Scams exploiting real-time social engineering—such as phishing, impersonation, and phone fraud—remain a persistent and evolving threat across digital platforms. Existing defenses are largely reactive, offering limited protection during active interactions. We propose a privacy-preserving, AI-in-the-loop framework that proactively detects and disrupts scam conversations in real time. The system combines instruction-tuned artificial intelligence with a safety-aware utility function that balances engagement with harm minimization, and employs federated learning to enable continual model updates without raw data sharing. Experimental evaluations show that the system produces fluent and engaging responses (perplexity as low as 22.3, engagement≈0.80), while human studies confirm significant gains in realism, safety, and effectiveness over strong baselines. In federated settings, models trained with FedAvg sustain up to 30 rounds while preserving high engagement (≈0.80), strong relevance (≈0.74), and low PII leakage (≤0.0085). Even with differential privacy, novelty and safety remain stable, indicating that robust privacy can be achieved without sacrificing performance. The evaluation of guard models (LlamaGuard, LlamaGuard2/3, MD-Judge) shows a straightforward pattern: stricter moderation settings reduce the chance of exposing personal information, but they also limit how much the model engages in conversation. In contrast, more relaxed settings allow longer and richer interactions, which improve scam detection, but at the cost of higher privacy risk. To our knowledge, this is the first framework to unify real-time scam-baiting, federated privacy preservation, and calibrated safety moderation into a proactive defense paradigm.

In this study, we address the task of discerning gender through the textual content of social media, a crucial step in detecting and mitigating counterfeit account activity. Ensuring accurate gender portrayal on digital platforms is essential for creating a secure and inclusive cyberspace. While research exists for languages like English, Russian, and Arabic, Bangla remains underexplored. To address this, we compiled 15,000 Bangla posts from Facebook groups, profiles, pages, blogs, and forums. We trained seven traditional machine learning algorithms (NB, SVM, LR, DT, RF, SGD, KNN) and three deep learning models (MLP, LSTM, GRU), using stylometric features, Term Frequency-Inverse Document Frequency (TF-IDF), and word embeddings. Traditional models generally outperformed deep learning models, except with stylometric features. Notably, the Stochastic Gradient Descent (SGD) model with TF-IDF achieved the highest accuracy (78.33%) and F1-Score (87.67%). Additionally, Continuous Bag of Words (CBOW) out-performed Skip-Gram (SG) in training the word2vec model, with top accuracy and F1-Score of 75.13% and 79.92%, respectively. These findings represents a significant stride forward in the field of gender identification from Bangla text.

The proliferation of social media platforms, such as Facebook, has led to an exponential increase in diverse user-generated content, including text, images, and videos that could be pivotal in public health contexts. This research paper presents an innovative approach to summarizing Facebook posts that incorporate multiple modalities, aiming to generate concise and informative summaries. Utilizing the MT5 architecture for textual analysis and state-of-the-art computer vision for image and video processing, our approach achieves a seamless integration of diverse modalities. Rigorously evaluated against a comprehensive suite of metrics, including ROUGE and BLEU, our model demonstrates notable effectiveness, with GPT-4 achieving impressive scores across various metrics (BLEU-1: 0.82, BLEU-2: 0.68, BLEU-3: 0.59, ROUGE-1: 0.78, ROUGE-2: 0.60, ROUGE-L: 0.73, METEOR: 0.58, CIDEr: 1.75, SPICE: 0.37). These results highlight our methodology's capability to produce succinct and informative summaries, significantly enhancing decision-making processes in public health monitoring and response. This achievement not only underscores the potential of advanced models in handling the complexity of multi-modal content but also sets a new benchmark for future explorations in social media analytics and user experience enhancement.

With the proliferation of edge devices, there is a significant increase in attack surface on these devices. The decen-tralized deployment of threat intelligence on edge devices, coupled with adaptive machine learning techniques such as the in-context learning feature of Large Language Models (LLMs), represents a promising paradigm for enhancing cybersecurity on resource-constrained edge devices. This approach involves the deployment of lightweight machine learning models directly onto edge devices to analyze local data streams, such as network traffic and system logs, in real-time. Additionally, distributing computational tasks to an edge server reduces latency and improves responsiveness while also enhancing privacy by processing sensitive data locally. LLM servers can enable these edge servers to autonomously adapt to evolving threats and attack patterns, continuously updating their models to improve detection accuracy and reduce false positives. Furthermore, collaborative learning mechanisms facilitate peer-to-peer secure and trustworthy knowledge sharing among edge devices, enhancing the collective intelligence of the network and enabling dynamic threat mitigation measures such as device quarantine in response to detected anomalies. The scalability and flexibility of this approach make it well-suited for diverse and evolving network environments, as edge devices only send suspicious information such as network traffic and system log changes, offering a resilient and efficient solution to combat emerging cyber threats at the network edge. Thus, our proposed framework can improve edge computing security by providing better security in cyber threat detection and mitigation by isolating the edge devices from the network.

Our paper introduces a groundbreaking approach to healthcare information retrieval and engagement through a personalized chatbot system empowered by Federated Learning Based GPT. The system is designed to seamlessly aggregate and curate diverse healthcare data sources, including research papers, multimedia resources, and news articles. Leveraging Federated Learning techniques, the GPT model is trained on decentralized data sources to ensure privacy and security while providing personalized insights and recommendations. Users interact with the chatbot through an intuitive interface, accessing tailored information and real-time updates on medical research and news. The system's innovative architecture enables efficient processing of input files, parsing and enriching text data with metadata, and generating relevant questions and answers using advanced language models. By facilitating interactive access to a wealth of healthcare information, this personalized chatbot system rep-resents a significant advancement in healthcare communication and knowledge dissemination.

Federated Learning (FL) has emerged as a transfor-mative approach for enabling distributed machine learning while preserving user privacy, yet it faces challenges like communication inefficiencies and reliance on centralized infrastructures, leading to increased latency and costs. This paper presents a novel FL methodology that overcomes these limitations by eliminating the dependency on edge servers, employing a server-assisted Proximity Evaluation for dynamic cluster formation based on data similarity, performance indices, and geographical proximity. Our integrated approach enhances operational efficiency and scalability through a Hybrid Decentralized Aggregation Protocol, which merges local model training with peer-to-peer weight exchange and a centralized final aggregation managed by a dynamically elected driver node, significantly curtailing global communication overhead. Additionally, the methodology includes Decentralized Driver Selection, Check-pointing to reduce network traffic, and a Health Status Verification Mechanism for system robustness. Validated using the breast cancer dataset, our architecture not only demonstrates a nearly tenfold reduction in communication overhead but also shows remarkable improvements in reducing training latency and energy consumption while maintaining high learning performance, offering a scalable, efficient, and privacy-preserving solution for the future of federated learning ecosystems.

The widespread popularity of Federated Learning (FL) has led researchers to delve into its various facets, primarily focusing on personalization, fair resource allocation, privacy, and global optimization, with less attention puts towards the crucial aspect of ensuring efficient and cost-optimized communication between the FL server and its agents. A major challenge in achieving successful model training and inference on distributed edge devices lies in optimizing communication costs amid resource constraints, such as limited bandwidth, and selecting efficient agents. In resource-limited FL scenarios, where agents often rely on unstable networks, the transmission of large model weights can substantially degrade model accuracy and increase communication latency between the FL server and agents. Addressing this challenge, we propose a novel strategy that integrates a knowledge distillation technique with a Particle Swarm Optimization (PSO)-based FL method. This approach focuses on transmitting model scores instead of weights, significantly reducing communication overhead and enhancing model accuracy in unstable environments. Our method, with potential applications in smart city services and industrial IoT, marks a significant step forward in reducing network communication costs and mitigating accuracy loss, thereby optimizing the communication efficiency between the FL server and its agents.

Federated Learning (FL) has emerged as a critical technology for enabling distributed machine learning while preserving data privacy. However, traditional FL faces significant challenges, including communication inefficiencies and reliance on central infrastructures, which lead to increased latency and costs. This paper introduces an innovative FL methodology that addresses these issues by eliminating the dependency on edge servers and employing a server-assisted Proximity Evaluation that dynamically clusters nodes based on data similarity, performance metrics, and geographical proximity. We propose a Hybrid Decentralized Aggregation Protocol that combines local model development with direct peer-to-peer model weight exchanges and centralized aggregation conducted by a dynamically selected driver, substantially reducing global communication overhead. Our system also incorporates Decentralized Driver Selection, Check-pointing to mitigate network congestion, and a Health Status Verification Mechanism to enhance system robustness. Evaluated across five different datasets, our approach demonstrates up to a tenfold reduction in communication requirements, significantly accelerates training speed, and improves energy efficiency, all while maintaining strong learning performance. This methodology offers a scalable, efficient, and secure architecture for the future of federated learning implementations.

The expansion of social media has unlocked a real-time barometer of public opinion. This paper introduces a novel framework to analyze sentiment shifts in social network comment sections, a reflection of the broader public discourse over time. Leveraging a pre-trained uncased RoBERTa model, we predict emotional scores from user comments, mapping these to key sentiment trends such as Approval, Toxicity, Obscenity, Threat, Hate, Offensive, and Neutral. Our methodology employs machine learning techniques to train a dataset that connects emotional scores with these trends, generating trend probability scores. We utilize a bottom-up recursive algorithm to aggregate emotional scores within comment threads, enabling the prediction of trend scores using three distinct aggregation methods. The results demonstrate that our emotional prediction model achieves an AUC of 0.92, and XGBoost stands out with an F1 score exceeding 0.40. Our research elucidates the temporal evolution of online public sentiment, enhancing the understanding of digital social dynamics and offering insights for strategic online interaction, intervention, and content moderation.

Gender identification from social media content can play a crucial role in detecting and mitigating the risks posed by counterfeit accounts. Authentic gender representation can foster a safer and more diverse online environment. While research has been conducted on gender identification in languages such as English, Russian, and Arabic, there remains a gap in studies targeting Bangla and its related languages. This paper introduces a stylometric feature approach to discern the gender of authors from Bangla texts. Utilizing a dataset of 5,000 posts sourced from various Facebook groups, we trained seven traditional machine learning models. Among these, the Random Forest (RF) model notably excelled, achieving an accuracy of 73.37% and an F1-Score of 79.25%, thus setting a promising benchmark in gender identification from Bangla texts.

Identity attacks, such as impersonation, identity theft, and fraudulent account creation, pose significant threats to the security and trustworthiness of Online Social Networks (OSNs). In this paper, we propose a robust and secure framework to verify user identities without compromising their privacy by developing a multi-layered framework leveraging zero-knowledge proof (ZKP) and Hyperledger Fabric private blockchain. We introduce a blockchain-based government identity provider system, coupled with a zero-knowledge proof-based signup process for social networks. Our prototype authenticates user identities in multiple layers, effectively mitigating fraudulent, cloned, and multiple account creations. Our experiments with n (n = 50) users showed a 100% success rate for our system, highlighting its effectiveness compared to other OSNs.

The proliferation of social media offers a real-time reflection of public sentiments. Sentiment analysis on such platforms yields crucial insights for sectors like market research, politics, business strategy, and public health. In this study, we introduce an innovative framework to examine evolving sentiments in social media comments and understand their wider implications. Utilizing a pre-trained BERT base uncase model, we estimate emotional values from comments and align them with various sentiment trends such as Approval, Toxicity, and Neutral, among others. By leveraging machine learning, we train on a distinctive dataset, correlating emotional values with sentiment trends to generate trend likelihood scores. Through a bottom-up methodology, we compile emotional ratings across comment threads to forecast overarching sentiment scores. Our results reveal that the BERT base uncase model excels in emotional prediction, achieving an AUC of 0.91. Meanwhile, Decision Tree models stand out, registering an F1 score above 0.40 on a macro average basis.