This paper presents FAMDR, a Feature-Aligned Multimodal Denoising framework for Reliable Diagnostic Reconciliation. Existing approaches suffer from two major limitations: (1) an overemphasis on simplifying observational descriptions and (2) a failure to denoise the misleading content in radiological findings against clinical histories. Current methods often dismiss such cross-modal inconsistencies as noise rather than clinically significant signals. To bridge this gap, the framework integrates four synergistic components: (1) noise-aware multimodal alignment that preserves discriminative discrepancy features while ensuring semantic coherence, (2) cross-modal retrieval augmentation leveraging external medical knowledge to resolve ambiguous cases, (3) granular localization of noises at pixel and phrase levels using adaptive thresholding, and (4) medical noise uncertainty quantification to provide reliable confidence estimates. Evaluated on an extended MIMIC-CXR dataset enriched with expert-annotated noise and longitudinal records, FAMDR achieves superior accuracy in semantic denoising and inconsistency localization while preserving clinical interpretability. Its capability to generate actionable, uncertainty-aware reports advances safer and more reliable integration into diagnostic workflows.

Pruning is a promising approach to reduce the high inference cost of large language models (LLMs), but it often comes at the expense of performance. Motivated by the "functional localization" theory in neuroscience, we hypothesize that LLMs contain task-specific expert activation paths, where specific subsets of neurons are co-activated for particular tasks. This structure allows selective activation to preserve task performance while improving inference efficiency. We introduce Sparse Expert Activation Pruning (SEAP), a training-free pruning method for large language models. SEAP identifies task-relevant activation paths by analyzing the clustering patterns of hidden states and neuron activations on a multi-task calibration dataset. Cross-task transfer evaluations confirm the existence of such expert activation structures. SEAP constructs task-aware pruning masks by leveraging a task-expert calibration dataset, which provides representative samples across diverse tasks to reveal their activation signatures. It then employs a lightweight task router to dynamically select relevant computation paths based on the input task. This design significantly reduces inference cost without compromising accuracy. Experimental results show that SEAP retains model performance with only a 1.5% drop on most tasks at 20% sparsity, and at 50% sparsity, it surpasses strong pruning baselines such as WandA and FLAP by over 20%. These results highlight SEAP as a scalable and effective solution for efficient LLM inference.

Retrieval-augmented generation (RAG) is widely adopted for knowledge-intensive tasks, but unverified external knowledge can pose risks such as data injection and retrieval pollution, leading to unexpected generation. Existing defenses rely on patch-based fixes, which limit generalization and increase system latency. To address these issues, we propose RAG2RAG, a framework-level security solution specifically designed for RAG. Inspired by human intuition to reason about what can and cannot be said during RAG phase, RAG2RAG augments the main RAG module with a lightweight RAG-based security expert module composed of two components: (1) a Detective that dynamically retrieves supporting evidence, and (2) a Judge that makes final decisions based on retrieved context. The main and expert modules operate in parallel without causing noticeable delays. Experiments across two languages, six domains, and seven types of poisoning attacks demonstrate that RAG2RAG overall achieves higher accuracy and lower attack success rates than seven mainstream baselines. Furthermore, it integrates seamlessly with various RAG architectures, offering efficient protection across diverse threat scenarios.

Graph augmentation is a cornerstone of effective graph contrastive learning, yet existing methods often rely on random designed perturbations, which may distort latent semantics and impair representation quality. In this work, we argue that semantic consistency can be effectively approximated by low-frequency components in the spectral domain, offering a principled proxy for guiding augmentation. Based on this insight, we propose Frequency-Aware Graph Contrastive Learning (FA-GCL), a novel framework that explicitly preserves low-frequency signals while selectively perturbing high-frequency components. By aligning augmentation with frequency-aware decomposition, FA-GCL generates diverse yet semantically coherent views, mitigating semantic drift and enhancing representational discrimination. Extensive experiments across multiple benchmarks demonstrate that FA-GCL consistently outperforms state-of-the-art baselines with statistically significant gains, validating its exclusive merits.

Retrieval-augmented generation (RAG) provides an efficient solution for expanding the knowledge boundaries of large language models (LLMs), where the indexing serves as a compass to guide LLMs in locating query-relevant external knowledge. Nevertheless, current indexing methods commonly encounter a critical challenge: native indexing is convenient to construct, but it usually disrupts contextual associations and constrains the expressive capacity of rich knowledge. Conversely, knowledge indexing can structure contextual knowledge, but it is often based on preset schemas that limit its generalizability. To address it, we propose a universal and flexible knowledge indexing called pseudo-graph (PG) indexing. During the indexing construction phase, we use the advanced LLMs to transform the knowledge of each raw text into a concise and structured mind map, organizing intra-document knowledge. Subsequently, independent mind maps are linked by associating highly relevant topics or consistent facts across documents, thereby establishing inter-document knowledge connections. Eventually, using the resulting knowledge network PG as the knowledge indexing can circumvent the challenges associated with schema design reliant on preset knowledge and relationship types. During the knowledge retrieval phase, we develop a PG knowledge retriever to mimic human note-reviewing, adaptively navigating and recalling query-relevant knowledge from PG. Experimental results demonstrate that retrieving relevant pseudo-subgraphs from the PG via PG indexing and retriever significantly improves performance in fact-based Q&A, hallucination correction, and two multi-document Q&A tasks, achieving$F1_{QE}$improvements of 15.85%, 8.12%, 3.34%, and 5.73%, respectively, and outperforming the state-of-the-art baseline KGP-LLaMA. Our code is available at:https://github.com/IAAR-Shanghai/PGRAG.

Cross-domain recommendations in healthcare services differ from traditional ones in electronic commerce due to the need for heightened medical privacy protection for a small group of users, while ensuring the majority, who may lack sufficient medical knowledge, can understand the recommendations. To recommend doctors who provide online consultations to health video viewers and enable multimodal cross-domain recommendations from short video platforms (source domain) to online healthcare communities (target domain), this paper introduces a framework based on the User-Centric Synthetic Data Architect (UCSDA) and Pre-trained Large Language Model (PtLLM). UCSDA employs a user-centric, advanced selection-synthesis mechanism to filter users' cold interaction items and synthesize noise items, reducing privacy leakage risk. PtLLM focuses on necessary patient and doctor IDs during the recommendation decision process to generate explanations. The model's effectiveness and scalability were validated using three public datasets and a healthcare cross-domain recommendation dataset. In addition to traditional evaluation metrics, strong privacy metrics and the unique sentence ratio were used to assess privacy protection and interpretability. We also compared the characteristics of privacy protection and interpretability between e-commerce and healthcare recommendation scenarios.

The significance of Temporal Knowledge Graphs (TKGs) in Artificial Intelligence (AI) lies in their capacity to incorporate time-dimensional information, support complex reasoning and prediction, optimize decision-making processes, enhance the accuracy of recommendation systems, promote multimodal data integration, and strengthen knowledge management and updates. This provides a robust foundation for various AI applications. To effectively learn and apply both static and dynamic temporal patterns for reasoning, a range of embedding methods and large language models (LLMs) have been proposed in the literature. However, these methods often rely on a single underlying embedding space, whose geometric properties severely limit their ability to model intricate temporal patterns, such as hierarchical and ring structures. To address this limitation, this paper proposes embedding TKGs into projective geometric space and leverages LLMs technology to extract crucial temporal node information, thereby constructing the 5EL model. By embedding TKGs into projective geometric space and utilizing Möbius Group transformations, we effectively model various temporal patterns. Subsequently, LLMs technology is employed to process the trained TKGs. We adopt a parameter-efficient fine-tuning strategy to align LLMs with specific task requirements, thereby enhancing the model's ability to recognize structural information of key nodes in historical chains and enriching the representation of central entities. Experimental results on five advanced TKG datasets demonstrate that our proposed 5EL model significantly outperforms existing models.

Xun Liang, Simin Niu, Zhiyu Li, Sensen Zhang, Hanyu Wang, Feiyu Xiong, Zhaoxin Fan, Bo Tang, Jihao Zhao, Jiawei Yang, Shichao Song, Mengwei Wang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Retrieval-augmented citation generation (RACG) helps users trust the large language model output by retrieving evidence from reliable sources. However, most current RACG research focuses on single-language tasks, particularly in English, and overlooks the need for cross-lingual evidence retrieval and utilization in real-world applications. To address this issue, we introduce a plug-and-play Retrieval-Augmented Multilingual Citation Generation method (RAMCG) which uses a multilingual retriever to identify relevant evidence from a multilingual knowledge base. The evidence is then combined with the query and processed by a multilingual citation generator. The result is citations that are both accurate and comprehensive. Experiments show that RAMCG outperforms baseline methods in multilingual citation generation and is well-suited for practical use.

The rapid advancement of Large Language ModelS (LLMs) has inaugurated a transformative era in natural language processing, fostering unprecedented capabilities in text generation, understanding, and contextual analysis. However, effectively handling extensive contexts, which are crucial for many applications, remains a significant challenge due to the intrinsic limitations of the context window sizes of the models and the computational burdens associated with their operations. This study proposes an innovative framework that uses unsupervised natural language summarization to provide more efficient context handling. Our methodology is dubbed LONGPC. We demonstrate that our framework significantly reduces computational overhead and enhances LLMs’ performance across various datasets while maintaining or even enhancing the quality of the generated content.

Large Language Models, even when specifically trained to process long input contexts, struggle to capture relevant information located in the middle of their input. This phenomenon is known as the "lost-in-the-middle" problem. In this study, We propose a new method Found In The Distribution (FITD) with Latent Dirichlet Allocation (LDA) to solve the problem of "lost-in-the-middle". Specifically, we first apply the LDA to capture the critical information within the text. Based on the captured results, we generate corresponding probabilities and use a masking method to compress the prompt, enabling LLMs to correctly focus on these important pieces of information, even if they are situated in the middle. Finally, we demonstrate that our approach performs better in identifying pertinent information within long contexts.

Graph Meta-learning methods have improved the performance of few-shot node classification by means of applying meta-learning to the data in non-Euclidean domains. However, most works focus on adopting a single domain, ignoring the fact that tasks in various domains may be distinct, which can cause overfitting problems and thus limit generalizability. To tackle this challenge, we propose a novel Graph Meta-learning framework called Feature-Enhanced Cross-domain Graph Meta-learning that consists of two crucial modules: 1) A feature information extraction module that aims to capture discriminative node importance and simulate various node feature distributions under distinct domains; 2) A heterogeneous graph encoder module that leverages the enhanced node features and topological information to generate task-specific node embeddings with simple fine-tuning. Moreover, we meta-learn the parameters involved to ensure the generalizability in the unseen domains. Results show that our method markedly outperforms the existing state-of-the-art methods.

Online Medical Consultations (OMCs) are increasingly prevalent in Online Healthcare Communities (OHCs), with a considerable focus on doctor recommendations. However, existing research has predominantly centered on recommending individual specialized doctors, neglecting the exploration of more reliable interdisciplinary options for collaborative consultations. This paper introduces an innovative framework for interdisciplinary doctor recommendations for collaborative work in OMC scenarios, comprising two modules: one for calculating expertise and another for collaborative computing. Leveraging the medical knowledge graph, the former employs an expertise encoder to derive specialty embeddings for patient queries, doctor profiles, and historical consultations. The latter builds a collaborative circle for each doctor based on three dimensions: scientific citation, academic collaboration, and spatiotemporal proximity. This circle analysis delves into the potential for cooperation between doctors, aiming to identify optimal combinations. Experimental results on two real datasets demonstrate the superior performance of our proposed model compared to state-of-the-art methods. Ablation studies highlight the significant contribution of the three social network dimensions and their interactions to collaborative computing outcomes. Our model exhibits robustness across different parameter settings, as demonstrated through experiments. The study's results empower practitioners to develop a portfolio of recommendations, enhancing the effectiveness of cooperative consultations for complex diseases requiring multidisciplinary collaboration.

Cross-domain Graph Meta-learning (CGML) has shown its promise, where meta-knowledge is extracted from few-shot graph data in multiple relevant but distinct domains. However, several recent efforts assume target data available, which commonly does not established in practice. In this paper, we devise a novel Cross-domain Data Augmentation for Graph Meta-Learning (CDA-GML), which incorporates the superiorities of CGML and Data Augmentation， has addressed intractable shortcomings of label sparsity, domain shift, and the absence of target data simultaneously. Specifically, our method simulates instance-level and task-level domain shift to alleviate the cross-domain generalization issue in conventional graph meta-learning. Experiments show that our method outperforms the existing state-of-the-art methods.

Detecting anomalies on attributed graphs is a challenging task since labelled anomalies are highly labour-intensive by taking specialized domain knowledge to make anomalous samples not as available as normal ones. Moreover, graphs contain complex structure information as well as attribute information, leading to anomalies that can be typically hidden in the structure space, attribute space, and the mix of both. In this paper, we propose a novel model for graph anomaly detection named ProGAD. Specifically, ProGAD takes advance of label propagation to infer high-quality pseudo labels by considering the structure and attribute inconsistencies between normal and abnormal samples. Meanwhile, ProGAD introduces the prior knowledge of class distribution to correct and refine pseudo labels with a prototype-aware strategy. Experiments demonstrate that ProGAD achieves strong performance compared with the current state-of-the-art methods.

Researchers have applied knowledge graph embedding (KGE) techniques with advanced neural network techniques, such as capsule networks, for predicting drug-drug interactions (DDIs) and achieved remarkable results. However, most ignore molecular structure and position features between drug pairs. They cannot model the biomedical field's significant relational mapping properties (RMPs,1-N, N-1, N-N) relation. To solve these problems, we innovatively propose CDHse that consists of two crucial modules: 1) Entity embedding module, we obtain position feature obtained by PubMedBERT and Convolutional Neural Network (CNN), obtain molecular structure feature with Graphic Nuaral Network (GNN), obtain entity embedding feature of drug pairs, and then incorporate these features into one synthetic feature. 2) Knowledge graph embedding module, the synthetic feature is Householder projections and then embedded in the complex vector space for training. In this paper, we have selected several advanced models for the DDIs task and performed experiments on three standard BioKG to validate the effectiveness of CDHse.

Xun Liang, Shichao Song, Simin Niu, Zhiyu Li, Feiyu Xiong, Bo Tang, Yezhaohui Wang, Dawei He, Cheng Peng, Zhonghao Wang, Haiying Deng. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

The rapid development of Knowledge Graph (KG) technology has led to the emergence of Temporal Knowledge Graphs (TKGs), which hold significant research importance and value. Temporal Knowledge Graph Embedding (TKGE) techniques complement TKGs and predict links within them. The efficacy of TKGE hinges upon its capability to effectively model intrinsic temporal relation patterns. However, existing methodologies often need to capture temporal relation patterns adequately or establish intrinsic connections between evolving relations. We propose a highly robust TKGE framework called HouRP to address this limitation and incorporate a more comprehensive range of relational information. This framework introduces two types of Householder transformations into TKGE. Specifically, the Householder projections enable the generation of temporal-specific representations for each entity. At the same time, the relational Householder rotations facilitate high-dimensional rotations between projected entities, thereby capturing relation-specific properties. Our proposed model’s effectiveness is demonstrated through extensive experimentation on four widely used TKGs datasets.

Multimodal summarization for videos aims to craft concise representations from a plethora of information sources, such as videos and audio transcripts. Encouragingly, notable progress has been achieved in this domain. Nevertheless, existing works appear superficial when it comes to modal fusion, and the training process demands substantial computational power. This serves as a catalyst for our proposition of a novel Deeper and Faster Fusion Network (DFFN) for multimodal summarization, which aims to deeply integrate multimodal sources and expedite the training process of generating summaries. The proposed model consists of a forget fusion encoder and a cross-modal fusion decoder. Specifically, the forget fusion encoder employs a forget gate fusion mechanism for multi-mode deep fusion, which demonstrates a superior capability compared with the shallow fusion of each mode. Furthermore, the cross-modal fusion decoder utilizes one-dimensional convolution to augment the model’s comprehension of allegorical meaning and incorporates a multimodal-aware adapter to accelerate the training of multimodal summarization. Experimental results on the reorganized How2 dataset corroborate that the proposed model surpasses previous approaches both in terms of effectiveness and efficiency for multimodal summarization, especially obtaining a 264% improvement in terms of the efficiency.

Anomaly detection on graphs has recently attracted considerable attention due to its broad range of high-impact applications, including cybersecurity, financial transactions, and recommendation systems. Although many efforts have thus far been made, how to effectively handle the high inconsistency between users' behavior and labels, a fundamental issue in anomaly detection, has not yet received sufficient concern. Moreover, the inconsistency problem is hard to investigate and even deteriorates the performance of anomaly detectors. To this end, we propose a novel graph self-supervised learning framework, Capsule Graph Infomax (termed CapsGI), to overcome the inconsistency of anomaly detection. Inspired by the recent advances of capsules on images, we explore another possibility of reforming the node embedding by capsule ideas to represent the unique node's properties. Concretely, by disentangling heterogeneous factors underlying each node representation, we can establish node capsules such that their representation can reflect intrinsic node properties. To strengthen the connection among normal nodes, CapsGI further represents the part-whole contrastive learning between lower-level capsules (part) and higher-level capsules (whole) by explicitly considering the context graph relations. Extensive experiments on multiple real-world datasets demonstrate that our model significantly outperforms state-of-the-art models.

Node attribute forecasting has recently attracted considerable attention. Recent attempts have thus far utilize dynamic graph convolutional network (GCN) to predict future node attributes. However, few prior works have notice that the complex spatial and temporal interaction between nodes, which will hamper the performance of dynamic GCN. In this paper, we propose a new dynamic GCN model named meta-DGCN, leveraging meta spatial-temporal tasks to enhance the ability of dynamic GCN for better capturing node attributes in the future. Experiments show that meta-DGCN effectively modeling comprehensive spatio-temporal correlations between nodes and outperforms state-of-the-art baselines on various real-world datasets.

This paper studies the problem of exploring the user intent for session-based recommendations. Its challenges come from the uncertainty of user behavior and limited information. However, current endeavors cannot fully explore the mutual interactions among sessions and do not explicitly model the complex high-order relations among items. To circumvent these critical issues, we innovatively propose a HyperGraph Convolutional Contrastive framework (termed HGCC) that consists of two crucial tasks: 1) The session-based recommendation (SBR task) that aims to capture the beyond pair-wise relationships between items and sessions. 2) The self-supervised learning (SSL task) acted as the auxiliary task to boost the former task. By jointly optimizing the two tasks, the performance of the recommendation task achieves decent gains. Experiments on multiple real-world datasets demonstrate the superiority of the proposed approach over the state-of-the-art methods.

Alzheimer’s disease (AD) is a progressive neurodegenerative disease which has a serious impact on patients’ daily lives. Early detection, diagnosis and treatment of AD remain a major challenge in clinical practice. Current methods typically construct brain functionally connectivity networks based on similarities between regions of interest (ROIs) for each individual, neglecting the potential information from identical ROIs. We contend that considering similarities among identical ROIs holds crucial implications for AD classification. To address this issue, we propose a multi-view brain network construction method, which constructs brain network based on the similarity of identical ROIs for AD diagnosis. Firstly, the time series of the same ROIs of all subjects are extracted and recombined into a new set. For the newly composed set the brain connectivity network is obtained based on the similarity between the time series. Next, this constructed network undergoes graph convolutional network (GCN) processing to produce node embeddings. Finally, these embeddings are extracted for all ROIs within each subject, converted into one-dimensional feature vectors, and utilized for disease detection with support vector machine (SVM). To demonstrate the effectiveness of our method, we evaluate the performance of the proposed method in the widely recognized Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, and the results show that our method exhibits higher performance in identifying AD and MCI compared to existing methods.

Prevalence of fake news has seriously disrupted the information ecosystem and undermined social stability and public trust. It stimulated the development of automatic fake news detection method to tackle this dilemma. Most of the methods can be divided into two types, content-based and propagation-based methods. However, these methods overlook the historical background information of the news events contained in the target news in varying degrees. The environment constructed by the historical context of events related to a news event can reflect the direction of the public’s expectations of the current event with respect to future developments. The expectation can be used to study the direction of fake news that has not been fully explored. Therefore, considering the influence of public expectation, we propose a general fake news detection method based on cross-domain coarse-fine grained environments referred to as CFGE including three parts. Specifically, we first construct a cross-domain coarse-fine grained environment with the news related to the domains which were contained in the target news. Then the coarse-grained embedding and the fine-grained embedding of cross-domain environment are extracted respectively by the proposed environment information capture modules. Finally, based on the gate fusion method, coarse-grained embedding and fine-grained embedding are fused to predict fake news. Extensive experimentation substantiates the superiority of CFGE compared to alternative models, further affirming the efficacy of coarse-fine grained environments.

Despite the significant advances in RGB-D scene recognition, there are several major limitations that need further investigation. For example, simply extracting modal-specific features neglects the complex relationships among multiple modalities of features. Moreover, cross-modal features have not been considered in most existing methods. To address these concerns, we propose to integrate the tasks of cross-modal matching and modal-specific recognition, termed as Matching-to-Recognition Network (MRNet). Specifically, the cross-modal matching network enhances the descriptive power of the recognition network via a layer-wise semantic loss. The recognition network obtains multi-modal features from a two-stream CNN: global features are obtained by a higher-layer of a CNN to preserve the semantic content, and local layout features are learned by the graph attention network, thus better capturing the key object regions and modelling their relationships. Extensive experiments results demonstrate the MRNet achieves superior performance to state-of-the-art methods, especially for recognition solely based on single modality.

In this paper, we propose a novel embedded feature selection method, which is able to select the informative and discriminative features with the underlying submanifolds of data in intra-class being well preserved so as to improve the classification performance. Specifically, we first impose the l2,1-norm on both loss function and projection matrix with the aim to suppress the influence of noise and achieve the sparse features selection. Then, a retargeted learning technique is introduced into our model to further enhance the discriminant ability of the projection matrix. Moreover, we design an adaptive graph regularizer to fully utilize the intra-class local submanifold information, which can endow the projection matrix with more locality preserving power and prevent overfitting. What is more, our method jointly optimizes problems of the projection learning, target learning and graph learning instead of separately learning them as in most of existing works, which guarantees an overall optimality in algorithmic performance. Experimental results on both synthetic and real-world image data sets demonstrate the superiorities of our method on local submanifold structure exploration and classification task.

In this paper, we propose a novel semi-supervised dimensionality reduction method based on local structured feature learning with dynamic maximum entropy graph. The proposed method first learns a local discriminative embedding subspace from labeled data for preserving the intrinsic sub-manifold structure in each class by virtue of dynamic maximum entropy graph technique. Then, another auto-optimized k-nearest neighbor graph is constructed at learned embedded subspace to smooth the manifold of all labeled and unlabeled data such that each labeled sample and its neighbored unlabeled samples can be clustered into a same sub-manifold and possess the same label information. Most importantly, in order to guarantee an overall optimum, subspace learning and local structure graph optimizing are performed simultaneously rather than treat them as two irrelevant steps as done in most of graph-based semi-supervised learning methods, which can avoid the negative effects brought by noisy and redundant features. Experimental results on several real-world benchmarks demonstrate the superiorities of our method on sub-manifold structure exploration and classification task.

Graph contrastive learning (GCL) has emerged as a powerful tool to address real-world widespread label scarcity problems and has achieved impressive success in the graph learning domain. Albeit their remarkable performance, most current works mainly focus on designing sample augmentation methods, while the effect of negative sample selection strategy is largely ignored by previous works but rather practical and significant for graph contrastive learning. In this paper, we study the impact of negative samples on learning graph-level representations, and innovatively propose a Reinforcement Graph Contrastive Learning (ReinGCL) for negative sample selection. To be concrete, our model consists of two major components: a graph contrastive learning framework (GCLF), and a selection distribution generator (SDG) for producing the selection probabilities based on RL. The key insight is that Re-inGCL attempts to leverage SDG to guide GCLF and narrow the divergence between the augmented positive pairs, so as to further improve graph representation learning. Extensive experiments demonstrate that our approach significantly yields superior performance compared to the state-of-the-art.

Session-based recommendation (SBR) aims to predict the user’s action at the next timestamp according to an anonymous yet short interaction sequence (i.e., session). Almost all the existing SBR solutions for user preference are only based on the current session without exploiting the high-order relations among other sessions, which may restrict the SBR representation ability and even deteriorate the performance. To this end, we propose a Hyper-relation alignment hyperGraph Convolutional Network, called Hyra-GCN, for better inferring the user preference of the current session. Specifically, we first model session-based data as a hyper-graph capable of representing high-order relationships to exploit item transitions over sessions in a more subtle manner. Subsequently, we explore self-supervised learning on item-session hypergraphs, so as to alleviate the problem of data sparsity. Experimental results on real-world datasets demonstrate the effectiveness of our proposed Hyra-GCN against state-of-the-art baselines.

This paper introduces ChinaOpen, a dataset sourced from Bilibili, a popular Chinese video-sharing website, for open-world multimodal learning. While the state-of-the-art multimodal learning networks have shown impressive performance in automated video annotation and cross-modal video retrieval, their training and evaluation are primarily conducted on YouTube videos with English text. Their effectiveness on Chinese data remains to be verified. In order to support multimodal learning in the new context, we construct ChinaOpen-50k, a webly annotated training set of 50k Bilibili videos associated with user-generated titles and tags. Both text-based and content-based data cleaning are performed to remove low-quality videos in advance. For a multi-faceted evaluation, we build ChinaOpen-1k, a manually labeled test set of 1k videos. Each test video is accompanied with a manually checked user title and a manually written caption. Besides, each video is manually tagged to describe objects / actions / scenes shown in the visual content. The original user tags are also manually checked. Moreover, with all the Chinese text translated into English, ChinaOpen-1k is also suited for evaluating models trained on English data. In addition to ChinaOpen, we propose Generative Video-to-text Transformer (GVT) for Chinese video captioning. We conduct an extensive evaluation of the state-of-the-art single-task / multi-task models on the new dataset, resulting in a number of novel findings and insights.

Imbalanced graph classification is ubiquitous yet challenging in many real-world applications. Existing methods typically follow the same convention of treating graph instances as discrete individuals and exploit graph neural networks (GNNs) to predict graph labels. Despite their success, they only propagate intra-graph information within a single graph while disregarding extra supervision globally derived from other graphs. In fact, the inter-graph learning plays a vital role in providing more supervision for minority graphs. However, it is disadvantageous to accurately derive reliable inter-graph supervision because the redundancy information from majority graphs is introduced to obscure the representations of minority graphs during the propagation process. To tackle this issue, we propose a novel method that integrates the restricted random walk kernel with the global graph information bottleneck (GIB) to improve imbalanced graph classification. Specifically, the restricted random walk kernel is proposed to perform the inter-graph learning with learnable graph filters and produce kernel outputs. To ensure that the redundant information of majority graphs does not plague kernel outputs, we model the entire kernel learning as a Markovian decision process and employ the global GIB manner to optimize it. Extensive experiments on real-world graph benchmark datasets verify the competitive performance of the proposed method.

Temporal Knowledge Graph Embedding (TKGE) is a crucial technique for performing Temporal Knowledge Graph Completion (TKGC). The effectiveness of TKGE largely depends on the ability to model intrinsic relation patterns. However, as we know, most existing TKGE models usually embed KGs into a single geometric space such as Euclidean, hyperbolic or hyperspherical space to maintain their specific geometric structures (e.g., chain, hierarchy, and ring structures). None of the existing methods can simultaneously model relation patterns of chain, hierarchy, ring structures, and relation mapping properties. This paper constructs a hybrid interaction TKGE model HyIE, which learns spatial structures interactively between the Euclidean, hyperbolic and hyperspherical spaces. HyIE performs two Householder transformations of head and tail entities parameterized by relations in a high-dimensional mixed vector space. The curvature of hyperbolic and hyperspherical spaces depends on the product of both relation and temporal. The core of HyIE lies in implementing transformations and interactions of vectors in Euclidean, hyperbolic and hyperspherical spaces, and Household transformation of head and tail entities. Theoretically, HyIE can model crucial relation patterns and mapping properties simultaneously. Experimental results on five temporal knowledge graph benchmarks show that our HyIE achieves state-of-the-art performance.

Recently, action recognition has achieved impressive performance, mainly due to the aid of deep convolutional neural networks and large datasets. Traditionally, most efforts in action recognition have focused on capturing motion information by dense optical flow, but optical flow extraction is very time-consuming. Moreover, prior arts seek to improve accuracy but neglect the part-whole relationship between objects in videos, which may be self-defeating and even deteriorate the performance of methods. To circumvent the above challenges, we present a novel collaborative multipath capsule network (CMCN) for action recognition. In particular, we propose a plug-and-play collaborative multipath block containing spatiotemporal, channel, and motion units, which are complementary and crucial information for action recognition. We exploit the interaction of these three units and selectively emphasize informative spatial-temporal motion to reduce the expensive computational costs. Subsequently, we explore a new capsule voting procedure to reduce the computation used in the capsule dynamic routing mechanism. The critical insight is that the same type of capsules simulates the same entity in different positions, and their voting results should be consistent. This strategy lessens the number of learning parameters that backward pass in the training process, and thus strengthens part-whole relationships in a video. Extensive experiments on multiple real-world datasets for action recognition demonstrate that our model significantly outperforms state-of-the-art models.

Weakly Supervised Semantic Segmentation with image-level annotation uses localization maps from the classifier to generate pseudo labels. However, such localization maps focus only on sparse salient object regions, it is difficult to generate high-quality segmentation labels, which deviates from the requirement of semantic segmentation. To address this issue, we propose a dual-aware domain mining and cross-aware supervision (DDMCAS) method for weakly-supervised semantic segmentation. Specifically, we propose a dual-aware domain mining (DDM) module consisting of graph-based global reasoning unit and salient-region extension controller, which produces dense localization maps by exploring object features in salient regions and adjacent non-salient regions simultaneously. In order to further bridge the gap between salient regions and adjacent non-salient regions to generate more refined localization maps, we propose a cross-aware supervision (CAS) strategy to recover missing parts of the target objects and enhance weak attention in adjacent non-salient regions, leading to pseudo labels of higher quality for training the segmentation network. Based on the generated pseudo-labels, extensive experiments on PASCAL VOC 2012 dataset demonstrate that our method outperforms state-of-the-art methods using image-level labels for weakly supervised semantic segmentation.

As the scope of receptive field and the depth of Graph Neural Networks (GNNs) are two completely orthogonal aspects for graph learning, existing GNNs often have shallow layers with truncated-receptive field and far from achieving satisfactory performance. In this article, we follow the idea of decoupling graph convolution into propagation and transformation processes, which generates representations over a sequence of increasingly larger neighborhoods. Though this manner can enlarge the receptive field, it has two critical problems unsolved: how to find the suitable receptive field to avoid under-smoothing or over-smoothing? and how to balance different diffusion operators for better capturing the local and global dependencies? We tackle these challenges and propose a S calable, A daptive G raph C onvolutional N etworks ( SAGCN ) with Transformer architecture. Concretely, we propose a novel non-heuristic metric method that quickly finds the suitable number of diffusing iterations and produces smoothed local embeddings that enable the truncated receptive field to become scalable and independent of prior experience. Furthermore, we devise smooth2seq and diffusion-based position schemes introduced into Transformer architecture for better capturing local and global information among embeddings. Experimental results show that SAGCN enjoys high accuracy, scalability and efficiency on various open benchmarks and is competitive with other state-of-the-art competitors.

Recently, with the development of temporal knowledge graph technology, more and more Temporal Knowledge Graph Embedded (TKGE) models have been developed. The effectiveness of TKGE largely depends on the ability to model intrinsic relation patterns and capture specific information about entities and relations. However, existing approaches can capture only some of them with insufficient modeling capacity, and none has a “deep” architecture for modeling the entries in a quadruple at the same dimension. In this article, we propose a more powerful KGE framework named DuCape , which combines a dual quaternion and capsule network in modeling for the first time to make up for the defects of existing TKGE models. In dual quaternion vector space, the head entity learns a k -dimensional rigid transformation parametrized by relation and time, falling near its corresponding tail entity. Further, we employ the embeddings of entities, relations, and time trained from dual quaternion vector space as the input to capsule networks. Experimental results on several basic datasets show that the DuCape model constructed in this article is superior to existing state-of-the-art models.

This paper studies the problem of learning complex relationships between multi-labels for image recognition. Its challenges come from the rich and diverse semantic information in images. However, current methods cannot fully explore the mutual interactions among labels and do not explicitly model the label co-occurrence. To overcome these shortcomings, we innovatively propose CGML that consists of two crucial modules: 1) an image representation learning module that aims to complete the feature extraction of an image whose features are expressed in the form of primary capsules; 2) a label adaptive graph convolutional network module that leverages the popular graph convolutional networks with an adaptive label correlation graph to model label dependencies. Experiments show that our approach obviously outperforms the existing state-of-the-art methods.

Graph neural networks (GNNs) have become a practical paradigm for learning graph-structured data, which can generate node representations by recursively aggregating information from neighbor nodes. Recent works utilize self-supervised tasks to learn transferable knowledge from source domain graphs and improve the GNNs performance on target domain graphs. However, there are considerable low-quality and incorrect-labeled graphs in the source domain, which leads to the negative transfer problem in target domain graphs. To tackle this challenge, we propose RSS-GNN, a reinforced sample selection for GNNs transfer learning. The critical insight is that RSS-GNN attempts to use reinforcement learning (RL) to guide transfer learning and narrow the graph divergence between the source and the target domain. We leverage a selection distribution generator (SDG) to produce the probability for each graph and select high-quality graphs to train GNNs. We innovatively designed a reward mechanism to measure the quality of the selection process and employ the policy gradient to update SDG parameters. Extensive experiments demonstrate that our approach can be compatible with various GNNs frameworks and yields superior performance compared to state-of-the-art methods.

The human recognition system has presented the remarkable ability to effortlessly learn novel knowledge from only a few trigger events based on prior knowledge, which is called insight learning. Mimicking such behavior on Knowledge Graph Reasoning (KGR) is an interesting and challenging research problem with many practical applications. Simultaneously, existing works, such as knowledge embedding and few-shot learning models, have been limited to conducting KGR in either “seen-to-seen” or “unseen-to-unseen” scenarios. To this end, we propose a neural insight learning framework named Eureka to bridge the “seen” to “unseen” gap. Eureka is empowered to learn the seen relations with sufficient training triples while providing the flexibility of learning unseen relations given only one trigger without sacrificing its performance on seen relations. Eureka meets our expectation of the model to acquire seen and unseen relations at no extra cost, and eliminate the need to retrain when encountering emerging unseen relations. Experimental results on two real-world datasets demonstrate that the proposed framework also outperforms various state-of-the-art baselines on datasets of both seen and unseen relations.

Existing graph contrastive methods have benefited from ingenious data augmantations and mutual information estimation operations that are carefully designated to augment graph views and maximize the agreement between representations produced at the aftermost layer of two view networks. However, the design of graph CL schemes is coarse-grained and difficult to capture the universal and intrinsic properties across intermediate layers. To address this problem, we propose a novel fine-grained graph contrastive learning model (FGCL), which decomposes graph CL into global-to-local levels and disentangles the two graph views into hierarchical graphs by pooling operation to capture both global and local dependencies across views and across layers. To prevent layers mismatch and automatically assign proper hierarchical representations of the augmented graph (Key view) for each pooling layer of the original graph (Query view), we propose a sematic-aware layer allocation strategy to integrate positive guidance from diverse representations rather than a fixed layer manually. Experimental results demonstrate the advantages of our model on graph classification task. This suggests that the proposed fine-grained graph CL presents great potential for graph representation learning.

Graph convolutional network (GCN) is a novel framework that utilizes a pre-defined Laplacian matrix to learn graph data effectively. With its powerful nonlinear fitting ability, GCN can produce high-quality node embedding. However, generalized GCN can only handle static graphs, whereas a large number of graphs are dynamic and evolve over time, which limits the application field of GCN. Facing the challenge, GCN with recurrent neural network (e.g., RNN) is naturally combined to acquire dynamic graph changes through joint training. However, these methods must use the node information during the entire timeline and ignore two subtle factors: the influence of nodes change with time and are related to the frequency of events. Therefore, we propose a stable and scalable dynamic GCN method using a fine-grained attention mechanism named FADGC. We use GCN to obtain static node vectors at each timestep and integrate node influence factors with multi-head attention for graph time-series learning. Experiments on multiple datasets show that our approach can better capture the inherent special characteristics of different dynamic graphs and achieve higher performance compared with related approaches.

From the perspective of the spatial domain, Graph Convolutional Network (GCN) is essentially a process of iteratively aggregating neighbor nodes. However, the existing GCNs using simple average or sum aggregation may neglect the characteristics of each node and the topology between nodes, resulting in a large amount of early-stage information lost during the graph convolution step. To tackle the above challenge, we innovatively propose an adaptive attention graph capsule network, named AA-GCN, for graph classification. We explore various propagation mechanisms of graphs and present an attention mechanism combined with graph propagation and capsules to generate capsule nodes, preserving the spatial topology between nodes. We also propose a graph adaptive attention mechanism to investigate the context information in different global GCN layers, so as to effectively improve the next dynamic routing connection and the final graph classification. Experiments show that our proposed algorithm achieves either state-of-the-art or competitive results across all the datasets.

Semantic segmentation aims to assign correct category label for every pixel in the image. Pixel-level classification and locality are thus both essential. To this end, previous methods try to fuse the multi-scale features of backbone network, including high-level category distinguishing cues and low-level location details. Other methods are designed to incorporate contextual cues or boundary guidance. However, previous methods are usually designed to refine the multi-scale feature fusion after the last fusion stage under the guidance of context or boundary, resulting in separate processing pipeline. This post-processing strategy is hard to remedy the context inconsistency and blurring boundary within fused features, which inevitably generates object segmentation with inconsistent context and blob-like contour. In this paper, we aim to bridge this gap by seamlessly introducing the context or boundary guidance into the multiple feature fusion operations. In this way, multi-scale features are effectively combined while maintaining context consistency and sharp object boundary, leading to enhanced semantic segment coherence. Experimental results on Cistyscapes and ADE20K datasets show the superiority of the proposed method.

Semantic segmentation is a dense pixel-level prediction task, which requires both strong classification and locality ability simultaneously. Previous methods tackle this challenging problem by enlarging receptive field, incorporating contextual cues, fusing multi-level features, etc. Recently, several pure Transformer methods push this forward by its intrinsic characteristics of global relation modeling between any pixels. The receptive field is enlarged to the whole image as while. However, current pure Transformer methods still has shortcomings, such as poor ability for pixel-level locality because of single-scale image patch splitting, the simple upsampling of bilinear interpolation and discrete positional encoding. To this end, we propose a pure multi-scale Transformer architecture with pixel-shuffle upsampling kernel and semantically continuous positional encoding. In addition, current Transformers naturally lack the ability to encode object shape due to the one-dimensional processing. Hence we propose to explicitly enforce the boundary constraints on the decoder. Experimental results demonstrate our method achieves state-of-art results on both Cityscapes and ADE20K datasets.

Sparsity and smoothness are two main factors that affect the performance of Graph Convolutional Networks (GCNs). Sparsity ensures that models have the first-class generalization ability, while smoothness benefits to reduce noise and make edges reliable. As real-world graphs are often incom-plete and noisy, most GCNs learn node embeddings only acting them as ground-truth information, which unavoidably lead to suboptimal solutions. This paper proposes a co-embedding network (CoNet), jointly learns embeddings by fusing the global and local dependencies to capture the uni-versal and intrinsic properties. We proposed NodeNet and EdgeNet modules, which aggregate global node information and refine local topology structure respectively. Moreover, we further introduce two piplines of variational auto-encoders to fuse the intermediate latent variables of each module to ob-tain co-embeddings via Knowledge Distillation strategy. Ex-tensive experiments on multiple benchmarks show that our proposed approach achieves better performance than existing methods on the graph node classification task.

Weakly supervised semantic segmentation with image-level labels is a challenging problem that typically relies on the initial responses generated by the classification network to locate object regions. However, such initial responses only cover the most discriminative parts of the object and may incorrectly activate in the background regions. To address this problem, we propose a Cross-pixel Dependency with Boundary-feature Transformation (CDBT) method for weakly supervised semantic segmentation. Specifically, we develop a boundary-feature transformation mechanism, to build strong connections among pixels belonging to the same object but weak connections among different objects. Moreover, we design a cross-pixel dependency module to enhance the initial responses, which exploits context appearance information and refines the prediction of current pixels by the relations of global channel pixels, thus generating pseudo labels of higher quality for training the semantic segmentation network. Extensive experiments on the PASCAL VOC 2012 segmentation benchmark demonstrate that our method outperforms state-of-the-art methods using image-level labels as weak supervision.

Recently, the link prediction task on Hyper-relational Knowledge Graphs (HKGs) has been a hot spot, which aims to predict new facts beyond binary relations. Although previous models have accomplished considerable achievements, there remain three challenges: i) the previous models neglect the existence of False Positive Entities (FPEs), which are true entities in the binary triples, yet becomes false when encountering the query statements of HKGs; ii) Due to the sparse interactions, the models are not capable of coping with long-tail hyper-relations, which are ubiquitous in the real-world; iii) The models are generally transductive learning processes, and have difficulty in adapting new hyper-relations. To tackle the above issues, we firstly propose the task of few-shot link prediction on HKGs and devise hyper-relation-aware attention networks with a contrastive loss, which are empowered to encode all entities including FPEs effectively and increase the distance between the true entities and FPEs through contrastive learning. With few-shot references available, the proposed model then learns the representations of their long-tail hyper-relations and predicts new links by calculating the likelihood between queries and references. Furthermore, our model is inductive and can be scalable to any new hyper-relation effortlessly. Since it is the first trial on few-shot link prediction for HKGs, we also modify the existing few-shot learning approaches on binary relational data to work with HKGs as baselines. Experimental results on three real-world datasets show the superiority of our model over various state-of-the-art baselines.

While Graph Convolutional Networks (GCNs) have been extended to various fields of artificial intelligence with their powerful representation capabilities, recent studies have revealed that their ability to capture the part-whole structure of the graph is limited. Furthermore, though many GCNs variants have been proposed and obtained state-of-the-art results, they face the situation that much early information may be lost during the graph convolution step. To this end, we innovatively present an Graph Capsule Network with a Dual Adaptive Mechanism (DA-GCN) to tackle the above challenges. Specifically, this powerful mechanism is a dual-adaptive mechanism to capture the part-whole structure of the graph. One is an adaptive node interaction module to explore the potential relationship between interactive nodes. The other is an adaptive attention-based graph dynamic routing to select appropriate graph capsules, so that only favorable graph capsules are gathered and redundant graph capsules are restrained for better capturing the whole structure between graphs. Experiments demonstrate that our proposed algorithm has achieved the most advanced or competitive results on all datasets.

We develop a RGB-D scene recognition model based on object-scene relation(RSBR). First learning a Semantic Network in the semantic domain that classifies the label of a scene on the basis of the labels of all object types. Then, we design an Appearance Network in the appearance domain that recognizes the scene according to local captions. We enforce the Semantic Network to guide the Appearance Network in the learning procedure. Based on the proposed RSBR model, we obtain the state-of-the-art results of RGB-D scene recognition on SUN RGB-D and NYUD2 datasets.

We proposed Image2Vec, an unsupervised algorithm that learns feature representations for variable-size pieces of Chinese character images.

Scene recognition is challenging due to intra-class diversity and inter-class similarity. Previous works recognize scenes either with global representations or with intermediate representations of objects. By contrast, we investigate more discriminative sequential representation of object-to-scene relations (SOSRs) for scene recognition. Particularly, we develop an Attention-Preserving Memory-Learning (APML) model, which enforces the Memory Network of the semantic domain to guide the Learning Network of the appearance domain in the learning procedure. Accordingly, we allocate semantics-preserving attention to different objects, which is more effective to seek the key encoded SOSR and discard the misleading encoded SOSR between objects and scene without requiring extra labeled data. Based on the proposed APML networks, we obtain the state-of-the-art results of RGB-D scene recognition on SUN RGB-D and NYUD2 datasets.

A least squares support vector machine (LS-SVM) offers performance comparable to that of SVMs for classification and regression. The main limitation of LS-SVM is that it lacks sparsity compared with SVMs, making LS-SVM unsuitable for handling large-scale data due to computation and memory costs. To obtain sparse LS-SVM, several pruning methods based on an iterative strategy were recently proposed but did not consider the quantity constraint on the number of reserved support vectors, as widely used in real-life applications. In this article, a noniterative algorithm is proposed based on the selection of globally representative points (global-representation-based sparse least squares support vector machine, GRS-LSSVM) to improve the performance of sparse LS-SVM. For the first time, we present a model of sparse LS-SVM with a quantity constraint. In solving the optimal solution of the model, we find that using globally representative points to construct the reserved support vector set produces a better solution than other methods. We design an indicator based on point density and point dispersion to evaluate the global representation of points in feature space. Using the indicator, the top globally representative points are selected in one step from all points to construct the reserved support vector set of sparse LS-SVM. After obtaining the set, the decision hyperplane of sparse LS-SVM is directly computed using an algebraic formula. This algorithm only consumes O(N2) in computational complexity and O(N) in memory cost which makes it suitable for large-scale data sets. The experimental results show that the proposed algorithm has higher sparsity, greater stability, and lower computational complexity than the traditional iterative algorithms.

In complex network, the links among the nodes fabricate thousands of polygons, where triangles are the most foundamental ones. As more links are added to the network, more polygons are split to be triangles. Discovering the triangles can help understand the current status of the network. Due to the limitations of internal memory, discovering all the triangles in a large-scale real-world network, e.g., online social network (OSN), is challenging and has intrigued researchers for years. Currently, parallel computation frameworks, e.g., MapReduce, have been employed to address this issue. The last reducer is required to load all the neighbors of each node into internal memory, which causes “the curse of the last reducer.” The scale-free characteristic in real-world networks further exacerbate the applicability of the parallel triangle discovery algorithms. The current efforts on this issue mainly focus on dividing a large network into several small subnetworks and later employ internal memory algorithms. In this paper, we address this issue from a different perspective by employing the power-law degree distribution in scale-free networks. We theoretically and empirically prove that one node in scale-free network has a limited number of neighbors with a higher degree. Employing this finding, an novel parallel triangle discovery algorithm called degree partition (DePart) is designed. Since only the portion of neighbors with higher degrees are loaded into internal memory in the reducers, DePart solves “the curse of the last reducer” problem and works efficiently in clusters with limited internal memory. We develop DePart using MapReduce framework and examine the performance of DePart on five large real-world networks. The empirical testing results demonstrate that DePart has a favorable tradeoff between internal memory and total disk space and outperforms the current state-of-the-art MapReduce-based triangle discovery algorithms.

In the research of social networks, their structural characteristics, user social behaviors, and user relationships are elements that are important to understand social networks, to predict user behaviors, and to manage social networks. In this article, we took as the research object of social networks the mobile communication network, which is closely connected with people's real lives. We studied the structural characteristics using the complex network analysis methods and derived the laws that exist in the structure of communication social networks. We analyzed users' social behaviors or social patterns from the perspective of age, gender, social scope, age differences, and time. In addition, we extracted various salient features of user's calling behavior, and used the XGBoost and logistic regression (LR) fusion method to establish the Kinship-XL model, which is able to improve the performance and speed. Through the experimental verification, the Kinship-XL model can determine whether there is a kinship between users or not.

Convex edges are located on convex surfaces of hidden geometry spaces, resulting in shorter geometrical distance than observed. Convex edges reveal the close relationships between two connected nodes and have a key role in many research fields and applications, including social, trust, and protein-protein interaction networks. We first quantitatively define convex edges using hidden geometry space defined by Ricci curvature and show these to be capable of predicting highly interacted edges in social networks. From empirical studies of social networks, we find that convex edges often occur on two connected nodes with similar popularity and allow for frenemy (or versatile node) identification, and that almost all nodes maintain an upper bound of convex edges. These findings present further evidence of the limited attention in an attention economy from a different perspective. Using three well-known synthetic networks, we show that convex edges can reflect the formation of a network from scatter (concave) to aggregation (convex). Taken together, our results show that convex edges are a novel mechanism to investigate social networks from different disciplines and offer insights into many network phenomena.

Given the edge list of a social network, the node embedding method learns the structural features for every node and embeds the features into a vector space. The current related work on node embedding exploits only a portion of existing networks, e.g., static networks. However, social networks are inherently hierarchical and dynamic systems in which the topology changes constantly and the strength of influence of information among neighbors varies with different numbers of hops. We propose a highly efficient node embedding method, DNPS, that is faster and more accurate than state-of-the-art methods and that can further boost the training progress, especially under dynamic conditions. In this paper, we attempt to model the hierarchical and dynamic features of social networks by designing a damping-based sampling algorithm corresponding to a local search-based incremental learning algorithm, which can easily be extended to large-scale scenarios. We conduct extensive experiments on six real-world social networks with three challenging tasks, including missing link prediction, dynamic link prediction, and multi-label classification. The results of the experiments on these tasks demonstrate that the proposed method significantly outperforms the existing methods with different settings.

This paper studies the problem of learning node embeddings (a.k.a. distributed representations) for dynamic networks. The embedding methods allocate each node in network with a d-dimensions vector, which can generalize across various tasks, such as item recommendation, node labeling, and link prediction. In practice, many real-world networks are evolving with nodes/links added or deleted. However, most of the proposed methods are focusing on static networks. Although some previous researches have shown some promising results to handle the dynamic scenario, they just considered the added links and ignored the deleted ones. In this work, we designed a joint learning of added and deleted links model, named RDEM, for dynamic network embedding.

Social network user identification is of great value and significance for predicting users' behavior, analyzing and regulating social networks, and studying the interaction between users. This paper studies the identities of social network users, divides them into organizational and individual users in terms of their identities, and explicitly defines and identifies both types of users. This paper also distinguishes a user as an organization or an individual according to the contents of text, multimedia, and their time series published in a social network. During the identification, the content (topic) complexity and normalization of the user in text content are measured; the picture features and time-series content of the user are being analyzed, and the machine-operable method that identifies a user as an organization or an individual is proposed from different perspectives so as to perform this identification. Finally, in order to verify the feasibility and effectiveness of the identification method proposed in this paper, an experiment was made to the data collected from Sina Weibo and the probability model identification method was used to make a comparative analysis. Results indicate that the identification method used in this paper can effectively distinguish between a user as an organization or an individual.

Identification of anonymous identical users of cross-platforms refers to the recognition of the accounts belonging to the same individual among multiple Social Network (SN) platforms. Evidently, cross-platform exploration may help solve many problems in social computing, in both theory and practice. However, it is still an intractable problem due to the fragmentation, inconsistency, and disruption of the accessible information among SNs. Different from the efforts implemented on user profiles and users' content, many studies have noticed the accessibility and reliability of network structure in most of the SNs for addressing this issue. Although substantial achievements have been made, most of the current network structure-based solutions, requiring prior knowledge of some given identified users, are supervised or semi-supervised. It is laborious to label the prior knowledge manually in some scenarios where prior knowledge is hard to obtain. Noticing that friend relationships are reliable and consistent in different SNs, we proposed an unsupervised scheme, termed Friend Relationship-based User Identification algorithm without Prior knowledge (FRUI-P). The FRUI-P first extracts the friend feature of each user in an SN into friend feature vector, and then calculates the similarities of all the candidate identical users between two SNs. Finally, a one-to-one map scheme is developed to identify the users based on the similarities. Moreover, FRUI-P is proved to be efficient theoretically. Results of extensive experiments demonstrated that FRUI-P performs much better than current state-of-art network structure-based algorithm without prior knowledge. Due to its high precision, FRUI-P can additionally be utilized to generate prior knowledge for supervised and semi-supervised schemes. In applications, the unsupervised anonymous identical user identification method accommodates more scenarios where the seed users are unobtainable.

The semisupervised least squares support vector machine (LS-S3VM) is an important enhancement of least squares support vector machines in semisupervised learning. Given that most data collected from the real world are without labels, semisupervised approaches are more applicable than standard supervised approaches. Although a few training methods for LS-S3VM exist, the problem of deriving the optimal decision hyperplane efficiently and effectually has not been solved. In this paper, a fully weighted model of LS-S3VM is proposed, and a simple integer programming (IP) model is introduced through an equivalent transformation to solve the model. Based on the distances between the unlabeled data and the decision hyperplane, a new indicator is designed to represent the possibility that the label of an unlabeled datum should be reversed in each iteration during training. Using the indicator, we construct an extended candidate set consisting of the indices of unlabeled data with high possibilities, which integrates more information from unlabeled data. Our algorithm is degenerated into a special scenario of the previous algorithm when the extended candidate set is reduced into a set with only one element. Two strategies are utilized to determine the descent directions based on the extended candidate set. Furthermore, we developed a novel method for locating a good starting point based on the properties of the equivalent IP model. Combined with the extended candidate set and the carefully computed starting point, a fast algorithm to solve LS-S3VM quasi-optimally is proposed. The choice of quasi-optimal solutions results in low computational cost and avoidance of overfitting. Experiments show that our algorithm equipped with the two designed strategies is more effective than other algorithms in at least one of the following three aspects: 1) computational complexity; 2) generalization ability; and 3) flexibility. However, our algorithm and other algorithms have similar levels of performance in the remaining aspects.

Learning distributed representations of symbolic data were introduced by Hinton[1], and first developed in modeling networks for learning the node vectors by Perozzi et al (2014). In this work, we proposed Dnps, a novel nodes embedding approach for acquiring distributed representations of large-scale dynamic social networks. Dnps is suitable for many types of social networks: dynamic/static, directed/undirected, and weighted/unweighted. Recently, several works of nodes embedding were proposed. However, they were designed for static networks, such as language networks. To address this problem, first, we develop a damping based positive sampling (DpS) algorithm to learn the hierarchical structure of social networks. Then, we devise a local search based DpS algorithm to obtain incremental information of network evolution. Finally, we show Dnps's potentials on future link prediction task for three real-life large-scale dynamic social networks. The results show that Dnps consistently outperforms all baseline methods and exhibits an improvement of 12%, 6%, 4% on Digg, Flickr and YouTube over the second-highest level, respectively. Moreover, Dnps is also scalable. For example, Dnps can speed up the training process in 2 ~ 36 times compared with benchmarks on Flickr network. The source codes of the project is available online1.

We present an algorithm (LsNet2Vec) that, given a large-scale network (millions of nodes), embeds the structural features of node into a lower and fixed dimensions of vector in the set of real numbers. We experiment and evaluate our proposed approach with twelve datasets collected from SNAP. Results show that our model performs comparably with state-of-the-art methods, such as Katz method and Random Walk Restart method, in various experiment settings.

The last few years have witnessed the emergence and evolution of a vibrant research stream on a large variety of online social media network (SMN) platforms. Recognizing anonymous, yet identical users among multiple SMNs is still an intractable problem. Clearly, cross-platform exploration may help solve many problems in social computing in both theory and applications. Since public profiles can be duplicated and easily impersonated by users with different purposes, most current user identification resolutions, which mainly focus on text mining of users’ public profiles, are fragile. Some studies have attempted to match users based on the location and timing of user content as well as writing style. However, the locations are sparse in the majority of SMNs, and writing style is difficult to discern from the short sentences of leading SMNs such as Sina Microblog and Twitter. Moreover, since online SMNs are quite symmetric, existing user identification schemes based on network structure are not effective. The real-world friend cycle is highly individual and virtually no two users share a congruent friend cycle. Therefore, it is more accurate to use a friendship structure to analyze cross-platform SMNs. Since identical users tend to set up partial similar friendship structures in different SMNs, we proposed the Friend Relationship-Based User Identification (FRUI) algorithm. FRUI calculates a match degree for all candidate User Matched Pairs (UMPs), and only UMPs with top ranks are considered as identical users. We also developed two propositions to improve the efficiency of the algorithm. Results of extensive experiments demonstrate that FRUI performs much better than current network structure-based algorithms.

Mining global patterns on big data distributed in many remote locations is a challenging task since transmitting big data in different remote data servers to the central server is prohibitively expensive. In this paper, we present a succinct distributive big data clustering algorithm based on local-remote coordination (DBDC-LRC) that aims to reduce the cost of big data transmission while maintaining an acceptable overall clustering accuracy. The algorithm is divided into three phases. In the first phase, the idea of Canopy algorithm is improved in the search for representative points with a clustering assumption that the decision boundary should lie in a low-density region, during which controllable thresholds are optimized. Noticing that in data mining a hyperellipsoid is more adaptable in shaping unknown data than a hypercube, we employ Mahalanobis distance as opposed to the Euclidean distance in determining the representative points in different remote data servers. Given that only a limited number of representative points, instead of all the remote data, are transmitted to the central server for clustering, the transmitting cost is reduced significantly. In the second phase, a weighted clustering method is used in mining the global patterns from the gathered information of representative points from various remote data servers. In the third phase, the mined global patterns are sent back to the original remote server and the related data are labeled with the same patterns according their representative points nearby. In this phase, Bayesian method is used to resolve the conflicts that one point is covered by several representative points in its neighborhood. Experiments show that DBDC-LRC is highly suitable for mining patterns from distributive big data, and the advantages of this approach include low cost, high accuracy, high robustness, and good expansibility.

In bus hijacking, the availability of instant information in the scene may help the decision-making largely. In this paper, we discussed the significant value of the information acquisition in bus hijacking emergency from a qualitative analysis and quantitative description. Furthermore, we proposed an effective emergency decision support architecture for bus hijacking based on massive information in social networks. Last but not least, as to the core part of images discrimination, we build an image anomaly detection algorithm model. In the first step of the model, we conduct a Scale Invariant Feature Transform (SIFT) detection for images, and extract local feature descriptor; In the second step, the image feature vectors of the key points are subjected to further K-means clustering, so that we get the unified K-dimensional feature vectors; In the third step, we make the image classification with Support Vector Machine (SVM) classifier. This algorithm model achieves the image discrimination for bus hijacking emergency successfully, so that the information inside the bus could be transmitted to the outside effectively, and therefore provide a significant value for emergency decision-making.

Community detection is one of the most popular issues in analyzing and understanding the networks. Existing works show that community detection can be enhanced by proper assignments of weights onto the edges of a network. Large numbers of edge weighting schemes have been developed to cope with this problem. However, hardly has a satisfied balance between the local and global weightings been found. In this paper, the problem of the local and global weighting balance is first proposed and discussed. The SimRank is next introduced as a novel edge weighting method. Furthermore, the fast Newman algorithm is extended to be applicable for a weighted network. Combined with the edge weighting techniques, the extended algorithm enhances the performance of the original algorithm significantly through exhaustive experiments. And by comparing with several weighting methods, the experiments demonstrate that the proposed algorithm is superior and more robust for different kinds of networks.

In this paper, we present Cdear, a novel system to detect fake reviews by using sentiment analysis on attributes of products. We formulate review spam detection as an opinion coincided problem. Specifically, we try to capture the sentiment diverse of attributes of products among different consumers. To our knowledge, this is the very first attempt to use sentiment analysis on attributes of products in reviews to detect review spam. To evaluate the effectiveness of our system, we conduct the experiments on the real life datasets and employ 20 experts to assess the reliability of reviews by carrying out a simulation of shopping online. Meanwhile, we developed an evaluation system to help those experts to assess the reviews, thus to ensure the scientificity of experiments. Comparison between the automatic results of proposed system and human evaluated results demonstrates that the sentiment-based method matches approximately with human perception of reviews' reliability with 82.66% accuracy.

Standard support vector machines (SVMs) have kernels based on the Euclidean distance. This brief extends standard SVMs to SVMs with kernels based on the Mahalanobis distance. The extended SVMs become a special case of the Euclidean distance when the covariance matrix in a reproducing kernel Hilbert space is degenerated to an identity. The Mahalanobis distance leads to hyperellipsoidal kernels and the Euclidean distance results in hyperspherical ones. In this brief, the Mahalanobis distance-based kernel in a reproducing kernel Hilbert space is developed systematically. Extensive experiments demonstrate that the hyperellipsoidal kernels slightly outperform the hyperspherical ones, with fewer SVs.

Support vector machine (SVM) classifiers often contain many SVs, which lead to high computational cost at runtime and potential overfitting. In this paper, a practical and effective method of pruning SVM classifiers is systematically developed. The kernel row vectors, with one-to-one correspondence to the SVs, are first organized into clusters. The pruning work is divided into two phases. In the first phase, orthogonal projections (OPs) are performed to find kernel row vectors that can be approximated by the others. In the second phase, the previously found vectors are removed, and crosswise propagations, which simply utilize the coefficients of OPs, are implemented within each cluster. The method circumvents the problem of explicitly discerning SVs in the high-dimensional feature space as the SVM formulation does, and does not involve local minima. With different parameters, 3000 experiments were run on the LibSVM software platform. After pruning 42% of the SVs, the average change in classification accuracy was only - 0.7%, and the average computation time for removing one SV was 0.006 of the training time. In some scenarios, over 90% of the SVs were pruned with less than 0.1% reduction in classification accuracy. The experiments demonstrate the existence of large numbers of superabundant SVs in trained SVMs, and suggest a synergistic use of training and pruning in practice. Many SVMs already used in applications could be upgraded by pruning nearly half of their SVs.

Support vector machines (SV machines, SVMs) have many merits that distinguish themselves from many other machine-learning algorithms, such as the nonexistence of local minima, the possession of the largest distance from the separating hyperplane to the SVs, and a solid theoretical foundation. However, SVM training algorithms such as the efficient sequential minimal optimization (SMO) often produce many SVs. Some scholars have found that the kernel outputs are frequently of similar levels, which insinuate the redundancy of SVs. By analyzing the overlapped information of kernel outputs, a succinct separating-hyperplane-securing method for pruning the dispensable SVs based on crosswise propagation (CP) is systematically developed. The method also circumvents the problem of explicitly discerning SVs in feature space as the SVM formulation does. Experiments with the famous SMO-based software LibSVM reveal that all typical kernels with different parameters on the data sets contribute the dispensable SVs. Some 1% ~ 9% (in some scenarios, more than 50%) dispensable SVs are found. Furthermore, the experimental results also verify that the pruning method does not alter the SVMs' performances at all. As a corollary, this paper further contributes in theory a new lower upper bound on the number of SVs in the high-dimensional feature space.

In this paper, we analyze the China venture capital guiding fund policy based on options model. Options theory determines the present value of a future uncertainty interest. Following this principle, we propose the methods to valuate the policy with both Monte Carlo simulation and numerical analysis technique on Black-Scholes method. Our work is a remarkable step towards the quantitative analysis of public policies using options theory.

Error hypersurfaces are very valuable to study because of their unique status in multilayer perceptron research. Given the architecture of a multilayer perceptron, if the pattern sets are different, so are the respective error hypersurfaces in the multilayer perceptron. Using the theory of groups and Polya Theorem, this paper constructs classes of congruent pattern sets and classes of congruent error hypersurfaces, and proves that the number of classes of congruent pattern sets is equal to the number of congruent error hypersurfaces. Calculation results lead to much fewer classes of congruent error hypersurfaces than the total error hypersurfaces, and show that as the input dimension N increases, the former number increases at a much lower rate than the latter number, thus simplifying the understanding of the complexity of classes of error hypersurfaces.

This paper presents two compensation methods for multilayer perceptrons (MLPs) which are very difficult to train by traditional Back Propagation (BP) methods. For MLPs trapped in local minima, compensating methods can correct the wrong outputs one by one using constructing techniques until all outputs are right, so that the MLPs can skip from the local minima to the global minima. A hidden neuron is added as compensation for a binary input three-layer perceptron trapped in a local minimum; and one or two hidden neurons are added as compensation for a real input three-layer perception. For a perceptron of more than three layers, the second hidden layer from behind will be temporarily treated as the input layer during compensation, hence the above methods can also be used. Examples are given.