Decentralised Finance (DeFi) applications involve a large volume of funds and exhibit diverse user behaviours, including malicious activities such as smart contract exploits and financial scams. Existing approaches struggle to capture complex behaviours. To address this gap, we propose a general Blockchain User Behaviour Analysis (BUBA) pipeline for DeFi security. The pipeline presents an automated action formation process that takes blockchain transactions as inputs and outputs user actions. In addition, BUBA introduces a dual Graph Neural Network (GNN) model that jointly captures user action features, contract and token interactions, and heterogeneous graph structure information to produce rich behavioural embeddings, enabling effective clustering of semantically meaningful user behaviours. We evaluate the proposed pipeline on Uniswap V3, where it outperforms baseline methods in identifying and differentiating suspicious behaviours. A further case study on Sushiswap V2 demonstrates the generalisability of the pipeline across DeFi applications.
Oshani Seneviratne, Fernando Spadea, Adrien Pavao, Aaron Micah Green · 5 authors
Temporal Web analytics increasingly relies on large-scale, longitudinal data to understand how users, content, and systems evolve over time. A rapidly growing frontier is the \emph{Temporal Web3}: decentralized platforms whose behavior is recorded as immutable, time-stamped event streams. Despite the richness of this data, the field lacks shared, reproducible benchmarks that capture real-world temporal dynamics, specifically censoring and non-stationarity, across extended horizons. This absence slows methodological progress and limits the transfer of techniques between Web3 and broader Web domains. In this paper, we present the \textit{FinSurvival Challenge 2025} as a case study in benchmarking \emph{temporal Web3 intelligence}. Using 21.8 million transaction records from the Aave v3 protocol, the challenge operationalized 16 survival prediction tasks to model user behavior transitions.We detail the benchmark design and the winning solutions, highlighting how domain-aware temporal feature construction significantly outperformed generic modeling approaches. Furthermore, we distill lessons for next-generation temporal benchmarks, arguing that Web3 systems provide a high-fidelity sandbox for studying temporal challenges, such as churn, risk, and evolution that are fundamental to the wider Web.
Bikki Kumar, Dev Karan, Adarsh Kandu, Ashish Khari
The social impact of decentralised online communities, such as blockchain-based social networks, is complex because their decentralisation allows users to exercise greater freedom and independence. A novel Dynamic Graph Neural Network with Temporal Knowledge Distillation (DGNN-TKD) is proposed to model and predict influence patterns. DGNN-TKD differs from standard Graph Neural Networks (GNNs), which function under the assumption of static graphs. It tracks the temporal evolution of a graph, and introduces a knowledge distillation mechanism that enables the transfer of influence embeddings over time. We propose a novel multi-dimensional influence metric that captures agent reputation, engagement and trust, supplemented with robust attention-based temporal aggregation. In experiments on decentralized social network datasets, DGNN-TKD surpasses current dynamic GNNs in influence prediction, community detection, and misinformation detection in decentralized governance/Web3 applications. This framework connects graph-based learning and social dynamics and serves as a powerful tool to study decentralized phenomena.
Centralized e-commerce recommenders face privacy risks, while Federated Recommendation Systems (FRS) suffer from accuracy loss in sparse environments and rely on untrusted aggregators. We propose BL-ZPRS, a framework utilizing bilayer zk-SNARKs for end-to-end trustworthiness. Its lower-layer User-to-Anchor (U2A) paradigm restores collaborative signals via verifiable vectors without exposing raw data, while an upper-layer ZKP proves FedAvg integrity. Evaluations on the Amazon Review dataset show BL-ZPRS achieves accuracy comparable to centralized models with superior resistance to poisoning attacks, effectively balancing privacy and integrity.
Non-fungible tokens (NFTs) have become a key asset class in Web3 markets, where visual artwork, textual narratives, and on-chain transaction patterns jointly determine value, yet their pricing dynamics remain volatile, opaque, and difficult to explain. Existing NFT valuation methods typically either ignore the multimodal nature of NFTs or treat assets as independent samples, failing to exploit the rich relational structures induced by shared creators, collections, and ownership patterns, and offering limited interpretability for high-stakes financial decisions. To address these challenges, we propose NFT-Insight, a multimodal graph transformer framework that unifies visual, textual, and blockchain information on a heterogeneous NFT graph and explicitly links structural and content signals to valuation behavior. The framework identifies closely related NFTs via a joint similarity measure in the multimodal embedding space, propagates information through a relation-specific graph attention network and a global transformer encoder, and adopts a regularization strategy that encourages consistent valuations for highly similar assets while still allowing data-driven differentiation. In addition, NFT-Insight integrates attention-based and SHAP-based explanations into a unified analysis pipeline, enabling joint study of valuation behavior and feature attributions at the level of related NFT pairs. Experiments on three large-scale, real-world NFT datasets show that NFT-Insight consistently outperforms strong unimodal, multimodal, and graph-based baselines, reducing MAE and RMSE by up to about 20% in static valuation (withR2up to 0.904), achieving robust cross-market performance with averageR2≈ 0.84 andr≈ 0.93, and attainingR2= 0.911 in temporal forecasting. Interpretability analysis reveals that visual, textual, blockchain, and graph-relational features achieve a high alignment between SHAP importance and attention weights (average Spearman correlation above 0.8), and case studies highlight meaningful valuation patterns driven by rarity, speculative trading, and temporal market shocks. Overall, the proposed framework offers a multimodal graph–based perspective on explainable NFT valuation and market forecasting, and provides a general template for incorporating complex relational and content interactions into graph-based learning in decentralized digital economies.
Lourdusamy Selvam, S. Abarna, R. Santhana Krishnan, Azarudeen K · 6 authors
Ensuring vaccine integrity requires rigorous cold-chain management, as even minor temperature deviations can compromise safety, cause financial losses, and endanger public health. Conventional monitoring systems, dependent on centralized databases and manual oversight, remain prone to delays, inaccuracies, and tampering. To address these limitations, we propose an integrated framework that combines deep learning and blockchain for secure, predictive, and real-time cold-chain monitoring. A distributed network of IoT sensors captures temperature, humidity, vibration, and GPS data at one-minute intervals. Long Short-Term Memory (LSTM) networks forecast short-term temperature trends, Autoencoders (AE) detect anomalies, and one-dimensional Convolutional Neural Networks (1D CNN) classify shipment states as Safe, At Risk, or Spoiled. A decision engine fuses outputs from these models into a unified risk score, enabling timely and data-driven interventions. Critical events and metadata are immutably recorded on the Ethereum blockchain, while raw sensor data is stored off-chain using IPFS to ensure auditability and efficiency. The system also supports automated alerts, real-time monitoring through Grafana dashboards, and adaptive model retraining for continuous improvement. Evaluation covers forecasting accuracy, anomaly detection precision, classification performance, and blockchain efficiency. By uniting predictive analytics with tamper-proof logging, the proposed framework reduces vaccine spoilage, strengthens supply chain resilience, and offers a scalable solution for broader temperature-sensitive logistics applications.
With the rise of Web3, Non-Fungible Tokens (NFTs) have become a new class of digital assets, driving demand for large-scale NFT recommendation systems. Each NFT can be associated to a rich set of semantic, stylistic, and thematic labels, forming a highly complex label space. Similar to e-commerce platforms where detailed product labels enable personalized recommendations, such semantic dependencies between labels can potentially enhance NFT recommendation performance. Thus, NFT recommendation can be naturally formulated as an extreme multi-label (XML) classification problem. Many existing probabilistic label tree (PLT)-based approaches address XML problem by recursively partitioning the label space, which greatly alleviates the demands on expensive computer resources. Yet, the highly skewed distribution of labels in datasets in XML makes tail labels more challenging to predict than head labels. In this paper, Our preliminary analysis reveals that inherent label dependencies can be leveraged to improve tail label recommendations for NFTs. We propose ChainTail, a dependency-aware framework that enhances PLT-based NFT label partitioning and prediction re-scoring. It includes: (1) a Dependency-aware partition module that partitions highly dependent NFT labels into subsets. (2) a Dependency-aware ReScore module that re-ranks prediction scores of labels to eliminate the label-priors. Our experimental results show that ChainTail boosts tail label recommendation on widely used item recommendation datasets.
There is a lack of standardized datasets for NFT (Non-Fungible Token) recommendation systems. This study presents a comprehensive dataset designed for NFT recommendation systems, incorporating both NFT-related data (e.g., images, textual descriptions, rarity and transaction data) and user-related data (e.g., purchase price, transaction duration, and NFT holding period). To create the dataset, a Data Collection Tool was developed to gather raw data via the OpenSea API, and a Data Preparation Tool was implemented for preprocessing and filtering. All data used in this study are publicly available and anonymized, ensuring that user privacy is fully preserved. The dataset is evaluated using NFT-NCFAE, a deep learningbased NFT recommendation model, with performance measured by Recall and NDCG evaluation metrics. The evaluation results demonstrate the suitability and value of the proposed dataset for NFT recommendation systems. By making the dataset and its associated tools publicly available, this work aims to establish a benchmark for future research and enable comparability across different models.
Social media platforms rely heavily on user interaction data to personalize content and advertisements, raising concerns regarding user privacy and data misuse. Although regulations such as the General Data Protection Regulation (GDPR) aim to address these concerns, enforcement remains under the control of the platforms themselves. To address these issues, this paper proposes a privacy-preserving recommender system that minimizes personal preference data exposure while maintaining the potential for equivalent personalization accuracy (including for advertisements) through local data processing, which could access the same amount or more of pure user data than server-side models. This system integrates a lightweight client-based machine learning model to infer user preferences locally, combined with Merkle tree-based Zero-Knowledge Proof (ZKP) scheme to anonymously authenticate user requests. The authors develop a working web app prototype and evaluate performance across a range of user devices. Results show minimal latency for the client-based model (under 5 milliseconds on most devices) and diverse proof generation times, ranging from 2.6 seconds to over 18 seconds, depending on hardware capability. Server-side verification remains consistent and fast under 250 milliseconds. Although proof generation latency remains a bottleneck for real-time applications, optimization strategies such as proof-caching, cross-application preferences synchronization, and native implementation provides a promising path toward privacy-preserving personalization in social media systems.
To our knowledge, this paper is the first to introduce an iterative semantic-clustering framework driven by Large Language Models (LLMs) for refined user profiling on Uniswap V4. Traditional numerical techniques such as K-Means falter in high-dimensional feature spaces and identify only three coarse user groups. In contrast, our LLM-based pipeline repeatedly refines clustering criteria through natural-language reasoning, producing 21 semantically interpretable sub-clusters. These fine-grained clusters can be hierarchically aggregated into three macro categories that align with the K-Means result, thereby preserving global consistency while revealing nuanced behavioral motivations and patterns. The proposed framework offers both a theoretical perspective and practical toolkit for decentralized-finance user analysis, opening new avenues for understanding on-chain behavior.
Decentralised energy ecosystems suffer from data-governance, scalability and adoption barriers. Although blockchain-based marketplaces can offer transparency and security in Local Energy Communities (LECs), most existing solutions struggle with rigid token models, limited performance, and steep usability barriers. Building on a previous framework, this study presents an enhanced marketplace that integrates a modular blockchain layer, custodial identity management, and a dual-token model for flexible licensing and pricing. Testing on a per-missioned Quorum network with asynchronous queueing demonstrated notable improvements in transaction throughput and user responsiveness under load, while the custodial onboarding flow simplified access for non-technical participants. A refined policy enforcement mechanism further aligns the system with emerging federation standards, mitigating earlier shortcomings related to performance, data sovereignty, and scalability. Benchmarking on a five-node Quorum Proof of Authority (PoA) deployment (one RPC node and four validator nodes) showed that all key operations, including license issuance and asset usage, consistently completed in under 8 seconds, confirming the system’s suitability for possible energy data applications. Planned extensions include cross-domain interoperability, self-service governance tools, and zero-knowledge proofs, underscoring this architecture’s potential as a robust, future-ready platform for federated energy data ecosystems.
Federated Learning (FL) is an innovative learning paradigm that allows multiple devices to collaboratively train a shared model without uploading the raw data to the cloud, thereby enhancing privacy and security. Leveraging Mobile Edge Computing (MEC), Hierarchical Federated Learning (HFL) can further reduce the communication overhead, thereby increasing the efficiency and scalability of FL systems by enabling model aggregation at the network edge. However, this framework often encounters security challenges, such as single points of failure and the risk of malicious model tampering. To address these challenges, researches have employed blockchain technology to enhance the security of FL systems, but most of these solutions incur significant resource burdens due to the intensive computation demands of blockchain consensus mechanisms, such as Proof-of-Work (PoW). In this work, we aim to explore a lightweight blockchain-enabled federated learning (LBFL) framework that utilizes the Delegated Proof-of-Stake (DPoS) consensus mechanism, which employs a simple voting process to elect a small number of candidate block producers (known as delegates) to aggregate the FL model and produce blocks. This framework significantly reduces the number of consensus nodes, thereby minimizing resource consumption during the consensus process. We study the joint optimization of mobile device association, bandwidth allocation, computing frequency management, and block producer selection, aiming to minimize the overall delay and energy consumption. To address the challenges posed by discrete and continuous decision variables, we decompose the problem into three sequential subproblems and solve them iteratively. Simulation results show that compared with the existing benchmarks, the proposed scheme can reduce overall delay and energy consumption by 15% to 22%.
In order to solve the problem of low security of high network information sharing, the architecture and implementation of agent-based information sharing and pushing system are proposed. Firstly, the information node registration algorithm is introduced to draw the organization network connected to the public network into an orderly block network and extract the abnormal information of network nodes; Secondly, a complete network identity security threat intelligence is generated through effective scheduling and transmission of network identity security threat intelligence; Finally, the distributed ledger technology is introduced to match it with IP network, and the private chain and data are extracted by constructing data structure on the information chain to realize the generation and operation of intelligence information sharing model. The experimental results show that the model designed in this paper can protect the privacy information that all network users need to protect in the process of sharing information resources, and its privacy protection intensity exceeds 0.950. Conclusion: The proposed sharing model can realize the safe sharing of user intelligence information in practical application, and the user’s private intelligence information will not be leaked during the sharing process, which reduces the threat degree of network security.
In its present form, traditional Web2 domain infrastructures have become vulnerable to increasing amounts of censorship, security breaches, and centralized authority exclusively handled by third party Certificate Authorities (CAs). These weaknesses and the single points of failure inherent, under spotlight give out the necessity of migrating to a safer decentralized Web3 model. Mygrate was developed using solidity, Hedera and hardhat. This paper elucidates Mygrate development, features, and applications. It offers a functionality to bridge Web2 domains with Web3 architecture by utilizing SSL certificates in combination with smart contracts to map domains to cryptographically signed InterPlanetary File System (IPFS) addresses, verified over decentralized blockchains such as Hedera. By solving a major part of the oracle problem, Mygrate emerged as an innovative solution, leveraging the power of Hedera Blockchain. This paper delves into the comprehensive development journey of Mygrate. It allows us to bring web2 hosted domain name source of trust and offset authenticity to web3 for easy identifiability. Moreover, there is increased security, decreased oracle fees, and money goes directly to the data providers who provide data as a public good. Through the integration of libraries like Ethereum-Waffle, ethers, chai,node-forge, Mygrate offers Oracle-free SSL verification, EVM assembly optimization, Frontrunning mitigation, gas optimization and security.
Federated Learning (FL) enables decentralized model training while ensuring data privacy between clients and the server, and is applicable in various domains such as healthcare, finance, and edge computing. Traditional FL approaches assume homogeneous model architectures across clients and servers, limiting their applicability in real-world scenarios where clients use diverse models. To address this, we introduce Feature Align-HFL, a novel framework designed for Heterogeneous Federated Learning (HFL) that aligns dissimilar models for effective knowledge sharing. Specifically, we demonstrate its application by aligning CNNs and U-Nets for a common classification task. By employing cosine similarity, Feature Align-HFL identifies and aligns common features across heterogeneous models. Experiments on the CIFAR-10 dataset show that CNN and adapted U-Net models achieved 90 percent and 86 percent accuracy, respectively, with feature map similarities reaching up to 96 percent. These results highlight the significant transferable knowledge between dissimilar architectures, indicating that Feature Align-HFL can effectively support knowledge sharing in heterogeneous FL settings.
As the next generation of the internet, Web 3.0 is expected to revolutionize the Internet and enable users to have greater control over their data and privacy. Federated learning (FL) enables data to be usable yet invisible during its use, thereby facilitating the transfer of data ownership and value. However, the issues of data size and blockchain computing power are of paramount importance for FL in Web 3.0. Due to the openness of Web 3.0, individuals can freely join or leave training and adjust data size, creating population uncertainty and making it difficult to design incentive mechanisms. Therefore, we propose a Poisson game-based FL incentive mechanism that motivates participants to contribute more data and computing power, considering the variability of data size and computing power requirements, and provides a feasible solution to the uncertainty of the number of participants using a Poisson game model. Additionally, our proposed FL architecture in Web 3.0 integrates FL with Decentralized Autonomous Organizations (DAO), utilizing smart contracts for contribution calculation and revenue distribution. This enables an open, free, and autonomous federated learning environment. Experimental evaluation shows that our incentive mechanism is feasible in blockchain with efficiency, robustness, and low overhead.
In the era emphasizing the privacy of personal data, verifiable federated learning has garnered significant attention as a machine learning approach to safeguard user privacy while simultaneously validating aggregated result. However, there are some unresolved issues when deploying verifiable federated learning in edge computing. Due to the constraint resources, edge computing demands cost saving measurements in model training such as model pruning. Unfortunately, there is currently no protocol capable of enabling users to verify pruning results. Therefore, in this paper, we introduce PrVFL, a verifiable federated learning framework that supports model pruning verification and heterogeneous edge computing. In this scheme, we innovatively utilize zero-knowledge range proof protocol to achieve pruning result verification. Additionally, we first propose a heterogeneous delayed verification scheme supporting the validation of aggregated result for pruned heterogeneous edge models. Addressing the prevalent scenario of performance-heterogeneous edge clients, our scheme empowers each edge user to autonomously choose the desired pruning ratio for each training round based on their specific performance. By employing a global residual model, we ensure that every parameter has an opportunity for training. The extensive experimental results demonstrate the practical performance of our proposed scheme.
Abstract In Web3 social platforms, i.e. social web applications that rely on blockchain technology to support their functionalities, interactions among users are usually multimodal, from common social interactions such as following, liking, or posting, to specific relations given by crypto-token transfers facilitated by the blockchain. In this dynamic and intertwined networked context, modeled as a financial network, our main goals are (i) to predict whether a pair of users will be involved in a financial transaction, i.e. the transaction prediction task , even using textual information produced by users, and (ii) to verify whether performances may be enhanced by textual content. To address the above issues, we compared current snapshot-based temporal graph learning methods and developed T3GNN, a solution based on state-of-the-art temporal graph neural networks’ design, which integrates fine-tuned sentence embeddings and a simple yet effective graph-augmentation strategy for representing content, and historical negative sampling. We evaluated models in a Web3 context by leveraging a novel high-resolution temporal dataset, collected from one of the most used Web3 social platforms, which spans more than one year of financial interactions as well as published textual content. The experimental evaluation has shown that T3GNN consistently achieved the best performance over time and for most of the snapshots. Furthermore, through an extensive analysis of the performance of our model, we show that, despite the graph structure being crucial for making predictions, textual content contains useful information for forecasting transactions, highlighting an interplay between users’ interests and economic relationships in Web3 platforms. Finally, the evaluation has also highlighted the importance of adopting sampling methods alternative to random negative sampling when dealing with prediction tasks on temporal networks.
Koki Koshikawa, Jong-Deok Kim, Won–Joo Hwang, Kien Nguyen · 5 authors
Blockchain holds significant potential in addressing the security, privacy, decentralization, and interoperability challenges prevalent in the Internet of Things (IoT), However, this advancement often comes at the expense of scalability. Therefore, enhancing the scalability of IoT blockchain systems while preserving other essential blockchain attributes is imperative. This paper aims to mitigate network latency in the blockchain network, a factor directly linked to blockchain scalability. Achieving this goal requires implementing an efficient peer selection method, moving beyond the reliance on default selection (i.e., the one in Bitcoin, Ethereum, etc.). In existing literature, Perigee has been introduced as a method that nearly optimizes the delay in the transaction transmission process. However, Perigee has not comprehensively addressed the complete transaction life cycle, which includes a crucial process-block transmission. In response to this limitation, we propose Dual Perigee, a solution that thoroughly considers and optimizes both transaction-oriented latency (TOL) and block-oriented latency (BOL). To show the effectiveness of Dual Perigee, we implemented and evaluated it within an emulated IoT-Blockchain system, comparing its performance with Perigee and the default peering method in Ethereum. The results reveal that Dual Perigee excels in reducing BOL compared to Perigee. Moreover, Dual Perigee exhibited a latency that was 43% and 80% lower than the default peering method and Perigee, respectively.