Knowledge Graphs (KGs) have emerged as powerful tools for organizing and representing structured knowledge, enabling advanced reasoning and intelligent applications across diverse domains. Traditional approaches to Knowledge Graph Construction (KGC) rely on rule-based systems or supervised learning methods that require extensive feature engineering and labeled data, often limiting their generalization capabilities. Recently, large language models (LLMs) have shown great promise in improving KGC tasks by offering contextualized representations that enhance entity and relation extraction from unstructured text. However, one major limitation of LLMs is their tendency to generate hallucinated or factually incorrect information, which undermines the reliability of the resulting knowledge graphs. To address this challenge, we propose RLRSKGC (Reinforcement Learning with Reward Shaping for Knowledge Graph Construction), a novel framework that integrates reinforcement learning (RL) with Transformer-based language models to improve factual consistency and reduce hallucinations in the knowledge extraction process. Our approach formulates KGC as a sequential decision-making problem and introduces reward shaping mechanisms that explicitly evaluate the accuracy, completeness, and structural coherence of the generated knowledge. We conduct comprehensive experiments on benchmark KGC datasets to validate the effectiveness of our framework, demonstrating that RLRS-KGC achieves superior performance in extracting high-quality, graph-structured knowledge from textual sources.

Telecom fraud in mobile communication networks has become a serious threat to user security and network integrity. Traditional graph neural networks (GNNs) struggle to effectively detect fraudulent activities due to the pervasive noise in real-world fraud data, where genuine fraud signals are often obscured by spurious interactions and feature corruption. To address this challenge, we propose a novel framework combining a Top-p Neighbor Sampler and an adaptive graph neural network module, which selectively aggregates reliable neighbor features while suppressing noise propagation. Experiments on a real-world telecom fraud dataset demonstrate that our model outperforms state-of-the-art methods in macro-F1, AUC, and recall for fraud detection. This work not only provides a practical solution for telecom fraud detection but also offers insights into handling noise contamination in graph-structured data.

Poor coverage quality is a common cause of poor wireless communication network quality, which seriously affects the user experience in mobile communication. Currently, the front line mainly adopts a manual trial-and-error method, which has problems such as low efficiency and high human cost. How to use artificial intelligence algorithms to quickly and accurately identify and solve the problem of poor coverage quality based on existing data is one of the important research directions in the field of wireless networks. The data of wireless networks is essentially spatio-temporal data, but most of the existing methods are based on time-domain and space-domain data for analysis and modeling, and the information mining in the spatio domain is not sufficient. In the spatio domain, the distribution of base stations is not uniform in Euclidean space, which increases the difficulty of spatio-temporal modeling. In view of the natural advantages of graph mining technology for modeling and processing unstructured data, this paper proposes a model named Spatio-Temporal Graph Message Propagation (STGMP) based on graph technology. This method uses spatio-temporal graphs to represent the historical states of related service cells, proposes a processing layer that combines the time and spatio domains, and maps the actual problem to a multi-classification task, thereby achieving the identification of the causes of poor coverage quality. This paper also conducts experiments on real data sets, and the results show that the proposed method STGMP is very effective.

The fragmented architecture, cloud-based infrastructure, and functionally virtualized network elements within the 5 G core network have significantly surged the volume and diversity of alarms generated on cloud network service platforms that it supports. Given the inherently cross-layered nature of failure scenarios on these platforms, identifying the root causes presents a significant challenge. Alarm association rule mining has become an effective means to address the problems of alarm correlation and root cause localization. In this paper, an explainable alarm association rule mining approach based on knowledge graph, referred to as ARK-G, is proposed. Initially, a cloud-network cross-layer alarm association knowledge graph (CA2KG) is constructed. Subsequently, the knowledge embedding based graph convolutional network is employed to perform knowledge graph embedding on CA2KG. This embedding is then utilized to enhance the RNNLogic algorithm, thereby facilitating cross-layer alarm association rule mining with interpretable paths. Finally, a weighted rule tree is derived from a subset of CA2KG and the generated explainable rules, enabling the deduction of the root alarm. Experimental results demonstrate that the proposed ARK-G approach for association rule mining yields a higher hit rate compared to the baseline model, which provides valuable assistance in the faults analysis of 5 G cloud-network platforms.

Click-Through Rate (CTR) prediction is a crucial task in online advertising and recommender systems. To achieve better prediction results, it is important to accurately model feature interactions. Through the embedding layer, the relationships and interactions between the original features are reflected between the embedding vectors. Additionally, for specific features, the different dimensions within the embedding vectors also represent different perspectives and angles, which should be assigned varying importance weights. However, existing research on CTR prediction tasks using convolution operations faces significant challenges. Firstly, the comprehensive exploration of attention mechanisms within and between embedding vectors is often overlooked, resulting in a limited understanding of feature importance. Additionally, the neglect of global feature interactions hinders the practical application of CTR prediction. In this paper, we propose Convolutional Attention Over Attention (ConvAOA), a novel model that fully and efficiently leverages convolution operations and attention mechanisms for CTR prediction. ConvAOA sequentially processes features using convolutional intra-attention and convolutional inter-attention mechanisms, enabling dynamic attention to feature importance and preservation of meaningful information within and across different features. Extensive experiments are conducted on four real-world datasets, demonstrating the significant performance of ConvAOA in CTR prediction scenarios. The results show that ConvAOA outperforms current mainstream models and is an efficient and advanced solution in the field of CTR prediction.

With AI and big data technologies, telecom operators are looking to change the traditional O&M model from reactive problem handling to proactive prevention and prediction. This paper proposes a model framework trained on multiple data sources for the 5G wireless network to support proactive O&M tasks based on complaint prediction. By grouping user complaints into base station complaint prediction, the model enhanced precision scores while maintaining high recall scores. The model has been integrated into the operator’s work order system to support intelligent operational optimization workflow.

Visual Question Answering (VQA) is a recent hot topic that involves multimedia analysis, computer vision (CV), natural language processing (NLP), and even a broad perspective of artificial intelligence, which is challenging and has obtained increasing attention. VQA needs a complete understanding of the spatial relationship, textual clues, as well as the common sense for an actual image. However, most existing approaches simply embed and concatenate the features of questions and images to predict answers. Treating all embeddings equally without consideration of relation consistency hinders the model performance. In this paper, we propose an explicit Multi-Granularity Cross-Attention network (MGCAN) that mutually learns the multi-modal branches. MGCAN jointly matches word-level representation with whole image, and patch-level representation with the whole question that infers the high-order vision-semantic relationship. Experiments conducted on VQA datasets demonstrate that the proposed MGCAN outperforms previous baselines. The cross-attention mechanism explicitly exploits the relevant visual and textual clues that lead to superior prediction.

Recent years have witnessed the emerging success of Graph Neural Networks (GNNs) for modeling graphical data. A GNN can model the spatial dependencies of nodes in a graph based on message passing through node aggregation. However, in many application scenarios, these spatial dependencies can change over time, and a basic GNN model cannot capture these changes. In this article, we propose a G raph S eq uence neural network with an A tt ention mechanism (GSeqAtt) for processing graph sequences. More specifically, two attention mechanisms are combined: a horizontal mechanism and a vertical mechanism. GTransformer, which is a horizontal attention mechanism for handling time series, is used to capture the correlations between graphs in the input time sequence. The vertical attention mechanism, a Graph Network (GN) block structure with an attention mechanism (GNAtt), acts within the graph structure in each frame of the time series. Experiments show that our proposed model is able to handle information propagation for graph sequences accurately and efficiently. Moreover, results on real-world data from three road intersections show that our GSeqAtt outperforms state-of-the-art baselines on the traffic speed prediction task.

Summary Being part of public transportations, the bike sharing system plays an important role in solving the “last mile” problem. With more advantages than traditional bike sharing systems relying on fixed docks, dockless bike sharing systems are rapidly expanding, which generally operated under cloud‐based architectures. Emerging Things‐Edge‐Cloud (TEC) architectures are showing many advantages than cloud‐based architectures including greater potential in reducing risks of privacy leakage due to their decentralized computing manners. However, even in the TEC architecture, differential attacks, which aims to mine information of individuals in the aggregated data, still cause major threats. To solve this problem, this article proposes a novel encoding and decoding framework within the TEC architecture for time series predictions in bike sharing systems, with considerations of guaranteeing differential privacy. In particular, we first construct a dynamic autoencoder based on the Long Short Term Memory (LSTM) network, then the collected raw temporal data are encoded into a hidden state in the edge end. In the cloud end, the trained output weights are used to reconstruct the current time series and predict the next‐period time series according to received hidden states. This framework not only improves the computation efficiency of bike sharing system by leveraging computing power at the node end but also provides safer data publications that meet the differential privacy requirements. Experiments are conducted on real‐world datasets, results demonstrate the effectiveness of the proposed framework in providing data services with both high utility in sequence predictions and high safety level for users' privacy.

Transport infrastructure generates a huge amount of city transportation data due to the significant increasing of advanced devices, such as sensing devices, mobile devices and real-time monitors. However, transportation big data cannot be fully analyzed and utilized by urban traffic data services currently. This paper proposes a novel City Traffic Data-as-a-Service (CTDaaS), which fuses data from distributed providers. Initially, we build an Internet of Traffic Data Service (IoTDS) model to identify associations and relationships among data resources. Then a CTDaaS agent is developed under Transparent Computing paradigm and service oriented architecture. It receives user requests, fuses knowledge from a variety of data sources according to different computing models, and responses differentiated Quality of Data (QoD). Finally, an application scenario, named Park and Ride (P+R), is implemented and evaluated to demonstrate how the service works using existing dynamic city traffic data.