Anti-money laundering (AML) detection is of vital importance in financial risk control. Although Graph Neural Networks (GNN) have yielded promising results, existing motif-based approaches primarily focus on node anomaly detection on simple graphs, which hinders the direct identification of anomalous edges in directed temporal transaction networks. Moreover, consecutive transaction relationships, termed dual-edge motifs, have rarely been considered in previous AML studies. To address these gaps, we propose the D-EMAML framework, which consists of: (1) Fast-Motif-Gen, a GPU-accelerated dual-edge motif graph generator with pruning; (2) D-EMGNN, an attention-enhanced heterogeneous GNN module that reduces motif-type information redundancy; (3) MELP, a label aggregation scheme projecting predictions from the motif graph to the original graph. Extensive experiments on real-world and synthetic datasets demonstrate significant improvements over representative baselines and validate the contribution of each component. To our knowledge, this is the first application of dual-edge motif graphs for GNN-based edge anomaly detection in AML.

The behavior of spreaders on social media evolves continuously, driven by shifting intentions and interactions with emerging news topics. Traditional approaches have focused on identifying misinformation spreaders, but have often relied on a static ground-truth label, limiting their applicability for implementing time-sensitive platform interventions. In contrast, our work tackles spreader behavior forecasting through an account-level credit score, modeling the temporal evolution of spreader behavior to capture the intent shifts that drive misinformation spreading. To this end, we propose a novel Intent-aware Neural Processes (INP) model, which focuses on tracking the evolving intent of spreaders over time. The model leverages a state transition structure and an intent state thinning algorithm to improve latent representations, enabling more accurate predictions of future spreader behavior. Experimental results on restructured datasets demonstrate the effectiveness of INP in identifying temporal risk regions for proactive misinformation intervention.

User credibility on social media is inherently dynamic, influenced by user interactions and engagement with evolving news topics. Existing approaches to credibility assessment often rely on static scores, limiting their applicability for implementing time-sensitive platform interventions. To address this, we redefine the problem by introducing time-grained credibility labeling for users based on their temporal engagement behaviors. Methodologically, we propose a novel State-Aware Neural Processes (SANP) model to capture user state transitions and evolving credibility over time. The model enhances Neural Processes with a dynamic state transition structure and a state-thinning algorithm for improved latent representation. Experimental results on restructured datasets validate the effectiveness of the SANP model in predicting user credibility dynamics and enabling more proactive misinformation mitigation.

Given the unwieldy glut of information in online health question-answer (Q&A) service, it is essential to understand what constitutes helpful answers in the medical domain. Despite the fact that studies have examined the impacts of answer content factors on answer helpfulness, there are two gaps that need further analysis. First, the empirical results of the existing relevant studies on the effect of answer emotion are inconsistent. Second, prior studies only have examined the independent impacts of answer content factors and question content cues on answer helpfulness. To fill this gap, a research model reflecting the impacts of emotional content and question-answer congruence on answer helpfulness was developed and empirically examined. Our empirical analyses confirm that emotional content and answer helpfulness are related to one another in the form of an inverted U-shape and indicate that two types of question-answer congruence (emotional intensity congruence and linguistic style matching) positively affect answer helpfulness. Theoretical and practical implications are discussed.

Fact-checking websites try to combat rumors but can't always keep up with breaking news. Existing spreader detection methods rely on fact-checking websites to label the user's historical activities. However, these methods overlook the context of news topics and fail to predict user activities with unchecked breaking news topics, called problematic news topics in previous literature. We propose a problematic news topics spreader prediction task to bridge this gap. Methodologically, we model the prediction task with Marked Temporal Point Processes (MTPP) based on the deep learning method. Moreover, we introduce a novel hierarchical contrastive learning approach incorporating uncertainty estimation to mitigate the impact of noise prevalent in real-world social media data. This integration facilitates improved data augmentation and sequence-level sampling methods. Our approach, dubbed Uncertainty-based Contrastive Learning for MTPP (UCL-MTPP), offers both high fidelity and variety in evaluation.

Non-coding variants associated with complex traits can alter the motifs of transcription factor (TF)-deoxyribonucleic acid binding. Although many computational models have been developed to predict the effects of non-coding variants on TF binding, their predictive power lacks systematic evaluation. Here we have evaluated 14 different models built on position weight matrices (PWMs), support vector machines, ordinary least squares and deep neural networks (DNNs), using large-scale in vitro (i.e. SNP-SELEX) and in vivo (i.e. allele-specific binding, ASB) TF binding data. Our results show that the accuracy of each model in predicting SNP effects in vitro significantly exceeds that achieved in vivo. For in vitro variant impact prediction, kmer/gkm-based machine learning methods (deltaSVM_HT-SELEX, QBiC-Pred) trained on in vitro datasets exhibit the best performance. For in vivo ASB variant prediction, DNN-based multitask models (DeepSEA, Sei, Enformer) trained on the ChIP-seq dataset exhibit relatively superior performance. Among the PWM-based methods, tRap demonstrates better performance in both in vitro and in vivo evaluations. In addition, we find that TF classes such as basic leucine zipper factors could be predicted more accurately, whereas those such as C2H2 zinc finger factors are predicted less accurately, aligning with the evolutionary conservation of these TF classes. We also underscore the significance of non-sequence factors such as cis-regulatory element type, TF expression, interactions and post-translational modifications in influencing the in vivo predictive performance of TFs. Our research provides valuable insights into selecting prioritization methods for non-coding variants and further optimizing such models.

Consumer product recommendation articles posted in Social Shopping Community (SSC) have become an important source of purchase information for other potential consumers. However, less effort has been put into understanding and predicting the popularity of such a distinctive form of consumer-generated content. In this study, we built a rich and comprehensive dataset comprising author-related features, article-related features, and engagement behavior information collected from post.smzdm.com. We constructed machine learning models to predict article popularity, and used the SHapley Additive exPlanations (SHAP) approach to visualize and explain feature importance in the prediction. The results show that with all identified features, LGBMClassifier gives best results with most evaluation metrics, and that the author-related feature set has better predictive capability than the article-related one. To the best of our knowledge, ours is the first study to investigate the popularity of consumers’ product recommendation articles in SSCs.

In this paper, we investigated the customer churn prediction problem in the Internet funds industry. We designed a novel feature embedded convolutional neural networks (FE-CNN) method that can automatically learn features from both the dynamic customer behavioral data and static customer demographic data and can utilize the advantage of convolutional neural networks to automatically learn features that capture the structured information. Our results show that our FE-CNN model outperforms the other traditional machine learning models with hand-crafted features, such as logistic regression (LR), support vector machines (SVM), random forests (RF) and neural networks (NN) in terms of accuracy, area under the receiver operating characteristics curve (AUC) and top-decile lift. Furthermore, we found that after adding the demographic data feature to the basic CNN model, the performance of the FE-CNN model improved. Overall, we found that the FE-CNN is the most powerful way to solve the problem of customer churn prediction in the Internet funds industry. Our FE-CNN method can also be applied to other fields that have both dynamic data and static data.

Boom Index can reveal the periodicity fluctuation trends and degree of the real estate market, so as to achieve the purpose of early warning. This article begins with the construction of indicators system of real estate from two aspects: external factors and internal factors. Then this article screened the 15 indicators through time-difference correlation analysis, and then checked that if there were significant difference between the same category indicators through the use of Wald-wolfowitz Runs Test. Finally, this article used state-space model to compose SWI boom index, and its movement is in line with the actual cycle of real estate market. At the same time it was compared with CI, the two indexes have a higher correlation. This illustrates the use of state-space model is effective.