footprints, creating growing demand for scalable artificial intelligence systems capable of supporting evidence-based decision-making. This study proposes and evaluates a human–AI collaborative analytical pipeline for multi-class sentiment and aggression analysis of large-scale social media data (N = 15,064 messages) related to an urban infrastructure project. The proposed framework integrates standard NLP preprocessing, machine learning-based classifiers, temporal aggregation, and controlled large language model (LLM)-assisted classification within a structured analytical workflow that incorporates expert validation and oversight. A stratified manual validation procedure (n = 301) demonstrated substantial inter-annotator agreement (κ = 0.70) and stable multi-class classification accuracy (80%). The results indicate that combining sentiment polarity and aggression detection as complementary linguistic indicators improves sensitivity to shifts in discourse dynamics and enables early identification of emerging social tension. The study demonstrates the potential of human–AI collaborative analytical frameworks for transparent, interpretable, and predictive large-scale social media analysis in decision-support contexts.

This paper presents a hybrid deep learning approach for personality trait classification from textual data. With the rapid growth of social media platforms, analyzing personality traits from user-generated text has become an important research area in natural language processing. The proposed system combines Convolutional Neural Networks (CNN) for effective feature extraction and Long Short-Term Memory (LSTM) networks for capturing contextual and sequential dependencies in text. The model utilizes TF-IDF for feature representation and is trained and evaluated on a labeled personality dataset based on standard personality traits. Extensive preprocessing techniques, including text cleaning, tokenization, and normalization, are applied to improve data quality and model performance. Experimental results show that the proposed CNN–LSTM model achieves an accuracy of 98%, outperforming traditional machine learning models such as Support Vector Machine (56%), Random Forest (53%), and K-Nearest Neighbors (31%). The improved performance of the hybrid model is attributed to its ability to learn both local semantic features and long-term contextual relationships in textual data. Furthermore, the model demonstrates strong generalization capability and robustness when applied to unseen data. The results indicate that the proposed approach is highly effective for real-world applications such as personalized recommendation systems, mental health analysis, user behavior prediction, and human-computer interaction. Keywords — Personality Trait Classification; Deep Learning; CNN-LSTM; Natural Language Processing; Text Mining; Machine Learning.

Abstract Career selection is a critical decision-making process that significantly influences an individual’s professional growth and long-term success. However, students often struggle to identify suitable career paths due to the abundance of unstructured information and the lack of personalized guidance systems. Traditional career recommendation platforms primarily rely on keyword-based search and static data representations, which limit their ability to understand user intent and provide context-aware suggestions. This paper presents the design and implementation of an AI-based career recommendation system aimed at delivering personalized guidance through intelligent information retrieval and dynamic content generation. The system is designed using a modular architecture that integrates a frontend interface, backend services, a structured relational database, and AI-driven components. To enhance recommendation accuracy, the system employs transformer-based embedding techniques to convert textual career data into high-dimensional vector representations. Semantic similarity between user queries and stored profession data is computed using cosine similarity, enabling context-aware retrieval of relevant career options beyond exact keyword matching. Additionally, a dynamic filtering mechanism is applied to ensure that only the most relevant recommendations are presented to the user. To further support personalized guidance, the system incorporates a large language model (LLM) to generate structured learning paths tailored to specific professions. These learning paths include foundational concepts, required skills, tools, and project-based progression, providing users with actionable insights for career development. The system is implemented using a full-stack approach, with an administrative interface that enables continuous updating and management of career-related data. Experimental evaluation demonstrates that the proposed system improves the relevance of career recommendations and enhances user understanding through AI-generated guidance. The proposed approach effectively combines semantic search and generative intelligence, offering a scalable and intelligent solution for next-generation career recommendation systems focused on personalized guidance. Keywords: Artificial Intelligence, Career Recommendation System, Semantic Search, Natural Language Processing, Personalized Guidance, Machine Learning, Large Language Models

With the rapid growth of electronic health records (EHRs), clinical notes, and physician summaries, healthcare systems are generating vast amounts of unstructured textual data. Unlocking meaningful insights from this information especially to support early detection of mental health conditions like depression remains a significant challenge. While Natural Language Processing (NLP) offers powerful tools to address this, there’s still a need to explore its effectiveness in real-world clinical contexts. In this study, we apply a range of NLP techniques to clinical text to detect early signs of depression. Our pipeline includes domain-specific preprocessing steps like tokenization, lemmatization, and lexical normalization. We use TF-IDF and contextual embeddings for feature extraction, followed by classification using traditional models (Logistic Regression, Random Forest) and deep learning approaches (LSTM, ClinicalBERT). We obtained promising results, Logistic Regression and LSTM models achieved perfect ROC-AUC scores of 1.000, with F1-scores of 0.800, reflecting strong balance between precision and recall. ClinicalBERT achieved high precision (1.000) but struggled with recall (0.400), resulting in a lower F1-score of 0.571. Random Forest, by contrast, performed poorly across most metrics. These findings show the potential of combining classic and modern NLP methods for early depression detection and suggest that even simpler models can deliver strong results with well-engineered features. We hope this work supports further efforts in building intelligent, interpretable clinical decision-support tools in mental health care.

Mental health disorders like depression and anxiety pose global challenges, requiring accurate, non-invasive detection methods. Classical modes of diagnosis are typically based on self-reported symptoms or clinical evaluation, which could be subjective and protracted in time. To address these limitations, this study proposes NeuroHAGWO-Net, an advanced artificial intelligence-based framework for automated mental health status detection using multimodal data. The proposed model integrates electroencephalogram (EEG) signals and behavioral textual data to enable early and reliable mental health screening. EEG signals are pre-processed with Empirical Mode Decomposition (EMD) for noise removal, while behavioral text data is transformed into embeddings using Bidirectional Encoder Representations from Transformers (BERT) models. The hybrid BiLSTM-CNN architecture captures temporal dependencies and spatial patterns in EEG data, enhanced by integrating behavioral embeddings for multimodal analysis. Features are selected using a novel Hybrid Ant-Grey Wolf Optimization (HAGWO) approach, combining Ant Colony Optimization (ACO) and Modified Grey Wolf Optimization (mGWO), respectively. The AI-based mental health detection is performed using NeuroVisionNet, integrating EfficientNetV2 and Temporal CNNs (T-CNNs). The model's performance is validated on two datasets: behavioral data and EEG signals data. On behavioral data, it achieves an accuracy of 0.9945, precision of 0.9874, sensitivity of 0.9935, specificity of 0.9915, F1-Score of 0.9909, Matthews Correlation Coefficient (MCC) of 0.9925, Negative Predictive Value (NPV) of 0.9905, False Positive Rate (FPR) of 0.0151, and False Negative Rate (FNR) of 0.0092. With its strong accuracy and efficiency in detecting mental health situations under diverse data modalities, NeuroHAGWO-Net Model proves to be a robust tool for early mental health screening and clinical support using modern optimization techniques and deep learning architectures.

In today’s digital world, Open-Source Intelligence (OSINT) is of critical importance in cyber threat analysis. However, when the studies in the current literature are examined, it has been realized that there is no comprehensive and automatic framework that allows the processing of visual and textual data obtained from social media platforms, especially Telegram, in an integrated manner with artificial intelligence. In order to fill this gap in literature, an AI-supported OSINT framework is proposed in this study, in which social media data is classified using GPT-based natural language processing models and configured for cyber threat intelligence. As a method, the text and images obtained from Telegram channels are collected automatically, classified according to categorical crime headings via GPT-based models, and the obtained outputs are configured in STIX 2.1 format and integrated into the OpenCTI platform. In addition, the System also provides pattern detection and relationship analysis by establishing correlations between different data types. The findings confirm that artificial intelligence-assisted classification provides superior performance compared to traditional methods in the accurate and rapid detection of threat contents. In addition, it has been observed that the data presented via visual panels and timelines with the OpenCTI platform accelerates the decision-making processes. This study not only provides a scalable and more easily applicable model for cybersecurity professionals but also makes an important contribution to the transformation of raw social media data into meaningful and actionable threat intelligence.

Abstract Introduction In England, pharmacists are using their clinical skills and knowledge to support patients and healthcare professionals. Currently, pharmacists provide specialist services in GP surgeries, work as part of multidisciplinary teams, as well as offer urgent care for minor illnesses and vaccinations in the community.[1] From 2026, all newly qualified pharmacists will be independent prescribers. To keep pace with this rapid expansion pharmacists must be able to confidently navigate complex career pathways and training. Aim To develop ‘pharmacist personas’ that guide pharmacists through the different career stages helping them choose pathways that fit with their individual goals and motivations while simultaneously offering opportunities for professional development and learning. Personas are a way of representing a group of people who share similar characteristics. Methods An anonymised online cross-sectional survey was sent to pharmacists registered with the General Pharmaceutical Council (GPhC; n ~ 16 273) who gave permission to be contacted when joining the Centre for Postgraduate Pharmacy Education (CPPE). The survey was distributed via email using the CPPE database. It employed three types of questionnaire (1) closed, multiple choice, (2) Likert-type and (3) free text questions. The questions focused on pharmacists’ individual and professional needs, goals, motivations and pain points, (which refer to the specific challenges and problems faced by participants).[2] The multiple choice and Likert-type questions were analysed using Stata. Inferential statistics were applied for subgroup comparisons including t-test for continuous variables and Fisher’s exact tests for categorical variables with significance set at p < 0.05. The free-text data were analysed using Leximancer, a text-mining software application that automatically codes large qualitative datasets by identifying major themes and concepts. Results A total of 253 respondents accessed the survey. Of 183 respondents who reported their main sector of employment, 28.4% worked in primary care, 24.6% in community pharmacy, 23.0% in secondary care, and 24.1% across sectors such as education, integrated care boards, and the pharmaceutical industry. Analysis of motivations by sector showed no significant differences between them with one exception—working conditions (primary: 67.3% (n = 35), community: 53.3% (n = 24), secondary: 38.1%, (n = 16), p = 0.01). The analysis of free text data on motivation and career identified 10 descriptive codes for each sector: learning & progression; more responsibility; developing clinical skills/specialisms; Independent Prescriber (IP) qualifications; plans to move sector; retirement/leave profession; working conditions including pay; work-life balance; move to another sector and miscellaneous. In total, 127 respondents were placed into one of five archetype categories, refined from the initial seven types. Nearly half of all respondents (n = 62, 48.8%) were categorised under ‘aspirational/early adopter/patient driven’ whilst 19.7% (n = 25) were given the ‘family/work-life balance’ archetype. From the remaining archetypes, 15.0% (n = 19) were ‘undecided/evolving’; 10.2% (n = 13) were ‘wild card/miscellaneous’ and 6.3% (n = 8) were ‘content/the settler.’ Conclusion Our analysis established common archetypes of pharmacists’ individual and professional needs, goals, motivations, and pain points. The limitation of the study is the low uptake of the survey, which weakens the representativeness of findings.

Abstract: Sentiment analysis of social media data has emerged as a crucial task for understanding public opinion, customer preferences, and emerging trends in the digital era. With the rapid growth of platforms such as Twitter, Facebook, and Instagram, vast amounts of user-generated content are produced daily, making manual analysis impractical. This study focuses on applying machine learning techniques to automatically classify and interpret sentiments expressed in social media posts. The proposed approach involves collecting large-scale textual data from social media platforms, followed by preprocessing steps such as tokenization, stop-word removal, and stemming to enhance data quality. Various supervised machine learning algorithms, including Naïve Bayes, Support Vector Machines (SVM), and Logistic Regression, are employed to categorize the data into positive, negative, or neutral sentiments. Feature extraction techniques such as Term Frequency-Inverse Document Frequency (TF-IDF) and word embeddings are utilized to improve model performance. The effectiveness of these models is evaluated using standard performance metrics such as accuracy, precision, recall, and F1-score. Experimental results demonstrate that machine learning-based approaches can achieve high accuracy in sentiment classification, with certain models outperforming others depending on the dataset characteristics. Additionally, the study highlights challenges such as handling sarcasm, slang, and multilingual content in social media data. Overall, this research emphasizes the potential of machine learning techniques in automating sentiment analysis and providing valuable insights for businesses, policymakers, and researchers. Future work may explore the integration of deep learning models and real-time analysis systems to further enhance performance and scalability.

Abstract: The rapid growth of social media platforms has resulted in an enormous amount of user- generated textual data expressing opinions, emotions, and attitudes about various events, products, and social issues. Analyzing this data through sentiment analysis helps organizations and researchers understand public opinion and behavioral trends. However, traditional sentiment analysis techniques often struggle to capture contextual meaning, sarcasm, and complex linguistic patterns present in social media text. To address these challenges, this study proposes a context-aware sentiment analysis approach using transformer-based models. Transformer architectures such as Bidirectional Encoder Representations from Transformers (BERT) and related models utilize self-attention mechanisms to capture contextual relationships between words in a sentence. These models are capable of understanding long- range dependencies and semantic context more effectively than traditional machine learning approaches. In this research, social media textual data is preprocessed through cleaning, tokenization, and normalization techniques before being analyzed by transformer-based models for sentiment classification. The proposed system aims to classify social media posts into sentiment categories such as positive, negative, and neutral while considering contextual dependencies within the text. Performance evaluation is conducted using standard metrics including accuracy, precision, recall, and F1-score. The results demonstrate that transformer- based sentiment analysis models significantly improve contextual understanding and classification performance in social media data. This study highlights the potential of transformer models for developing more accurate and context-aware sentiment analysis systems for real-world applications. Keywords: Sentiment Analysis, Transformer Models, Context-Aware Analysis, Social Media Mining, Natural Language Processing, BERT.

The rapid expansion of user-generated content on digital platforms, particularly YouTube, has significantly transformed the way people communicate and share opinions online. However, this growth has also led to a substantial rise in toxic comments, including insults, threats, abusive language, obscenity, and identity-based hate speech. Such content not only harms individuals but also disrupts healthy online discussions and creates an unsafe digital environment for users. Traditionally, moderation of comments has relied heavily on manual efforts, where human moderators review and filter inappropriate content. While this approach can be effective to some extent, it is highly time-consuming, labor-intensive, and impractical for handling the massive volume of comments generated every second on large platforms. As a result, there is a growing need for an automated and intelligent system that can efficiently detect and filter toxic content in real time, ensuring a safer and more respectful online community. To address this challenge, machine learning techniques can be employed for automated text classification. In this approach, raw textual data from comments is first preprocessed through cleaning steps such as removing punctuation, stopwords, and irrelevant characters.

Abstract - Dark patterns are deceptive design techniques used in digital platforms to manipulate users into making unintended decisions, such as unwanted subscriptions or sharing personal information.An intelligent AI-based framework is proposed for detecting such dark patterns in digital interfaces. The aim is to develop an automated and scalable system capable of identifying such patterns in real time. The proposed methodology involves web scraping to collect textual data from websites, followed by preprocessing and feature extraction using TF-IDF. A hybrid ensemble model combining CatBoost and AdaBoost is employed for accurate classification of deceptive patterns such as bait-and-switch, forced continuity, and hidden costs. The results demonstrate high detection accuracy and improved performance compared to traditional rule-based and single-model approaches. The system also integrates a real-time alert mechanism through a web application, enabling continuous monitoring and timely intervention. The proposed solution enhances detection efficiency, scalability, and adaptability in dynamic web environments, contributing to improved user awareness and promoting ethical design practices. Key Words:Dark Patterns, Machine Learning, Artificial Intelligence, TF-IDF, Ensemble Learning, Real-time Detection.

Abstract—The rapid spread of fake news across digital plat-forms poses significant challenges to information reliability and public trust. To address this issue, this paper proposes a hybrid fake news detection framework that integrates statistical feature extraction techniques with transformer-based contextual repre-sentations. The proposed system combines TF-IDF features with BERT embeddings and employs a dual-stage classification strat-egy to achieve a balance between computational efficiency and classification accuracy. The framework processes textual data through preprocessing, feature extraction, and classification stages within a unified architecture. Experimental evaluation on a benchmark dataset demonstrates that the hybrid approach outperforms traditional machine learning and standalone deep learning models in terms of precision, recall, and F1-score. The results highlight the effectiveness of combining statistical and semantic representations for improving fake news detection in real-world scenarios. This work contributes to the development of scalable and efficient AI-based solutions for combating misinformation. Keywords: Fake news detection, Natural Language Pro-cessing, Hybrid model, Transformer models, BERT, Machine learning, Misinformation

Pre-production phase of visual art is a very significant stage because it entails intellectual ideation, story development and search of design. The need to possess smart looking systems that can be utilized to augment the traditional ideation work is growing as well as requirements of fast and diverse creative effort are escalating. The article dwells upon the application of Large Language Models (LLMs) to generate creative concepts in pre-production in visual art. With their abilities to manipulate and generate semantically rich textual data, LLCs are in a good position to be utilized to assist in supporting the early-stage artistic processes. The study proposes a formal methodology that would involve timely engineering, notion generation, and evaluation into a human-AI work system. System architecture is a developed system that assists in the conversion of user specified inputs to structured creative concepts like design of characters, descriptions of scenes and thematic scripts. The paper also explains how LLC can be incorporated with digital art tools in such a way that the ideation process through text may be incorporated into a visual representation without interruption. The obtained outcomes of the experiment show that the workflows that are assisted by LLM have a positive influence on the diversity, originality, and the quality of idea generation as compared to the traditional methods of idea generation. The generated concepts are evaluated using a detailed evaluation framework to assess the quality of the concepts generated by using various measures such as coherence, relevance, aesthetic potential and diversity. In addition, the user study, which will be carried out with artists and designers, will assist in receiving the concept of the practical applicability and usability of the offered approach. The findings demonstrate that LLMs can be regarded as efficient co-creative partners that help users overcome the issue of creative paralysis and expand the scope of their conceptual exploration without losing their artistic control. Despite these advantages, the originality, bias and creative evaluation problems are still present, which proves the need of more research. The discussion of the future directions, including multimodal integration, personalization of AI tools, and the development of the standardized ways of creativity measurement, conclude the paper. Overall, the work is applicable to the field of computational creativity as it demonstrates the possibility of using LLM to enhance the pre-production process related to the visual art and rebrand the human-AI collaboration in the creative industries.

Background: Emergency triage systems using machine learning traditionally rely on structured tabular data (vital signs), creating a “contextual blind spot” that ignores diagnostic information embedded in unstructured clinical narratives. Hybrid AI models that fuse tabular and text data may improve predictive discrimination, but the magnitude and conditions under which fusion adds value remain unclear. Methods: Five databases (PubMed, Scopus, Web of Science, IEEE Xplore, ACM Digital Library) were searched from 1 January 2015 to 15 December 2025. Eligible studies employed Hybrid AI models integrating structured and unstructured emergency department data with quantitative baseline comparisons. Twenty-five studies (N ≈ 4.8 million encounters) met inclusion criteria. We extracted marginal performance gains (ΔAUC), calibration metrics, and demographic reporting. Synthesis followed SWiM principles with subgroup meta-regression testing our novel “Complexity Gradient” hypothesis. Results: Hybrid models demonstrated superior discrimination compared to tabular baselines, with effect magnitude dependent on clinical task complexity. Low-complexity tasks (tachycardia prediction) showed minimal gains (median ΔAUC + 0.036, IQR: 0.02–0.05), while high-complexity tasks (hypoxia, sepsis) demonstrated substantial improvement (median ΔAUC + 0.111, IQR: 0.09–0.13). Meta-regression confirmed complexity significantly moderated effect size (R2 = 0.42, p = 0.003). Only 12% (3/25) of studies reported calibration metrics (Brier scores: 0.089–0.142). Zero studies stratified performance by race/ethnicity; 88% (22/25) failed to report training data demographics. Discussion: The complexity gradient framework explains when multimodal fusion adds predictive value: tasks where diagnostic signal resides in narrative features (temporality, negation) rather than physiological measurements. However, systematic absence of calibration reporting and fairness auditing prevents clinical deployment. Seventy-two percent of studies had high risk of bias in the analysis domain due to retrospective designs without temporal validation. Conclusions: Hybrid triage models show promise for complex diagnostic tasks but require mandatory calibration reporting and demographic performance stratification before clinical implementation. We propose minimum reporting standards including Brier scores, race-stratified metrics, and temporal validation protocols.

ABSTRACT Research on fire safety in traditional villages has primarily focused on disciplines such as social anthropology, safety engineering, and architectural science, with existing studies often limited to a single disciplinary perspective. Framed within resilience theory, this study employs a mixed-methods approach to collect statistical, textual, and spatial-visual data through questionnaires, interviews, and field investigations across 14 ethnic minority settlements in Guangxi and Guizhou provinces to analyze fire safety measures in traditional Dong villages in China. The analysis covers pre-disaster prevention, disaster response, and post-disaster recovery, with a focus on technological, institutional, and cultural dimensions. The study also examines the impact of social changes on fire prevention capabilities in these villages. The findings indicate that the decline of traditional social structures has led to reduced community involvement in firefighting efforts. Furthermore, modern firefighting technologies have not been fully adapted to local needs and cultural contexts, limiting their effectiveness. The paper concludes by proposing strategies to modernize fire prevention by integrating indigenous knowledge with advanced technologies. This research contributes to both the theoretical and practical aspects of fire resilience in traditional rural communities.

Social networking, such as X (formerly Twitter), is the most effective way of having a large human interaction, but it is currently being overwhelmed with automated accounts that simulate human behavior and spread misinformation and manipulate opinions. The detection of such spambots is crucial in ensuring the integrity of information but many of the conventional techniques are based on opaque, black-box models that cause one to contemplate what is happening. Data used in this study is Cresci-15 and Cresci-17 to analyze interpretable ML techniques towards the identification of spambots and fraudulent followers. Both text and feature-based data are utilised and the preprocessing methods of normalisation, tokenization and removal of extraneous information are applied. RFE is a feature dimensionality reduction method that utilizes recursive selection, and resampling algorithms such as SMOTE and SMOTEENN are used to correct the issue of class imbalance. DT, RF, SVM, NB, XGBoost, AdaBoost, Stacking Classifier, and Voting Classifier are some of the various ML techniques evaluated. The findings show that the Stacking Classifier is highly accurate, achieving 99.9% on the Cresci-15 dataset and 99.5% on the Cresci-17 dataset. Moreover, explainable AI models such as LIME and SHAP allow one to visualize the importance of each feature, thus enhancing model transparency and making it easier to make decisions. These results emphasize the effectiveness of incorporating the feature selection method, advanced resampling as well as ensemble learning techniques along with the interpretable means to the dependable determination of the automated accounts in the social networks.