Graph-level anomaly detection is significant in diverse domains. To improve detection performance, counterfactual graphs have been exploited to benefit the generalization capacity by learning causal relations. Most existing studies directly introduce perturbations (e.g., flipping edges) to generate counterfactual graphs, which are prone to alter the semantics of generated examples and make them off the data manifold, resulting in sub-optimal performance. To address these issues, we propose a novel approach, Motif-consistent Counterfactuals with Adversarial Refinement (MotifCAR), for graph-level anomaly detection. The model combines the motif of one graph, the core subgraph containing the identification (category) information, and the contextual subgraph (non-motif) of another graph to produce a raw counterfactual graph. However, the produced raw graph might be distorted and cannot satisfy the important counterfactual properties: Realism, Validity, Proximity and Sparsity. Towards that, we present a Generative Adversarial Network (GAN)-based graph optimizer to refine the raw counterfactual graphs. It adopts the discriminator to guide the generator to generate graphs close to realistic data, i.e., meet the property Realism. Further, we design the motif consistency to force the motif of the generated graphs to be consistent with the realistic graphs, meeting the property Validity. Also, we devise the contextual loss and connection loss to control the contextual subgraph and the newly added links to meet the properties Proximity and Sparsity. As a result, the model can generate high-quality counterfactual graphs. Experiments demonstrate the superiority of MotifCAR.

Modeling the dynamic interactions between users and items on knowledge graphs is crucial for improving the accuracy of recommendation. Although existing methods have made great progress in modeling the dynamic knowledge graphs for recommendation, they usually only consider the mutual influence between users and items involved in the interactions, and ignore the influence propagation from the interacting nodes (i.e., users and items) on dynamic knowledge graphs. In this article, we propose an influence propagation-enhanced deep co-evolutionary method for recommendation, which can capture not only the direct mutual influence between interacting users and items but also influence propagation from multiple interacting nodes to their high-order neighbors at the same time on the dynamic knowledge graph. Specifically, the proposed model consists of two main components: the direct mutual influence component and the influence propagation component. The former captures direct interaction influence between the interacting users and items to generate the effective representations for them. The latter refines their representations via aggregating the interaction influence propagated from multiple interacting nodes. In this process, a neighbor selection mechanism is designed for selecting more effective propagation influence, which can significantly reduce the computational cost and accelerate the training. Finally, the refined representations of users and items are used to predict which item the user is most likely to interact with. The experimental results on three real-world datasets illustrate that the effectiveness and robustness of PIDKG outperform all state-of-the-art baselines and the efficiency of it is faster than most comparative baselines.

Sequential recommendation is to predict the next item by capturing the item transformation in the user historical sequences. Although existing methods have achieved convincing results in recommendation tasks, they ignore the relation between discontinuous items within sequences or those across different sequences. In this paper, we focus on how to fully capture the potential item relation within and across sequences. Specifically, we propose a novel graph collaborative optimization-based method named GOSR, which not only comprehensively explores the real item relation from many perspectives by constructing and optimizing the item relation graph but also collaborates user-item interaction graph to model the interactive behaviors of users and items. Our model converts the loose item sequences to a tight item relation graph and continuously optimize the graph by comprehensively considering the item relation within and across sequences. That is beneficial to dynamically capture the actual item relation. Furthermore, we employ the attention module to extract the long short-term preferences and characters of users and items by collaborating the user-item interaction graph. Finally, we use the refined representations of users and items to predict the next item that a user is most likely to interact with. To the best of our knowledge, this is the first attempt to explore the item relation by constructing and optimizing the item relation graph for better recommendation. Extensive experiments on three popular benchmark datasets demonstrate that GOSR outperforms several state-of-the-art methods.

Existing anomaly detection models for time series are primarily trained with normal-point-dominant data and would become ineffective when anomalous points intensively occur in certain episodes. To solve this problem, we propose a new approach, called DiffAD, from the perspective of time series imputation. Unlike previous prediction- and reconstruction-based methods that adopt either partial or complete data as observed values for estimation, DiffAD uses a density ratio-based strategy to select normal observations flexibly that can easily adapt to the anomaly concentration scenarios. To alleviate the model bias problem in the presence of anomaly concentration, we design a new denoising diffusion-based imputation method to enhance the imputation performance of missing values with conditional weight-incremental diffusion, which can preserve the information of observed values and substantially improves data generation quality for stable anomaly detection. Besides, we customize a multi-scale state space model to capture the long-term dependencies across episodes with different anomaly patterns. Extensive experimental results on real-world datasets show that DiffAD performs better than state-of-the-art benchmarks.

Graph anomaly detection is attracting remarkable multidisciplinary research interests ranging from finance, healthcare, and social network analysis. Recent advances on graph neural networks have substantially improved the detection performance via semi-supervised representation learning. However, prior work suggests that deep graph-based methods tend to learn spurious correlations. As a result, they fail to generalize beyond training data distribution. In this article, we aim to identify structural and contextual anomaly nodes in an attributed graph. Based on our preliminary data analyses, spurious correlations can be eliminated with causal subgraph interventions. Therefore, we propose a new graph-based anomaly detection model that can learn causal relations for anomaly detection while generalizing to new environments. To handle situations with varying environments, we steer the generative model to manufacture synthetic environment features, which are exerted on realistic subgraphs to generate counterfactual subgraphs. Further, these counterfactual subgraphs help a few-shot anomaly detection model learn transferable and causal relations across different environments. The experiments on three real-world attributed graphs show that the proposed approach achieves the best performance compared to the state-of-the-art baselines and learns robust causal representations resistant to noises and spurious correlations.

Pinpointing the geographic location of an IP address is important for a range of location-aware applications spanning from targeted advertising to fraud prevention. The majority of traditional measurement-based and recent learning-based methods either focus on the efficient employment of topology or utilize data mining to find clues of the target IP in publicly available sources. Motivated by the limitations in existing works, we propose a novel framework named GraphGeo, which provides a complete processing methodology for street-level IP geolocation with the application of graph neural networks. It incorporates IP hosts knowledge and kinds of neighborhood relationships into the graph to infer spatial topology for high-quality geolocation prediction. We explicitly consider and alleviate the negative impact of uncertainty caused by network jitter and congestion, which are pervasive in complicated network environments. Extensive evaluations across three large-scale real-world datasets demonstrate that GraphGeo significantly reduces the geolocation errors compared to the state-of-the-art methods. Moreover, the proposed framework has been deployed on the web platform as an online service for 6 months.

Traditional supervised keyphrase extraction models depend on the features of labelled keyphrases while prevailing unsupervised models mainly rely on structure of the word graph, with candidate words as nodes and edges capturing the co-occurrence information between words. However, systematically integrating all these multidimensional heterogeneous information into a unified model is relatively unexplored. In this paper, we focus on how to effectively exploit multidimensional information to improve the keyphrase extraction performance (MIKE). Specifically, we propose a random-walk parametric model, MIKE, that learns the latent representation for a candidate keyphrase that captures the mutual influences among all information, and simultaneously optimizes the parameters and ranking scores of candidates in the word graph. We use the gradient-descent algorithm to optimize our model and show the comprehensive experiments with two publicly-available WWW and KDD datasets in Computer Science. Experimental results demonstrate that our approach significantly outperforms the state-of-the-art graph-based keyphrase extraction approaches.

Understanding the behavior and characteristics of web users is valuable when improving information dissemination, designing recommendation systems, and so on. In this work, we explore various methods of predicting the ratio of male viewers to female viewers on YouTube. First, we propose and examine two hypotheses relating to audience consistency and topic consistency. The former means that videos made by the same authors tend to have similar male‐to‐female audience ratios, whereas the latter means that videos with similar topics tend to have similar audience gender ratios. To predict the audience gender ratio before video publication, two features based on these two hypotheses and other features are used in multiple linear regression (MLR) and support vector regression (SVR). We find that these two features are the key indicators of audience gender, whereas other features, such as gender of the user and duration of the video, have limited relationships. Second, another method is explored to predict the audience gender ratio. Specifically, we use the early comments collected after video publication to predict the ratio via simple linear regression (SLR). The experiments indicate that this model can achieve better performance by using a few early comments. We also observe that the correlation between the number of early comments (cost) and the predictive accuracy (gain) follows the law of diminishing marginal utility. We build the functions of these elements via curve fitting to find the appropriate number of early comments (approximately 250) that can achieve maximum gain at minimum cost.

According to the classical communication theories, known as Gate keeping and Selective Exposure, individuals tend to have selective behavior when they disseminate and receive information based on their psychological preferences. Selective behavior related to these two theories have been broadly studied separately. While, thanks to the advent of Online Social Networks (OSNs), larger-scale feedback and user information can be collected. In this paper, based on these data, We analyze the correlation among users' properties (such as age, gender, and cultural background) and analyze their selective behavior by tagging users as disseminators and/or audiences in YouTube, Flickr, and Twitter. We find that despite enormous amount of content available in OSNs, users have a comparatively small selective range and do exhibit selective behavior properties. In particular, they pay the most attention to the content published by disseminators that share similar properties, i.e., gender, age, and country. Nonetheless, we also find significant differences and commonalities among the three OSNs with respect to selective behavior. In particular, (i) the proportion and properties of disseminators, audiences, and dual-role users are quite different for the three networks, (ii) the global level of information spread in Flickr is almost two times than that in Twitter and YouTube is approximately the median one, (iii) For a given country, the global level of information spread is different for different OSNs. For a given OSN, it is different for different countries, (iv) despite ubiquitous presence of dual-role users in OSNs, most of such users are very active as either disseminators or audiences, but not both. Our findings are not only useful for understanding these two theories, but also have applications ranging from advertising and recommendation systems to developing predicting models.