Peter Zeng, Weiling Li, Amie J. Paige, Zhengxiang Wang, Panagiotis Kaliosis, Dimitris Samaras, Gregory J. Zelinsky, Susan Brennan, Owen Rambow. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

During conversation, speakers collaborate on spontaneous referring expressions, which they can then re-use in subsequent conversation with the same partner.Understanding such referring expressions is an important ability for an embodied agent so that it can carry out tasks in the real world.This requires integrating and understanding language, vision, and conversational interaction.We study the capabilities of seven state-of-the-art Large Vision Language Models (LVLMs) as overhearers to a corpus of spontaneous conversations between pairs of human discourse participants engaged in a collaborative object-matching task.We find that such a task remains challenging for current LVLMs, which fail to show a consistent performance improvement as they overhear more conversations from the same discourse participants repeating the same task for multiple rounds.We release our corpus and code 1 for reproducibility and to facilitate future research.

Community discovery is a prominent issue in com-plex network analysis. Symmetric non-negative matrix factorization (SNMF) is frequently adopted to tackle this issue. The use of a single feature matrix can depict network symmetry, but it limits its ability to learn node representations. To break this limitation, we present a novel Relaxed Symmetric NMF (RSN) approach to boost an SNMF-based community detector. It works by 1) expanding the representational space and its degrees of freedom with multiple feature factors; 2) integrating the well-designed equality-constraints to make the model well-aware of the network’s intrinsic symmetry; 3) employing graph regularization to pre-serve the local geometric invariance of the network structure; and 4) separating constraints from decision variables for efficient optimization via the principle of alternating-direction-method of multi-pliers. RSN’s effectiveness is verified through empirical studies on six real social networks, show-casing superior precision in community discovery over existing models and baselines.

To achieve collaboration among multiple services, accurately understanding quality of service (QoS) data becomes an important task since the known QoS data varies over time and tends to be extremely sparse, which makes missing QoS data analysis a challenge. Latent Factorization of Tensor (LFT) has shown its potential to capture the pattern of service interactions effectively. However, the existing work has two shortcomings, i.e., a) most of them use simple feature spaces to represent QoS data, which cannot achieve global intermodal correlation, b) the representation ability of LFT models is restricted by the commonly used first-order optimizer due to their bilinear and nonconvex nature. To address the above issues, this work innovatively proposes a Second-order LFT Network (STN) It innovatively uses a fully-connected latent factor space to construct the feature space of each modality of QoS data and employs the principle of Hessian-free optimization for integrating second-order information. Experimental results on the Response-Time industrial QoS dataset show that STN achieves better understanding of QoS data for obtaining higher QoS prediction accuracy than its peers with affordable computational burden.

The efficiency of Intelligent Transport Systems (ITS) runs on high-quality traffic data, however in real-world deployments, sensors failures, communication interruptions or other issues often lead to missing data, which affects the performance of ITS. Aiming at traffic data’s complex spatio-temporal characteristics, although the latent factorization of tensors (LFT) model has been widely used for missing-value completion, its non-convex objective function makes it difficult for first-order optimization methods to approximate high-quality second-order stationary points, therefore limiting the improvement of the completion accuracy. To address the issues, this paper proposes an incomplete tensor complementation model combining Tucker decomposition and second-order optimization strategy to improve the complementation accuracy and convergence stability. To address the issues, this paper proposes a Second-order Latent Factorization of Tensors based on Tucker Decomposition (SLTD), and efficiently solves it via Gauss-Newton approximation, so that it can significantly improve the model performance while keeping the computational cost low. Experimental results on real traffic datasets (in terms of average vehicle speed) from four cities verify the effectiveness of SLTD. Results show that the proposed model outperforms existing prevailing methods in terms of accuracy and provides a better solution for traffic data completion.

Monitoring the slagging status during the tapping process of a converter is an important task in the metallurgical process. Currently, the monitoring of the slagging process during converter tapping mainly relies on manual visual inspection methods, which suffer from problems such as low efficiency, strong subjectivity, and low accuracy. Fortunately, monitoring of the slagging during converter tapping is essentially an object detection task. Considering the harsh scenarios and tiny targets, the object detection model for this task should be carefully designed. To address this issue, this paper proposes an efficient object detection model named ESD-YOLO based on an improved YOLO algorithm. ESD-YOLO aims to enhance detection accuracy and real-time performance, especially for small targets such as minor slag streams in complex scenarios. It integrates feature guidance from DINOv2, to strengthen fundamental feature learning capabilities, while introducing an efficient channel attention module to optimize multi-scale feature information fusion. Additionally, the Normalized Wasserstein Distance (NWD) loss function is employed to further elevate focus on small target features. Experimental results demonstrate that ESD-YOLO achieves 96.8% mAP and 92.6 FPS on an actual converter slagging dataset, outperforming mainstream object detection models and reaching state-of-the-art performance.

Visual information processing is an important part of industrial intelligence. In industrial settings, eliminating camera lens fog-induced blur poses a formidable challenge. Traditional dehazing methods based on imaging principles are difficult to meet the needs of industrial scenes. In recent years, methods based on complex deep networks, such as transformers, have shown better dehazing performance. However, their practicability is restricted by low adaptability to industrial scenarios. It is necessary to realize an effective and lightweight model that can be applied in industrial sites. For such a purpose, an effective image dehazing model named MFDehaz-Net is proposed. MFDehaz-Net has two specially designed components, i.e., Multi-scale Fusion Block and Point-Depth wise Block, which help it achieve a deep fusion of image features of different scales to obtain better global understanding. Besides, to reduce the damage to the original color of the image caused by image dehazing, MFDehaz-Net integrates the supervision signal in the frequency domain with a specially designed loss function. Experimental results demonstrate that MFDehaz-Net outperforms SOTA models in terms of dehazing ability with much shorter inference time.

High-dimensional and incomplete (HDI) data are frequently encountered in big date-related applications for describing restricted observed interactions among large node sets. How to perform accurate and efficient representation learning on such HDI data is a hot yet thorny issue. A latent factor (LF) model has proven to be efficient in addressing it. However, the objective function of an LF model is nonconvex. Commonly adopted first-order methods cannot approach its second-order stationary point, thereby resulting in accuracy loss. On the other hand, traditional second-order methods are impractical for LF models since they suffer from high computational costs due to the required operations on the objective's huge Hessian matrix. In order to address this issue, this study proposes a generalized Nesterov-accelerated second-order LF (GNSLF) model that integrates twofold conceptions: 1) acquiring proper second-order step efficiently by adopting a Hessian-vector algorithm and 2) embedding the second-order step into a generalized Nesterov's acceleration (GNA) method for speeding up its linear search process. The analysis focuses on the local convergence for GNSLF's nonconvex cost function instead of the global convergence has been taken; its local convergence properties have been provided with theoretical proofs. Experimental results on six HDI data cases demonstrate that GNSLF performs better than state-of-the-art LF models in accuracy for missing data estimation with high efficiency, i.e., a second-order model can be accelerated by incorporating GNA without accuracy loss.

The frame-based method is not suitable for counting repetitive actions in event stream, since the framing process will disrupt the temporal information of events. For accurate count of repetitive actions in events, we propose a framework based on threefold ideas: a) converting event stream into time series, b) searching candidates of repetitive actions based on the ascending and descending trends of event time series, and c) checking the candidates with a fast dynamic time warping based method. For accurate counting repetitive actions, an action enhancement method for event time series and a Mann-Kendall test incorporated dynamic candidate selection algorithm are innovatively proposed. The experimental results on artificially synthesized and normally recorded event datasets demonstrate that our framework can count repetitive actions in event stream with high accuracy. All codes, datasets and examples of visualization can be found at https://github.com/ZYL618/action_count_in_events3.

Community detection sheds light on various graph mining tasks such as social recommendation, which is becoming a long-standing issue in the realm of complex network analysis. A non-negative matrix factorization (NMF) is frequently used to tackle this task. However, it builds based on the principle of linear representation and has difficulty in capturing non-linear features from irregularly non-Euclidean data. To address this issue, this study boosts an NMF-based community detector by combining with a graph convolution module, and a novel Graph Convolution and Graph-Laplacian bi-regularized, Symmetric non-negative matrix factorization (GCGS) model is proposed relying on two main ideas: a) taking a graph convolution network (GCN) as a non-linear constraint module on the feature matrix to ensure its non-linearity; and b) adopting graph regularization to preserve the local geometric features of the network topology. A non-negative and multiplicative update (NMU) algorithm is then derived to solve the unified objective function. Extensive experimental results on six real networks indicate that GCGS achieves higher precision in community detection than its peers.

A dynamic weighted graph is frequently encountered in real-world industrial applications like Internet of Things, which can be modeled into a Third-order Incomplete (ToI) tensor. Correspondingly, each element of the ToI tensor represents an observed link of dynamic weighted graph. A nonnegative tensor CP decomposition (NTC)-based link prediction model has proven to be efficient in predicting the missing links of a dynamic weighted graph. However, the learning objective of existing NTC model is usually built via a standard Euclidean distance, which restricts model prediction ability due to its low generalization. To address this issue, this paper presents an Adaptive Nonnegative Tensor CP decomposition (ANTC) model with two ideas include: a) adopting the$\beta$-divergence to build the learning objective for improving the generalization of the model; and b) implementing hyper-parameters self-adaptation via utilizing a differential evolutionary algorithm. Empirical studies on four dynamic weighted graphs generated by a real application illustrate that the proposed ANTC model achieves higher prediction accuracy and computational efficiency than state-of-the-art predictors in predicting the missing links.

We present ViPRA-Haplo, a de novo strain-specific assembly workflow for reconstructing viral haplotypes in a viral population from paired-end next generation sequencing (NGS) data. The proposed Viral Path Reconstruction Algorithm (ViPRA) generates a subset of paths from a De Bruijn graph of reads using the pairing information of reads. The paths generated by ViPRA are an over-estimation of the true contigs. We propose two refinement methods to obtain an optimal set of contigs representing viral haplotypes. The first method clusters paths reconstructed by ViPRA using VSEARCH Deorowicz et al. 2015 based on sequence similarity, while the second method, MLEHaplo, generates a maximum likelihood estimate of viral populations. We evaluated our pipeline on both simulated and real viral quasispecies data from HIV (and real data from SARS-COV-2). Experimental results show that ViPRA-Haplo, although still an overestimation in the number of true contigs, outperforms the existing tool, PEHaplo, providing up to 9% better genome coverage on HIV real data. In addition, ViPRA-Haplo also retains higher diversity of the viral population as demonstrated by the presence of a higher percentage of contigs less than 1000 base pairs (bps), which also contain k-mers with counts less than 100 (representing rarer sequences), which are absent in PEHaplo. For SARS-CoV-2 sequencing data, ViPRA-Haplo reconstructs contigs that cover more than 90% of the reference genome and were able to validate known SARS-CoV-2 strains in the sequencing data.

Complete and valid spatio-temporal traffic data play a vital role in intelligent transportation systems applications, such as congestion avoidance and route guidance. However, traffic data from real-world scenarios is usually incomplete or corrupted due to communication or sensors malfunctions, which makes the traffic analytics difficult. Since traffic data contains complex spatio-temporal patterns, it is very challenging to develop an efficient learning model that can accurately recover incomplete traffic data. To tackle this issue, this work propose a Tucker Decomposition-based Latent factorization of tensors (TDL) model with two interesting ideas: 1) modeling spatio-temporal traffic data as an incomplete third-order tensor and building a Tucker decomposition based learning objective according to the density-oriented principle for precisely recovering missing traffic data; and 2) adopting a proportional-integral-derivative (PID) control principle-incorporated parameters learning scheme for achieving high computational efficiency. Empirical studies on four traffic speed datasets generated from different cities demonstration that the proposed TDL model achieves significant performance gain in both accuracy and computational efficiency compared with state-of-the-art models.

An Undirected Weighted Network (UWN) is frequently encountered in a big-data-related application concerning the complex interactions among numerous nodes. A Symmetric High-Dimensional and Incomplete (SHDI) matrix can smoothly illustrate such a UWN, which contains rich knowledge like node interaction behaviors and local complexes. To extract desired knowledge from an SHDI matrix, an analysis model should carefully consider its topology for describing a UWN's intrinsic symmetry precisely. Representation learning to a UWN borrows the success of a pyramid of symmetry-aware models like a Symmetric Nonnegative Matrix Factorization (SNMF) model whose objective function utilizes a sole Latent Factor (LF) matrix for representing SHDI's symmetry precisely. However, they suffer from the following drawbacks: 1) their computational complexity is high; and 2) their modeling strategy narrows their representation features, making them suffer from low learning ability. Aiming at addressing the above critical issues, this paper proposes a Multi-constrained Symmetric Nonnegative Latent-factor-analysis (MSNL) model with two-fold ideas: 1) introducing multi-constraints composed of multiple LF matrices, i.e., inequality and equality ones into a data-density-oriented objective function for precisely representing the intrinsic symmetry of an SHDI matrix with broadened feature space; and 2) implementing an alternating direction method of multipliers (ADMM)-incorporated learning scheme for efficiently solving such a multi-constrained model. Empirical studies on three SHDI matrices from a real bioinformatics or industrial application demonstrate that the proposed MSNL model achieves higher representation accuracy than state-of-the-art models do, as well as promising computational efficiency.

Service-oriented applications commonly involve high-dimensional and sparse (HiDS) interactions among users and service-related entities, e.g., user-item interactions from a personalized recommendation services system. How to perform precise and efficient representation learning on such HiDS interactions data is a hot yet thorny issue. An efficient approach to it is latent factor analysis (LFA), which commonly depends on large-scale non-convex optimization. Hence, it is vital to implement an LFA model able to approximate second-order stationary points efficiently for enhancing its representation learning ability. However, existing second-order LFA models suffer from high computational cost, which significantly reduces its practicability. To address this issue, this paper presents a Momentum-accelerated Hessian-vector algorithm (MH) for precise and efficient LFA on HiDS data. Its main ideas are two-fold: a) adopting the principle of a Hessian-vector-product-based method to utilize the second-order information without manipulating a Hessian matrix directly, and b) incorporating a generalized momentum method into its parameter learning scheme for accelerating its convergence rate to a stationary point. Experimental results on nine industrial datasets demonstrate that compared with state-of-the-art LFA models, an MH-based LFA model achieves gains in both accuracy and convergence rate. These positive outcomes also indicate that a generalized momentum method is compatible with the algorithms, e.g., a second-order algorithm, which implicitly rely on gradients.

Occupational pneumoconiosis (OP) staging is a vital task concerning the lung healthy of a subject. To perform artificial intelligence (AI)-assisted OP staging via chest X-ray image representational learning and classification commonly adopted to address it, where a Convolutional Neural Network (CNN) has proven to be efficient. However, unlike commonly encountered image classification tasks, OP staging relies heavily on the profusion level of opacities. The opacities in chest X-ray overlap with other tissues in the lung area and are hard to be represented by a standard CNN, thereby leading to inaccurate staging results. Aiming at implementing accurate determination of the opacities caused by pneumoconiosis, this study incorporates a dark channel prior method into a bi-branch learning structure, thereby establishing a Bi-branch Dark Channel Differential Convolutional Neural Network (BDCNN) for accurate AI-assisted OP staging. Its ideas are two-fold: a) extracting opacities caused by pneumoconiosis from chest X-ray with a dark channel prior-based dehazing method, and b) realizing multiple feature fusion via a bi-branch structure to ensure high staging accuracy. Experimental results on six real OP data cases demonstrate that the proposed BDCNN outperforms state-of-the-art models in obtaining accurate staging results for occupational pneumoconiosis.

Latent Factor (LF) models are highly effective in representing high-dimensional and incomplete (HDI) matrices. Hessian free (HF) optimization is an efficiency second-order algorithm to minimize the object function in LF models. An HF-based second-order LF model can achieve better accuracy representation results than first-order ones with affordable computational burden. However, its low rank representation ability relies on a more complex training process, multiple hyper-parameters work cooperatively to decide the results. Thus, these hyper-parameters should be turned with care since they are mutually influenced. The heavy hyper-parameters turning work reduces the practicability of a second-order LF model. To address this issue, this study incorporates the principle of particle swarm optimization into the second-order LF model to propose an adaptive second-order LF (ASLF) model. Experimental results on three HDI matrices reveal that ASLF model can be fine-tuned adaptively with acceptable computation burden.

A non-negative latent factor analysis (NLFA)-based recommender can make precise recommendations by correctly representing the non-negative characteristic of industrial data. It commonly relies on a nonconvex and bilinear optimization process, where the effects of first-order solvers maybe significantly reduced. Higher order solvers like a Newton-type method are expected to make a breakthrough; however, its computation efficiency and scalability are greatly limited due to the numerous parameters involved in a Hessian matrix. To address this issue, this article proposes an approach for assimilating second-order information for building NLFA-based recommenders. The key idea is an inner second-order solver that employs a Hessian-free method for avoiding the highly expensive manipulations of a Hessian matrix. Empirical studies on eight data cases emerging from real industrial applications indicate that the proposed approach outperforms state-of-the-art models in prediction accuracy with affordable computational burden.

User-side Quality-of-Service (QoS) data are vital for efficient cloud service selection, while available QoS data in real applications are commonly described by a high-dimensional and Sparse (HiDS) matrix due to a) increasing users and services, and b) the impossibility of observing the full invoking mapping among users and services. A latent factor analysis (LFA) model has proven to be efficient in performing representation learning on such an HiDS QoS matrix. However, existing LFA models commonly adopt a first-order optimizer that cannot makes it approach the second-order stationary point of the learning objective, thereby resulting in accuracy loss. Aiming at addressing this issue, this study proposes to incorporate a Generalized Nesterov's Acceleration (GNA) method in to a Hessian-vector algorithm for LFA, thereby establishing a GNA-incorporated Hessian-vector-based LFA (GNHL) model with two-fold ideas: a) adopting the principle of a Hessian-vector method to acquire a proper Newton step efficiently, and b) incorporating a GNA method into its linear search for accelerating its convergence rate. Experimental results on six real QoS datasets demonstrate that a GNHL model outperforms state-of-the-art LFA models in generating highly accurate predictions for missing QoS data with low computational burden.

Large-scale undirected networks representation can be formulated into a non-convex optimization problem, which can be efficiently addressed by a symmetric non-negative latent factor analysis (SNLFA)-based approach. However, an SNLFA model commonly adopts a first-order optimization algorithm that cannot well handle its non-convex learning objective, thereby resulting in inaccurate representation to a target network. On the other hand, higher-order learning algorithms like a Newton-type one is expected to solve a non-convex optimization problem well, but its computation efficiency and scalability are greatly limited due to its direct manipulation of the Hessian matrix that can be huge in an SNLFA model. To address this issue, this paper proposes a Truncated Newton-method-based Symmetric Non-negative Latent Factor Analysis (TNS) model for utilizing second-order information in the latent factor analysis process without manipulating the Hessian matrix, thereby achieving high computational efficiency and scalability. Empirical studies indicate that TNS outperforms state-of-the-art models in prediction accuracy with an affordable computational burden.

Protein-protein interactions (PPIs) play a critical role in cellular activities. However, discover them experimentally is exhausted. How to predict the missing PPIs with the known PPI networks (PPINs) is a challenging task considering their large scale and extreme sparsity. To utilize the known data efficiently, this work proposes a Highly-Confident Protein Interactome Prediction (HPIP) model with two-fold ideas: a) employing a variational autoencoder model based on Gaussian distribution as a basic PPI predictor for processing the sparse input PPIN; b) embedding the basic PPI predictor into an elastic architecture which based on a block and probability summing strategy. Experimental results on two real world PPINs from the STRING database indicate that HPIP can effectively predict PPIs with high confidence.

Most existing researches on cloud workflow systems have focused on resource scheduling with the aims to minimize system delay under budget constraints or optimize system cost under deadline constraints. However, cloud providers cannot guarantee a failure-free cloud environment, a compact scheduling plan is prone to failure, thus, workflow system reliability has been identified as a critical and challenging issue in the volatile cloud environment. With the ability of cloud, it is easy for users to implement the active fault tolerance schemes, e.g., Scale-Out. However, it will lead to issues like security problem and extra management cost. In this paper, we first investigate Scale-Up and Scale-Hybrid schemes to fully explore the possibilities offered by the ability of cloud. We formally model the problem of optimizing the reliability of a cloud workflow system under budget constraints with these three fault-tolerance schemes. These optimization problems are discrete and non-convex. Thus, we propose a genetic algorithm based method for workflow fault tolerance (GA4WFT). Finally, we evaluate the effectiveness and efficiency of proposed GA4WFT with three different fault-tolerance schemes through experiments conducted on Amazon EC2 data.

Precisely understanding high-dimensional and sparse (HiDS) user-item interactions is the most important issue in a recommender system. A latent factor analysis (LFA)-based model has proven to be efficient in addressing it, while current models of this kind mostly rely on first-order optimizers. It is vital to implement an LFA-based model able to approach the second order stationary points efficiently for improving its representative learning ability. To do so, this work presents a Generalized-momentum-accelerated Hessian-vector Algorithm (GHA) for HiDS data. Its main idea includes a) adopting the principle of a Hessian-vector-product-based method to avoid operating a Hessian matrix directly, and b) incorporating a generalized momentum method into its parameter learning process for further enhancing its ability in approaching a stationary point. Experimental results on two industrial datasets demonstrate that when compared with state-of-the-art LFA-based models, a GHA-based LFA model achieves gains in accuracy and convergence rate. These positive outcomes also indicate that a generalized momentum method is compatible with algorithms implicitly relying on gradients like a second-order algorithm.

Human Endogenous Retrovirus type K (HERV-K) is the only HERV known to be insertionally polymorphic; not all individuals have a retrovirus at a specific genomic location. It is possible that HERV-Ks contribute to human disease because people differ in both number and genomic location of these retroviruses. Indeed viral transcripts, proteins, and antibody against HERV-K are detected in cancers, auto-immune, and neurodegenerative diseases. However, attempts to link a polymorphic HERV-K with any disease have been frustrated in part because population prevalence of HERV-K provirus at each polymorphic site is lacking and it is challenging to identify closely related elements such as HERV-K from short read sequence data. We present an integrated and computationally robust approach that uses whole genome short read data to determine the occupation status at all sites reported to contain a HERV-K provirus. Our method estimates the proportion of fixed length genomic sequence (k-mers) from whole genome sequence data matching a reference set of k-mers unique to each HERV-K locus and applies mixture model-based clustering of these values to account for low depth sequence data. Our analysis of 1000 Genomes Project Data (KGP) reveals numerous differences among the five KGP super-populations in the prevalence of individual and co-occurring HERV-K proviruses; we provide a visualization tool to easily depict the proportion of the KGP populations with any combination of polymorphic HERV-K provirus. Further, because HERV-K is insertionally polymorphic, the genome burden of known polymorphic HERV-K is variable in humans; this burden is lowest in East Asian (EAS) individuals. Our study identifies population-specific sequence variation for HERV-K proviruses at several loci. We expect these resources will advance research on HERV-K contributions to human diseases.

The OWL-based Web Service ontology is one of the most important standards for semantic service composition. Performance analysis of composite service processes specified in OWL-S enables us to understand whether the process meet the SLA requirements. In this work, we propose a Petri-net-based formal framework for OWL-S processes using non-markovian-stochastic-petri-nets (NMSPN) as the intermediate representation. The main innovation of this research includes a translation from OWL-S to non-markovian-stochastic-petri-nets and a well-defined control flow model for composite services built on OWL-S.

Radial basis function (RBF) networks are one of the most widely used models for function approximation and classification. There are many strange behaviors in the learning process of RBF networks, such as slow learning speed and the existence of the plateaus. The natural gradient learning method can overcome these disadvantages effectively. It can accelerate the dynamics of learning and avoid plateaus. In this letter, we assume that the probability density function (pdf) of the input and the activation function are gaussian. First, we introduce natural gradient learning to the RBF networks and give the explicit forms of the Fisher information matrix and its inverse. Second, since it is difficult to calculate the Fisher information matrix and its inverse when the numbers of the hidden units and the dimensions of the input are large, we introduce the adaptive method to the natural gradient learning algorithms. Finally, we give an explicit form of the adaptive natural gradient learning algorithm and compare it to the conventional gradient descent method. Simulations show that the proposed adaptive natural gradient method, which can avoid the plateaus effectively, has a good performance when RBF networks are used for nonlinear functions approximation.

The data of Semantic Web exist in machine readable format called RDF, in order to promote data exchange on the web based on their semantics. As an expressive knowledge representation language for the Semantic Web, Web Ontology Language (OWL) plays an important role in modeling information in a semantic way. However, due to the nature of knowledge bases, ontologies tend to be very large, distributed, and interconnected. Thus, maintaining constraints and enforcing data consistency for a group of ontologies become very challenging. In addition, frequent updates on ontologies necessitate an automatic approach to checking for potential constraint violations before any change takes place. In this study, we conducted a pioneer study and presented a framework for checking global constraints and ensuring integrity on data that span multiple ontologies. As an update is issued to a single site, global constraints that can be potentially violated are broken down into sub constraints that only involve a very small subset of ontologies. The checking of sub-constraints runs effectively in parallel and returns results about each subset. The collection of these results determines the violation of global constraints.