Accurate sleep staging is essential for assessing sleep quality and diagnosing sleep disorders, yet it heavily depends on large-scale, expertly labeled datasets, which are costly and time-consuming to produce. While existing methods aim to reduce this reliance, they often utilize data from a limited number of subjects, thereby restricting data diversity and hindering model generalization. To address these challenges, we propose SleepAC, a novel model designed to reduce the dependence on extensive manual annotations. It employs an adaptive sample selection strategy that prioritizes informative and diverse samples, starting with simpler ones and gradually adding more complex ones, while incorporating sleep-specific factors., enabling accurate classification with fewer labeled samples. Furthermore, SleepAC integrates a contrastive learning framework that generates hard negative samples across different sleep stages, effectively enhancing the classification of transitional stages, which are particularly difficult due to limited annotations. Experiments on four public datasets demonstrate that SleepAC achieves competitive accuracy and F1-scores, attaining approximately 95% of the fully supervised performance using only 20% of labeled data. These results underscore its effectiveness in low-resource settings, showcasing promising generalization across complex sleep dynamics while significantly reducing annotation costs.

Graph contrastive learning (GCL) has drawn much research attention for its ability to learn node representations in a self-supervised manner. However, the homophily assumption inherent in GNN encoders limits the direction (macro-level) and the process (micro-level) of message passing in current GCL frameworks, impairing the expressive power of GCL in non-homophilous graphs. This paper presents a novel framework that employs Macro and Micro Message Passing in GCL (M3P-GCL) to overcome these limitations and advance performance in both homophilous and non-homophilous graphs. Specifically, at the macro-level, we integrate structural and attribute views to enhance the direction of message passing, and employ an Aligned Priority-Supporting View Encoding (APS-VE) strategy to facilitate contrastive training; at the micro-level, we propose an Adaptive Self-Propagation (ASP) strategy based on role segmentation of self-loops to diversify the process of message passing in the encoder. These enhancements effectively address the limitations imposed by the homophily assumption. Experiments demonstrate that M3P-GCL outperforms both supervised and unsupervised baselines in the node classification task on various datasets with different levels of homophily.

Contactless palmprint recognition has recently emerged as a promising biometric technology. However, traditional methods that require sharing user data introduce substantial security risks. While federated learning offers privacy-preserving solutions, it often compromises recognition accuracy due to feature distribution drift caused by external factors such as lighting and devices. To address this issue, we propose an adaptive personalized federated learning framework (AdaptPFL). The central innovation lies in decomposing palmprint features into identity-related and contextual-related components using a feature decoupling mechanism. This design isolates the influence of external environmental factors on identity recognition through de-entanglement. Furthermore, two adaptive aggregation strategies are introduced to correct client drift: (1) Intra-Local Adaptive Aggregation (ILAA), which addresses intra-client drift by adaptively combining the two decoupled feature types; (2) Global-Local Adaptive Aggregation (GLAA), which corrects inter-client drift by adaptively aggregating model parameters. Experimental results demonstrate that AdaptPFL achieves superior performance compared to existing state-of-the-art methods.

People have a hard time using cloud computing because of rules concerning privacy and security in fields like healthcare and banking. Fully Homomorphic Encryption (FHE) lets computers work with encrypted data, but it puts a lot of burden on them. This paper introduces SMMHE (Secure Matrix Multiplication with FHE), a novel element-wise approach that enhances SIMD parallelism in contemporary FHE frameworks to mitigate costly rotation operations. SMMHE is 3.98× faster than the most common FHE-based multiplication techniques, according to thorough testing. This big speedup, which cuts the difference in performance between encrypted and plaintext computation by a lot, makes it much easier to use privacy-preserving cloud applications in real life, like classifying medical images. For encrypted MNIST classification, the amortized duration of 26 ms per image is an excellent example of this.

Graph Contrastive Learning (GCL) is a powerful self-supervised technique for learning node and graph representations. However, real-world graph data often exhibit imbalanced class distributions, which pose significant challenges to GCL’s effectiveness. Our experiments show that current state-of-the-art (SOTA) methods perform poorly under imbalanced settings. To address this, we propose a novel GCL framework called AIGCL for imbalanced node classification. This framework automatically and adaptively balances the node representations learned by GCL. Specifically, we introduce a new data augmentation method that retains more information from minority class nodes during graph augmentation. Additionally, we use an imbalance rate adaptive sampling strategy to balance the data. We also incorporate a Variational Graph Autoencoder (VGAE) with an encoder–decoder structure to pretrain the data and generate high-quality pseudo-labels. Our experiments demonstrate that under imbalanced settings, our model improves classification accuracy by 4 %-12 % compared to baseline models, significantly enhancing the performance of minority class nodes.

Time Series Classification (TSC) is a fundamental task in data mining and often suffers from class imbalance, particularly in real-world applications. Traditional methods often fail to capture high-order intrinsic dependencies among time series, especially when minority class samples are scarce. Effectively mining such associations to improve minority-class representation remains a significant challenge. To address this issue, we propose ITS2Graph, a graph-based generative adversarial learning framework that exploits high-order associations for imbalanced time series classification. An auto-encoder is employed to extract latent representations of time series, based on which pairwise similarities are computed to construct a graph, thereby reformulating TSC as a node classification task. To mitigate class imbalance, a graph generator synthesizes minority-class node features and their topological connections, while a Graph Convolutional Network (GCN) discriminator is trained to distinguish real from generated nodes. Experimental results on 22 real-world time series datasets demonstrate that ITS2Graph outperforms existing algorithms in imbalanced time series classification tasks.

Cloud computing offers inexpensive and scalable solutions for data processing, however privacy concerns often hinder the outsourcing of sensitive information. Homomorphic encryption provides a promising approach for secure computations over encrypted data. However, existing models often rely on restrictive assumptions, such as semi-honest adversaries and inaccessible public data. To address these limitations, we introduce the Secure Outsourcing Computation Toolkit (SOCT), which is a novel framework based on the threshold ElGamal cryptosystem. The toolkit employs a dual-server decryption architecture using a (2,2) threshold additively homomorphic ElGamal (TAHEG) algorithm. This architecture ensures that ciphertexts can be decrypted only with the cooperation of both servers, mitigating the risk of data breaches. The TAHEG algorithm requires the input of a secret key for every decryption operation, preventing unauthorized access to plaintext data. Moreover, the key generation process does not burden users with generating or distributing partial secret keys. We provide rigorous security proofs for our threshold ElGamal cryptosystem and associated secure computation functions. Experimental results demonstrate that SOCT achieves significant efficiency gains compared to existing toolkits, making it a practical choice for privacy-preserving data outsourcing.

Current studies mainly rely on overlapping users (who leave trajectories in both cities) as a medium to learn travelers' preference in the target city, however it is unrealistic to find overlapping users when two cities are far apart, thus a severe data scarcity issue exists for this problem. Besides, due to the mixture of mobility pattern from both cities, directly applying the model trained in the source city may lead to negative transfer in the target city. To tackle these issues, in this paper, we conceive and implement a novel framework called CrossPred to predict the cross-city mobility of long-distance travelers in the target city. Specifically, POI features including popularity, textual description, spatial distribution as well as sequential pattern are considered for cross-city POI matching, which further acts as a vital link for jointly modeling native user mobility preference in both source and target cities. Maximum Mean Discrepancy (MMD) is adopted to strengthen the shared POI features among cities and weaken the unique POI features, thereby promoting cross-city POI feature matching. Extensive experiments on real-world datasets demonstrate the effectiveness and superiority of the proposed framework.

This article studies a practically meaningful ship detection problem from synthetic aperture radar (SAR) images by neural network. We broadly extract different types of SAR image features and raise the intriguing question whether these extracted features are beneficial to: 1) suppress data variations (e.g., complex land-sea backgrounds, scattered noise) of real-world SAR images and 2) enhance the features of ships that are small objects and have different aspect (length-width) ratios, therefore resulting in the improvement of ship detection. To answer this question, we propose an SAR-ship detection neural network (called 3SD-Net for short), by newly developing bidirectional coordinate attention (BCA) and multiresolution feature fusion (MRF) and a center point distribution module (CPDM) based on CenterNet. In detail, we first develop BCA to make 3SD-Net focus on ship features as much as possible while ignoring the background noise. Second, we leverage MRF to enhance the spatial information of small-scale ships yet solve the nontrivial problem of small-scale and shallower pixels easily lost after deep convolution in SAR images. Moreover, considering the varying length-width ratio of arbitrary ships, we study the probability distribution around the ship center. We concentrate on enhancing the distribution function of the ship center, thereby significantly improving the performance of the basic CenterNet detector. This improvement is achieved without incurring additional computational and time costs. The experimental results obtained from the public SAR-Ship and SSDD datasets demonstrate the superior performance of our method compared with its competitors. Specifically, our 3SD-Net achieves average precision (AP) values of 91.66% and 90.22% on the two datasets, respectively, outperforming YOLOV7 (90.31% and 87.32%) and EfficientVit (90.06% and 90.08%). Source code will be released upon publication.

Attributed network has more network information, so more and more attention is paid to the embedding of attributed network. A few existing works have considered the node attributes plays a crucial role in the quality of network embedding. They use the non-negative matrix factorization (NMF) method to mine the network information of network structure and node attributes respectively. Considering the reconstruction error of NMF method, the original network information will be lost when the final network embedding is generated. In this paper, we propose a novel multi-view attributed network embedding model with manifold regularization (Mane). The manifold regularization is added to the model to better reflect the Riemann geometry structure of the network in the feature space to enhance the information. And the problem of missing information of NMF is solved. Our approach uses the NMF to get the non-negative coefficient matrix corresponding to network structure and node attributes. Then cooperative regularization and manifold regularization is added to obtain more information in the final network embedding. The model proposed in this paper has been verified by experiments on several real data sets. The result shows that the model is superior to the state-of-the-art algorithm in node classification task.

This paper studies a practically meaningful ship detection problem from synthetic aperture radar (SAR) images by the neural network. We broadly extract different types of SAR image features and raise the intriguing question that whether these extracted features are beneficial to (1) suppress data variations (e.g., complex land-sea backgrounds, scattered noise) of real-world SAR images, and (2) enhance the features of ships that are small objects and have different aspect (length-width) ratios, therefore resulting in the improvement of ship detection. To answer this question, we propose a SAR-ship detection neural network (call SAR-ShipNet for short), by newly developing Bidirectional Coordinate Attention (BCA) and Multi-resolution Feature Fusion (MRF) based on CenterNet. Moreover, considering the varying length-width ratio of arbitrary ships, we adopt elliptical Gaussian probability distribution in CenterNet to improve the performance of base detector models. Experimental results on the public SAR-Ship dataset show that our SAR-ShipNet achieves competitive advantages in both speed and accuracy.

KPI anomaly detection plays an important role in operation and maintenance. Due to incomplete or missing labels are common, methods based on VAE (i.e., Variational Auto-Encoder) is widely used. These methods assume that the normal patterns, which is in majority, will be learned, but this assumption is not easy to satisfy since abnormal patterns are inevitably embedded. Existing debias methods merely utilize anomalous labels to eliminate bias in the decoding process, but latent representation generated by the encoder could still be biased and even ill-defined when input KPIs are too abnormal. We propose a model-agnostic causal principle to make the above VAE-based models unbiased. When modifying ELBO (i.e., evidence of lower bound) to utilize anomalous labels, our causal principle indicates that the anomalous labels are confounders between training data and learned representations, leading to the aforementioned bias. Our principle also implements a do-operation to cut off the causal path from anomaly labels to training data. Through do-operation, we can eliminate the anomaly bias in the encoder and reconstruct normal patterns more frequently in the decoder. Our proposed causal improvement on existing VAE-based models, CausalDonut and CausalBagel, improve F1-score up to 5% compared to Donut and Bagel as well as surpassing state-of-the-art supervised and unsupervised models. To empirically prove the debias capability of our method, we also provide a comparison of anomaly scores between the baselines and our models. In addition, the learning process of our principle is interpreted from an entropy perspective.

Image dehazing is a common operation in autonomous driving, traffic monitoring and surveillance. Learning-based image dehazing has achieved excellent performance recently. However, it is nearly impossible to capture pairs of hazy/clean images from the real world to train an image dehazing network. Most of existing dehazing models that are learnt from synthetically generated hazy images generalize poorly on real-world hazy scenarios due to the obvious domain shift. To deal with this unpaired problem arisen by real-world hazy images, we present Cycle Spectral Normalized Soft likelihood estimation Patch Generative Adversarial Network (Cycle-SNSPGAN) for image dehazing. Cycle-SNSPGAN is an unsupervised dehazing framework to boost the generalization ability on real-world hazy images. To leverage unpaired samples of real-world hazy images without relying on their clean counterparts, we design an SN-Soft-Patch GAN and exploit a new cyclic self-perceptual loss which avoids using the ground-truth image to compute the perceptual similarity. Moreover, a significant color loss is adopted to brighten the dehazed images as human expects. Both visual and numerical results show clear improvements of the proposed Cycle-SNSPGAN over state-of-the-arts in terms of hazy-robustness and image detail recovery, with even only a small dataset training our Cycle-SNSPGAN. Code has been available athttps://github.com/yz-wang/Cycle-SNSPGAN.

The continuous development of wireless communication networks makes the interaction between users become frequent, and the analysis of network users has become a hot topic. Attributed network embedding combines user relationship and user attributes in the network to learn high-quality node representations. However, Existing attributed network embedding method ignore the edge information and the information enhancement of node attributes on the edge, so ultimately high-quality node representation cannot be obtained. In order to make full use of node attributes to enhance network edge information, we propose an Attributed Network Embedding via Edge Information Enhancing(ANEE) method. Our method uses random walk based on network structure and node attributes to obtain sequences. After that, In node sequence learning and edge sequence learning these two sequence learning, we adopt Self-Attention mechanism in multiple latent spaces to learn multiple latent correlation information in the sequence, so as to obtain better node representations. In the node sequence learning, the position information in the sequence is added as a constraint, and the node representation with a reasonable position is learned. In edge sequence learning, the attribute relation of adjacent nodes is added to edge sequence relation as additional information to enhance edge information. The performance of model proposed in this paper has been verified by experiments on several real data sets. The result shows that the model is superior to the state-of-the-art algorithm in link prediction task.

Data are widely collected via the IoT for machine learning tasks in in-home health monitoring applications and mislabeled training data lead to unreliable machine learning models in in-home health monitoring. Researchers have proposed a wide arrangement of algorithms to deal with mislabeled training data, in which one straightforward and effective solution is to directly filter noise from training data so that the negative effects of mislabeled data can be minimized. In essence, noise filtering might be a suboptimal solution because the mislabeled data are not completely useless. The features and distributions of mislabeled data are still useful for learning, especially when training data are insufficient. In this work, we propose a novel framework to learn from mislabeled training data through ambiguous learning (LeMAL). LeMAL mainly consists of two parts. First, it converts the original training data to ambiguous data. Second, an ambiguous learning algorithm is applied to the ambiguous data. In this work, we propose a novel distance-based ambiguous learning algorithm so that the ambiguous data can be used in a better way. Finally, we demonstrate that LeMAL can effectively improve learning performance over existing noise filtering methods.

The data in industrial informatics may be high-dimensional and mislabeled. Irrelevant or noisy features pose a significant challenge to the detection of high-dimensional mislabeling. The traditional method usually adopts a two-step solution, first finding the relevant subspace and then using it for mislabeling detection. This two-step method struggles to provide the optimal mislabeling detection performance, since it separates the procedures of feature selection and label error detection. To solve this problem, in this article, we integrate the two steps and propose a sequential ensemble noise filter (SENF). In the SENF, relevant features are selected and used to generate a noise score for each instance. Continuously, these noise scores guide feature selection in the regression learning. Thus, the SENF falls in the scope of sequential ensemble learning. We evaluate our approach on several benchmark datasets with high dimensionality and much label noise. It is shown that the SENF is significantly better than other existing label noise detection methods.

Resume block classification is the most significant step in resume information extraction. However, the existing algorithms applied to resume block classification are all the general text classification algorithms, which failed to consider the contextual order of each block within a resume. In order to improve the performance of resume block classification, we propose in this paper a block-level bidirectional recurrent neural network model that makes full use of the contextual order relationship among different resume blocks. The experimental results show that the average F1-score value of our model on three 1,400 real resume datasets is 6% to 9% higher than the existing methods.

Network embedding is an important method to learn low-dimensional vector representations of nodes in networks, which has wide-ranging applications in network analysis such as link prediction. Most existing network embedding models focus on the unsigned networks with only positive links. However, networks should have both positive and negative links in practical applications such as the trust and distrust relationships in social networks. It is certain that there are different properties between positive links and negative links, which means the network embedding models designed for unsigned networks are not suitable for signed networks. In this paper, we propose SNE-DML, a signed network embedding model with dynamic metric learning. The model learns positive and negative distance metrics respectively in the training process. We conduct sign prediction experiments on three datasets and compare with seven baselines including three signed network embedding models and four state-of-the-art unsigned network embedding models. The experimental results show the effectiveness of our model.

Self-organizing networks (SON) aim to offer high quality services while reducing both capital expenditure (CAPEX) and operational expenditure (OPEX). SON consists of three main functions: self- configuration, self-optimization, and self-healing. Comparing with self-configuration and self- optimization, there exits only few studies on self- healing. However, it plays an important role in maintaining network operation. Note that self- healing mainly includes fault detection, fault diagnosis, and fault compensation. In this paper, we focus on fault diagnosis and propose an ensemble learning based fault diagnosis system for a self- organizing cellular network. Specifically, in the proposed ensemble learning framework, the base learner is strengthened in each iteration and the final diagnosis result is obtained from the combination of all base classifications. Moreover, traditional classification algorithms are designed considering the premise of balanced data set. However, the classification accuracy of minority classes is not satisfactory. To deal with imbalanced training data sets, we applied the synthetic minority over- sampling technique (SMOTE) in the proposed system, which could also alleviate the difficulties caused by insufficient fault data. Simulation results show that the proposed system can achieve a high diagnosis accuracy, which can be further improved with the increase of training samples. In addition, the diagnosis accuracy of minority fault classes can be significantly improved with the application of SMOTE.

Graph classification is an important topic in graph mining research since it has many applications, such as, social web mining, function prediction of molecules for drug design, XML document classification and anomaly detection in program flows. The key difficulty in graph classification lies in selecting a subset of optimal features from a huge number of structural features. The features need to be highly discriminative and small in numbers for better classification accuracy and running time. In this paper, we propose a novel feature selection framework that selects an optimal feature subset by removing redundant subgraphs. Topologically similar subgraphs have similar discriminative powers and coverage. We cluster these subgraphs and select one subgraph as a feature. We also propose an efficient topological similarity-based clustering method that guarantees the efficiency of our framework. Empirical results show that the proposed framework achieves significantly improved classification accuracy and running time in comparison with the state-of-the-art methods.

A trust-aware recommender system (TARS) is widely used in social media to find useful information. S_Searching is one of the most effective recommender searching mechanisms of TARS. It is based on the scale-freeness of the trust network: a skeleton, which consists of hub nodes of the trust network, is involved in trust propagations. Trusts are first propagated from active users to the skeleton, and then recommenders are searched via the skeleton. One fundamental research issue in S_Searching is to search the skeleton for active users efficiently. Existing methods fully search the trust network to find hub nodes in the skeleton for active users. It has high computational cost. In this paper, we propose a novel iterative deepening-based skeleton searching strategy for S_Searching, in which a depth-limited search is run repeatedly. The depth limit is increased with each iteration until it reaches the maximum allowable trust propagation distance. Simulation results show that the computational complexity of our proposed strategy is much less expensive than that of existing methods.

Summary Today, the management of massive data collections draws much attention as data grids have been developed to deal with large computational problems and provide the opportunity for sharing geographically distributed resources for large‒scale data‒intensive applications. Therefore, finding an effective approach to discover data resources in order to promote better interactions between application communities or virtual organizations becomes a critical challenge. Traditional grid resource discovery models are mostly based on central and hierarchical architecture that can lead to bottlenecking with the expansion of the grid scale. Although the Peer‒to‒Peer (P2P) technique is integrated into the grid in order to improve the performance in recent years, each P2P structure still has drawbacks that require several compensatory strategies. In this paper, based on the unstructured super‒node‒based architecture from the P2P system, we design a structured logic resource tree in each domain in order to effectively alleviate the load on the super‒node, and we propose a query recording learning algorithm based on this hybrid architecture to reduce traffic in the network and greatly shorten the response time. The model and algorithm are validated by simulations and compared with the traditional super‒peer model and the flooding‒based approach. Copyright © 2014 John Wiley & Sons, Ltd.

One fundamental requirement of the trust-aware recommender system (TARS) is to efficiently find as many recommenders as possible for the active users. Existing approaches of TARS choose to search the entire trust network, which have very high computational cost. Though the trust network is the scale-free network, we show via experiments that TARS cannot find satisfactory number of recommenders by directly applying the classical searching mechanism of the scale-free network. This is because it chooses the local highest-degree node at each step of the trust propagation. Since the power of the trust network's degree distribution is not big enough, the selected nodes cannot cover superior number of users. In this paper, we propose an efficient searching mechanism, named S_Searching, for TARS based on the scale-freeness of trust networks: choosing the global highest-degree nodes to construct a Skeleton, and searching the recommenders via this Skeleton. Benefiting from the superior outdegrees of the nodes in the Skeleton, S_Searching can find the recommenders very efficiently. Experimental results show that S_Searching can find almost the same number of recommenders as that of conducting full search, which is much more than that of applying the classical searching mechanism in the scale-free network, while the computational complexity and cost is much less.

Trust-aware recommender system (TARS) suggests the worthwhile information to the users on the basis of trust. Existing works of TARS suffers from the problem that they need extra user efforts to label the trust statements. The authors propose a novel model named iTARS to improve the existing TARS by using the implicit trust networks: instead of using the effort-consuming explicit trust, the easy available user similarity information is used to generate the implicit trusts for TARS. Further analysis shows that the implicit trust network has the small-world topology, which is independent of its dynamics. The rating prediction mechanism of iTARS is based on the small worldness of the implicit trust network: the authors set the maximum trust propagation distance of iTARS approximately equals the average path length of the trust network's corresponding random network. Experimental results show that with the same computational complexity, iTARS is able to improve the existing TARS works with higher rating prediction accuracy and slightly worse rating prediction coverage.

For a learning task with a huge number of training instances, we sample some informative/important instances, which are then used for learning. Obtaining accurately labeling data is always difficult thus noise detection is required to filter out noises from sampled instances since the noises will degrade the learning performance. In this work, we propose to utilize unsampled instances to improve the performance of noise detection in sampled instances. Empirical study validates our idea that refined classifier can be achieved from noisy sampled instances by utilizing unsampled instances.

The performance of supervised learning could be improved when valuable data are selected for training. In this paper, we proposed three data selection methods based on fuzzy C-means algorithm. They are: center-based selection, border-based selection and bin-based selection. In center-based selection, the data with high degree of membership in each cluster are selected for training. In border-based selection, the data around the borders between clusters are selected. In bin-based selection, the data in each cluster are sorted based on their membership degrees. Then for each cluster, the sorted data are divided into bins. Finally, there is one data selected from each bin for training. The effects of them are empirically studied on a set of UCI data sets. Experimental results indicate that bin-based selection could effectively improve the performance of learning compared to randomly selecting training samples.

This paper proposes a novel method for data editing. The goal of data editing in instance-based learning is to remove instances from a training set in order to increase the accuracy of a classifier. To the best of our knowledge, although many diverse data editing methods have been proposed, this is the first work which uses semi-supervised learning for data editing. Wilson editing is a popular data editing technique and we implement our approach based on it. Our approach is termed semi-supervised nearest neighbor editing (SSNNE). Our empirical evaluation using 12 UCI datasets shows that SSNNE outperforms KNN and Wilson editing in terms of generalization ability.

As the cornerstone of effective patient-physician relationships in the traditional healthcare infrastructures, trust faces new opportunities and challenges in the ubiquitous healthcare. Ubiquitous healthcare enables the agents acquire more information on trust evaluation through effectively resource sharing. Yet ubiquitous healthcare also lays the agents in a more dynamic and uncertainty environment for the trust evaluations. This paper contributes to develop a distributed trust management for the ubiquitous healthcare. Our trust management infrastructure is responsible for evaluating the trust value and assigning access rights based on the trust value. Based on each agent's confidence of its personal experience on other agents, three naive Bayes classifier based algorithms are introduced for the trust evaluation: the robust experience algorithm, the weak experience algorithm and the no experience algorithm. The simulation results show the feasibility and effectiveness of our trust management in the ubiquitous healthcare.

Reputation system is a way to maintain trust in dynamic environments by collecting, distributing and aggregating feedbacks about the service providers' past behaviors. Most existing reputation systems assume that raters evaluate the ratee by means of numerical values. However, raters sometimes cannot express their judgments with exact numerical values, especially when the raters have uncertain or ambiguous opinions on the ratee. Our paper introduces a novel reputation system based on the methodology of Computing with Words (CW), in which the ratings and reputations of computation are words and propositions drawn from a natural language instead of numerical values. Our reputation system has a sound mathematical basis. At the same time, it is convenient for the raters to express their judgments and simple for the participants to understand the integrated reputation.

Activity recognition is a hot topic in context-aware computing. In activity recognition, machine learning techniques have been widely applied to learn the activity models from labeled activity samples. Since labeling samples requires human's efforts, most existing research in activity recognition focus on refining learning techniques to utilize the costly labeled samples as effectively as possible. However, few of them consider using the costless unlabeled samples to boost learning performance. In this work, we propose a novel semi-supervised learning algorithm named En-Co-training to make use of the unlabeled samples. Our algorithm extends the co- training paradigm by using ensemble method. Experimental results show that En-Co-training is able to utilize the available unlabeled samples to enhance the performance of activity learning with a limited number of labeled samples.

Many collaborative music recommender systems (CMRS) have succeeded in capturing the similarity among users or items based on ratings, however they have rarely considered about the available information from the multimedia such as genres, let alone audio features from the media stream. Such information is valuable and can be used to solve several problems in RS. In this paper, we design a CMRS based on audio features of the multimedia stream. In the CMRS, we provide recommendation service by our proposed method where a clustering technique is used to integrate the audio features of music into the collaborative filtering (CF) framework in hopes of achieving better performance. Experiments are carried out to demonstrate that our approach is feasible.