Anomalous links such as noisy links or adversarial edges widely exist in real-world networks, which may undermine the credibility of the network study, e.g., community detection in social networks. Therefore, anomalous links need to be removed from the polluted network by a detector. Due to the co-existence of normal links and anomalous links, how to identify anomalous links in a polluted network is a challenging issue. By designing a robust graph structure embedding framework, also called RGSE, the link-level feature representations that are generated from both global embedding view and local stable view can be used for anomalous link detection on contaminated graphs. Comparison experiments on a variety of datasets demonstrate that the new model and its variants achieve up to an average 5.2% improvement with respect to the accuracy of anomalous link detection against the traditional graph representation models. Further analyses also provide interpretable evidence to support the model's superiority.

Network embedding, also known as network repre-sentation, has attracted a surge of attention in data mining and machine learning community as a fundamental tool to treat net-work data. Most existing deep learning-based network embedding approaches focus on reconstructing the pairwise connections of micro-structure, which are easily disturbed by network anomaly or attack. Thus, to address the aforementioned challenge, we pro-pose a novel robust framework for attributed network embedding by preserving Community Information (AnECI). Rather than using pairwise connection-based micro-structure, we try to guide the node embedding by the underlying community structure learned from data itself as an unsupervised learning, as to own stronger anti-interference ability. Specially, we put forward a new modularity function for high-order proximity and overlapped community to guide the network embedding of an attributed graph encoder. We conducted extensive experiments on node classification, anomaly detection and community detection tasks on real benchmark data sets, and the results show that AnECI is superior to the state-of-art attributed network embedding methods.

Sparse representation has been a powerful technique for modeling high-dimensional data. As an unsupervised technique to extract sparse representations, sparse coding encodes the original data into a new sparse code space and simultaneously learns a dictionary representing high-level semantics. Existing methods have considered local manifold within high-dimensional data using graph/hypergraph Laplacian regularization, and more from the manifold could be utilized to improve the performance. In this article, we propose to further regulate the sparse coding so that the learned sparse codes can well reconstruct the hypergraph structure. In particular, we add a novel hypergraph consistency regularization term (HC) by minimizing the reconstruction error of the hypergraph incidence or weight matrix. Moreover, we extend the HC term to multi-hypergraph consistent sparse coding (MultiCSC) and automatically select the optimal manifold structure under the multi-hypergraph learning framework. We show that the optimization of MultiCSC can be solved efficiently, and that several existing sparse coding methods can fit into the general framework of MultiCSC as special cases. As a case study, hypergraph incidence consistent sparse coding is applied to perform semi-auto image tagging, demonstrating the effectiveness of hypergraph consistency regulation. We perform further experiments using MultiCSC for image clustering, which outperforms a number of baselines.

Small Group evolution has been of central importance in social sciences and also in the industry for understanding dynamics of team formation. While most of research works studying groups deal at a macro level with evolution of arbitrary size communities, in this paper we restrict ourselves to studying evolution of small group (size ≤ 20) which is governed by contrasting sociological phenomenon. Given a previous history of group collaboration between a set of actors, we address the problem of predicting likely future group collaborations. Unfortunately, predicting groups requires choosing from (n r) possibilities (where r is group size and n is total number of actors), which becomes computationally intractable as group size increases. However, our statistical analysis of a real world dataset has shown that two processes: an external actor joining an existing group (incremental accretion (IA)) or collaborating with a subset of actors of an exiting group (subgroup accretion (SA)), are largely responsible for future group formation. This helps to drastically reduce the (n r) possibilities. We therefore, model the attachment of a group for different actors outside this group. In this paper, we have built three topology based prediction models to study these phenomena. The performance of these models is evaluated using extensive experiments over DBLP dataset. Our prediction results shows that the proposed models are significantly useful for future group predictions both for IA and SA.

This paper presents a recommendation algorithm based on matrix operations (RAMO), which integrates collaborative filtering algorithm with information network-based approach. RAMO exploits information from different objects to increase the recommendation accuracy. Furthermore, a distributed recommendation algorithm DRAMD is proposed based on matrix decomposition using the framework MapReduce. DRAMD can be run across multiple cluster nodes to reduce the computation time. Test results on MovieLens dataset show that the algorithms not only have better recommendation effectiveness but improve the efficiency of the computation.

Imputation for missing data is an important task of data mining, which may influence the data mining result. In this paper, Missing Categorical Data Imputation Based on Similarity (MIBOS) is proposed to solve this problem. The algorithm defines a similarity model between objects with incomplete data, constructing the similarity matrix of objects and further gets the nearest undifferentiated object sets of each object to impute the missing data iteratively. In the imputing process, the imputed value will be directly applied to the same iteration and the following iterations. Experiments with three UCI benchmark data sets show the improvement of the proposed algorithm from perspectives of complete rate, accuracy and time efficiency.