With the emergence of location-based services and the proliferation of location-aware devices, a significant volume of location-based social media data, such as check-in data, has been generated. Consequently, this has led to the development of Point of Interest (POI) recommendation systems. However, most research has primarily focused on POI recommendations within a single region, with relatively little attention given to cross-region POI recommendations. In this study, a cross-region POI recommendation model named UTMFPR is proposed. Initially, within the user type feature extraction module, check-in features are extracted based on user check-in data. User type embeddings are then obtained using an attention network and a Sparse Autoencoder (SAE). Subsequently, within the multi-information fusion module, user home-town preferences are extracted based on check-in data and user interest drift are achieved through nonlinear mapping. The Neural Topic Model (NTM) is employed to uncover user travel intentions. A fusion mechanism is then designed to integrate user types, interest drift, and travel intentions, thereby deriving user target region preference. Finally, within the target region POI recommendation module, POI embeddings for target region are generated by integrating geographical information. These embeddings are subsequently combined with user target region preference to facilitate the recommendation process. Experiments were conducted on two real-world datasets, and the results indicate that UTMFPR surpasses existing methods in recommendation accuracy for cross-region POI recommendations.

With the rapid development of precision agriculture, the application of computer vision in the intelligent detection of crop pests has advanced significantly. Within the honeysuckle cultivation industry, accurate pest identification is crucial for ensuring the quality of the medicinal material. However, existing detection models often underperform in two key scenarios: firstly, when pests exhibit substantial scale variations in images due to different shooting distances, and secondly, when distinguishing between morphologically similar pests based on subtle features. To address these challenges, this paper proposes a symmetry-aware model termed the YOLO-Multi-dimensional Pyramid Attention Module (YOLO-MPAM), which incorporates structural symmetry principles as its fundamental design paradigm. The core component, a Multi-dimensional Pyramid Attention Module (MPAM), employs a symmetrical pyramid architecture to achieve balanced feature processing across different scales. This design ensures scale symmetry by treating features at different resolutions equivalently, while the attention mechanism establishes channel-spatial symmetry through coordinated processing of both dimensions. The symmetrical design enables the reliable capture of discriminative features across objects of varying sizes, effectively addressing the inherent asymmetries in pest distribution and scale variation.

Since different nodes in hypergraph have different contributions to hyperedge learning, it is a hot topic in the field of graph representation learning to introduce attention mechanism into hypergraph neural network to represent the importance of different nodes or hyperedges. However, the existing hypergraph attention network methods are mostly based on the assumption that node attributes and topology are perfectly given. In practice, node links and attributes are usually fuzzy concepts, which will lead to the uncertainty of feature learning and then affect the effectiveness of node representation. Based on this, inspired by the attention mechanism and fuzzy logic, this paper proposes a new fuzzy attention hypergraph neural network (HFATN), which is used to quantify the contribution ability of different vertices and hyperedges to better learn the vector representation of nodes in the hypergraph. HFATN consists of two modules: fuzzy attention vertex convolution and fuzzy attention hyperedge convolution. In the process of node convolution and hyperedge convolution, the attention mechanism based on hyperedge and node uncertainty characteristics is introduced respectively. By fuzzing the node set and hyperedge set on the hypergraph, the membership degree of nodes and hyperedges is calculated, which is used to extract effective features to deal with the uncertainty in the hypergraph data. Finally, we conduct experiments on three benchmark datasets for hypergraph node classification. The results show that compared with the latest TDHGNN model, the classification accuracy of FHATN on the three datasets is improved by 2.28 %, 8.99 % and 1.85 % respectively.

Anomaly detection, as a critical task in time series data analysis, plays a pivotal role in ensuring industrial production safety, enhancing the precision of climate predictions and improving early warning for ocean disaster. However, due to the high dimensionality, redundancy, and non-stationarity inherent in time series data, rapidly and accurately identifying anomalies presents a significant challenge. This paper proposes a novel model CiTranGAN, which integrates the advantages of Transformer architecture, generative adversarial networks, and channel-independence strategies. In this model, the channel-independent strategy eliminates cross-channel interference and mitigates distribution drift in high-dimensional data. To mitigate redundancy and enhance multi-scale temporal feature representation, we constructed a feature extraction module that integrates downsampling, convolution, and interaction learning. To overcome the limitations of the traditional attention mechanism in detecting local trend variations, a hybrid dilated causal convolution-based multi-scale self-attention mechanism is proposed. Finally, experiments were conducted on five real-world multivariate time series datasets. Compared with the baseline models, CiTranGAN achieves average improvements of 12.48% in F1-score and 7.89% in AUC. In the ablation studies, CiTranGAN outperformed the channel-independent mechanism, the downsampling–convolution–interaction learning module, and the multi-scale convolutional self-attention mechanism, with respective average increases in AUC of 1.63%, 2.16%, and 3.47%, and corresponding average improvements in F1-score of 1.70%, 4.33%, and 2.04%, respectively. These experimental results demonstrate the rationality and effectiveness of the proposed model.

This study develops a sophisticated dynamic, multi-period intelligent traffic allocation algorithm using multi-objective data mining techniques, designed to optimize the incorporation of renewable energy systems (RESs) and electric vehicles (EVs) within electrical power grids. Given the inherent intermittency and unpredictability associated with RESs and EVs, the algorithm utilizes Dynamic Optimal Network Reconfiguration (DONR) and Capacitor Bank Switching (CBS) to address these challenges effectively. This integrated approach aims to enhance grid stability and operational efficiency, focusing on reducing energy losses, improving voltage profiles, and achieving financial savings through optimized 24-hour grid operations. The core innovation of this research is the application of the Artificial Hummingbird Algorithm (AHA), which has been adapted for the first time to tackle this multi-faceted optimization problem. By considering the impacts of variable solar generation and the demands of diverse load profiles, including substantial EV penetrations, the AHA navigates complex decision spaces to find optimal solutions. This methodology was rigorously tested using an enhanced IEEE 33-bus benchmark system, where various scenarios were simulated to evaluate the computational effectiveness of the AHA compared to other prevailing methods. The results from these simulations clearly demonstrate the superior performance of the integrated DONR and CBS strategy, particularly in managing the dynamic and stochastic nature of load demands and renewable energy inputs in real-time scenarios. The method by dynamic reconfiguration may boost the overall savings to 6903.03 $/h and decrease inefficiencies at (87.95 kW + j64.72 kVAr).

Machine learning is good at learning general knowledge and predictive knowledge from known and limited environments, and perform well in similar environments. However, their performance in unknown environments is not satisfactory. Machine learning focus on correlation learning, but correlation is not causality. The non-causal part of the correlation forms spurious correlation that affect the model's generalization ability. Therefore, it is necessary to examine machine learning from a causal perspective. By constructing structural causal models, it is found that the targets of different domains are instrumental variables of each other and can be mutually represented. They have same essential features and different domain features, which lead to causal correlation and spurious correlation with labels respectively. In this paper, a causal-inspired contrastive learning supervised model is designed to strengthen essential features and weaken domain features. With the target of improving the ability of model to capture causal correlations and reducing the interference of spurious correlations, we join transfer and autoencoder with image classification model as a cross-domain contrastive learning model. Experiments show that the proposed framework has a simple and easy-to-implement structure, it performs well on public datasets. By adopting visualization technology, the effects of this method are intuitively demonstrated.

The rise of graph neural networks has greatly contributed to the development of recommendation systems, and self-supervised learning has emerged as one of the most important approaches to address sparse interaction data. However, existing methods mostly focus on the recommendation’s accuracy while neglecting the role of recommended item diversity in enhancing user interest and merchant benefits. The reason for this phenomenon is mainly due to the bias of popular items, which makes the long-tail items (account for a large proportion) be neglected. How to mitigate the bias caused by item popularity has become one of the hot topics in current research. To address the above problems, we propose a Popularity-Debiased Graph Self-Supervised for Recommendation (PDGS). Specifically, we apply a penalty constraint on item popularity during the data enhancement process on the user–item interaction graph to eliminate the inherent popularity bias. We generate item similarity graphs with the popularity bias removed to construct a self-supervised learning task under multiple views, and we design model optimization strategies from the perspectives of popular items and long-tail items to generate recommendation lists. We conduct a large number of comparison experiments, as well as ablation experiments, on three public datasets to verify the effectiveness and the superiority of the model in balancing recommendation accuracy and diversity.

The objective of cross-city recommendation is to suggest points-of-interest (POI) in the target city that may be of interest to users, based on their check-in records from their source city. Although significant progress has been made in studying user preference transfers, there is a lack of research focusing on personalized user preference transfers. Furthermore, the mining of user preferences from the source city is impacted by errors and missing information. To address these challenges, this paper proposes a Check-In Heterogeneous Hypergraph and Personalized Preference Transfers for Cross-City POI Recommendation Method (CHHPPT). Firstly, a check-in heterogeneous hypergraph network is introduced in the user source city preference-mining module. This network, through Heterogeneous Hypergraph Embeddings (HHE), captures user preferences in the source city, thereby mitigating the impact of errors and missing information on user preference. Subsequently, in the user-personalized preference transfer module, a user’s transferable features are obtained through a POI aggregation network. These features are then combined with a meta-network and transfer networks to achieve user-personalized preference transfer. Finally, in the target city point-of-interest recommendation module, a POI-geographical graph is constructed using the geographical information of POI. This graph, in conjunction with category information, yields a joint embedding representation. The final recommendation is achieved by integrating the user-personalized preference transfer embeddings with the target city’s POI embeddings. Extensive experiments conducted on two real-world datasets demonstrate the effectiveness of CHHPPT in cross-city recommendation tasks.

Traffic flow prediction based on road network is of great significance in logistics transportation planning and traffic management dispatch. Under the dual influence of spatial and temporal dependency factors, how to improve the accuracy of traffic flow prediction has become a current research hotspot. However, existing studies mostly use static graph structures to capture spatial dependency, ignoring the dynamic implicit graph information contained in traffic flow, resulting in insufficient spatial information learned by graph models and ignoring the temporal dependency impact of traffic flow. Therefore, this paper proposes a new traffic flow prediction model (IDIGI) that integrates dynamic implicit graph information. First, the model constructs a dynamic implicit graph based on time-segmented sensor traffic flow embedding representation, which is then integrated with topological graph to form a dynamic fusion graph; Secondly, it extracts the temporal dependency of traffic flow using down-sampling sequences; Finally, it compares favorably with six baseline models on four real datasets, significantly outperforming existing graph prediction models.

To address the problem of poor stability and low accuracy of community division caused by the randomness in the traditional label propagation algorithm (LPA), a community discovery algorithm that combines seed node influence and neighborhood similarity is proposed. Firstly, the K-shell values of neighbor nodes are combined with clustering coefficients to define node influence, the initial seed set is filtered by a threshold, and the less influential one in adjacent node pairs is removed to obtain the final seed set. Secondly, the connection strengths between non-seed nodes and seed nodes are defined based on their own weights, distance weights, and common neighbor weights. The labels of non-seed nodes are updated to the labels of seed nodes with which they have the maximum connection strength. Further, for the case that the connection strengths between a non-seed node and multiple seed nodes are the same, a new neighborhood similarity combining the information between the two types of nodes and their neighbors is proposed, thus avoiding the instability caused by randomly selecting the labels of seed nodes. Experiments are conducted on six classic real networks and eight artificial datasets with different complexities. The comparison and analysis with dozens of related algorithms are also done, which shows the proposed algorithm effectively improves the execution efficiency, and the community division results are stable and more accurate, with a maximum improvement in the modularity of about 87.64% and 47.04% over the LPA on real and artificial datasets, respectively.

An overlapping community discovery algorithm integrating cycle structure and label optimization strategies is proposed. Initially, the node superiority index is defined, and the seed node set is selected based on a threshold. Then labels of non-seed nodes are updated through node similarity to decrease the number of initial labels, which helps enhance the stability of the algorithm and reduce computational complexity. Furthermore, based on cycle structure, an initial label updating sequence is generated from the descending order of node cycle ratios, subsequently combining node importance and node similarity to define label updating rules. Moreover, in cases where indicators are identical, the quantity of shared cycles is used to determine the category of labels, thereby increasing the accuracy of community partitioning. Experiments indicate that compared with five other algorithms, the proposed algorithm demonstrates highest up to 64.1% and 41.5% improvement in EQ and NMI.

In the medical field,the recognition of medical entities is often influenced by their adjacent context,the cur-rent named entity recognition methods typically rely on BiLSTM to capture the global dependency relationships within text,lacking modeling of local dependencies between characters.To resolve this problem,a Chinese medical named en-tity recognition model LENER based on local enhancement was proposed.Firstly,the representation of characters was enriched by LENER utilizing multi-source information,including phonetic,graphic and semantic features.Secondly,relative position encoding was combined to perform local attention calculations on sequence segments divided by sliding windows,and local information was fused with global information obtained from BiLSTM through nonlinear computa-tion.Finally,the recognized entity heads and tails were combined by LENER to extract the entities.The experimental re-sults show that the LENER model has excellent entity recognition capabilities,and the F1 value is improved by 0.5%to 2%compared with other models.

In recent years, recommendation systems have made significant strides through the application of graph neural networks (GNNs). However, most of the existing methods primarily focus on modeling user–item interactions, often failing to account for the distinct roles of different types of neighboring nodes during the graph convolution process. Moreover, the intricate relationships between user and item attributes are frequently underexplored, which constrains further improvement in the performance of recommendation models. To overcome these challenges, this paper proposes an attribute-aware graph convolutional network recommendation model (AAGCNR). This model accounts for the complex interrelationships between user and item attributes while distinguishing between different types of neighboring nodes during graph convolution. First, a multi-head self-attention mechanism is applied to capture the semantic relationships among attributes across various semantic spaces. Additionally, a bilinear interaction module is employed to facilitate interactions between attributes. Since different neighboring nodes exert different influences on target nodes, the model performs convolutional aggregation by leveraging the interrelationships of these attributes throughout the graph convolution process. Experiments conducted on real-world datasets reveal that AAGCNR outperforms other benchmark algorithms, particularly in terms of RECALL and NDCG metrics.

As a kind of special graph of structured data, a hypergraph can intuitively describe not only the higher-order relation and complex connection mode between nodes but also the implicit relation between nodes. Aiming at the limitation of traditional distance measurement in high-dimensional data, a new method of hypergraph construction based on set pair theory is proposed in this paper. By means of dividing the relationship between data attributes, the set pair connection degree between samples is calculated, and the set pair distance between samples is obtained. Then, on the basis of set pair distance, the combination technique of k-nearest neighbor and ε radius is used to construct a hypergraph, and high-dimensional expression and hypergraph clustering are demonstrated experimentally. By performing experiments on different datasets on the Kaggle open-source dataset platform, the comparison of cluster purity, the Rand coefficient, and normalized mutual information are shown to demonstrate that this distance measurement method is more effective in high-dimensional expression and exhibits a more significant performance improvement in spectral clustering.

An improved Adam optimization algorithm combining adaptive coefficients and composite gradients based on randomized block coordinate descent is proposed to address issues of the Adam algorithm such as slow convergence, the tendency to miss the global optimal solution, and the ineffectiveness of processing high-dimensional vectors. The adaptive coefficient is used to adjust the gradient deviation value and correct the search direction firstly. Then, the predicted gradient is introduced, and the current gradient and the first-order momentum are combined to form a composite gradient to improve the global optimization ability. Finally, the random block coordinate method is used to determine the gradient update mode, which reduces the computational overhead. Simulation experiments on two standard datasets for classification show that the convergence speed and accuracy of the proposed algorithm are higher than those of the six gradient descent methods, and the CPU and memory utilization are significantly reduced. In addition, based on logging data, the BP neural networks optimized by six algorithms, respectively, are used to predict reservoir porosity. Results show that the proposed method has lower system overhead, higher accuracy, and stronger stability, and the absolute error of more than 86% data is within 0.1%, which further verifies its effectiveness.

A complex network in reality contains a large amount of information, but some information cannot be obtained accurately or is missing due to various reasons. An uncertain complex network is an effective mathematical model to deal with this problem, but its related research is still in its infancy. In order to facilitate the research into uncertainty theory in complex network modeling, this paper summarizes and analyzes the research hotspots of set pair analysis, rough set theory and fuzzy set theory in complex network modeling. This paper firstly introduces three kinds of uncertainty theories: the basic definition of set pair analysis, rough sets and fuzzy sets, as well as their basic theory of modeling in complex networks. Secondly, we aim at the three uncertainty theories and the establishment of specific models. The latest research progress in complex networks is reviewed, and the main application fields of the three uncertainty theories are discussed, respectively: community discovery, link prediction, influence maximization and decision-making problems. Finally, the prospect of the modeling and development of uncertain complex networks is put forward.

This paper proposes an improved adaptive lion swarm optimization (LSO) algorithm integrating the chaotic search strategy and information entropy to address the problem that the standard LSO algorithm has slow convergence and easily falls into the local optimum in later iterations. At first, an adaptive factor is introduced to improve tent chaotic mapping and used for population position initialization to enhance population diversity and realize uniform traversal while ensuring random distribution, ultimately improving the global search ability. Second, to address the problem that the cub selection strategy is blind, resulting in insufficient traversal in the early stage, a dynamic step-size perturbation factor is established using the second-order norm and information entropy. Adaptive parameters are used to dynamically adjust the selection probability of different cub behaviors based on the number of iterations to suppress the premature convergence of the algorithm. Finally, tent chaotic search is employed to adaptively adjust the search range and improve the individuals with poor fitness through multiple neighborhood points of the local optimal solution, further improving the algorithm’s search speed and accuracy. Experimental results on 18 benchmark functions revealed that the proposed algorithm yields superior performance in terms of convergence speed, optimization accuracy, and ability to jump out of the local optimal solution compared with the standard LSO, gray wolf optimizer, and particle swarm optimization algorithms. Furthermore, the improved LSO algorithm was used to optimize the initial weights and thresholds of the BP neural network, and the effectiveness of the proposed algorithm was further verified by studying the house price prediction problem using two real-world datasets.

An improved algorithm based on tent chaotic map and difference mechanism is proposed to address the problem of poor population diversity and easily falling into local optimum in the lion swarm optimization algorithm. At first, an improved tent chaotic map is introduced to initialize the population position to increase diversity and ergodicity, thereby improving the global search capability. Second, a perturbation factor combined with a differential evolution mechanism is introduced to realize the adaptive positioning of the lioness and improve its ability to jump out of the local optimum. Following that, the proposed algorithm’s superiority is tested on 10 multi-type benchmark functions and compared to six swarm intelligence algorithms. Finally, the improved algorithm is applied to optimize the initial weights and thresholds of BP neural networks. Then a new model is proposed and applied to the study of house price prediction. Experiments are conducted on two standard datasets, and results show that the proposed model’s convergence speed, accuracy, and stability are better than the other three methods. On the Boston dataset, the mean square error of the training set and test set is 0.0016 and 0.0052, respectively, and the absolute error of more than 98% of the data is within 3%. On the Californian dataset, the mean square error of the training set and test set is 0.0149 and 0.0151, respectively, with more than 92.3% of the data within 5% absolute error, further validating the effectiveness, higher accuracy, and convergence performance of the proposed model.

With the rapid development of cyber-physical systems (CPS), constructing information systems with high-quality services has become an urgent requirement in both academia and industry. Hypergraphs are widely used in graph data analysis tasks due to their wide node-aware range and reasonable interpretability. This article proposes a model known as the relation-aggregated hypergraph neural network, which aims to better describe and apply complex structures and relationships in the physical world, thereby improving the management efficiency and service quality of service recommendation in CPS. First, explicit and implicit hyperedges are constructed based on the explicit and implicit relationships between nodes, respectively. Second, a global attention mechanism is adopted to calculate the importance between neighbor nodes that share implicit hyperedges with the current node, generating a relation-aggregated hypergraph. Finally, by weighting the generated hypergraph association matrix and passing the aggregated features through the hypergraph neural network, we obtain the embedding representation of nodes. Node classification experiments were conducted on three public network datasets, and the accuracy rates were 83.7%, 88.8%, and 84.9%, respectively, which were higher than those of the baseline model.

Efficient and accurate porosity prediction is essential for the fine description of reservoirs, for which an optimized BP neural network (BPNN) prediction model is proposed. Aiming at the problem that the BPNN is sensitive to initialization and converges to local optimum easily, an improved shuffled frog leaping algorithm (ISFLA) is proposed based on roulette and genetic coding. Firstly, a roulette mechanism is introduced to improve the selection probability of elite individuals, thus enhancing the global optimization ability. Secondly, a genetic coding method is carried out by making full use of effective information such as the global and local optimal solutions and the boundary values of subgroups. Subsequently, the ISFLA algorithm is verified on 12 benchmark functions and compared with four intelligent optimization algorithms, and experimental results show its good optimization performance. Finally, the ISFLA algorithm is applied to the optimization of initial weights and thresholds of the BPNN, and a new model named ISFLA_BP is proposed to study the porosity prediction problem. The logging data is preprocessed by grey correlation analysis and deviation normalization, and then the effective prediction of porosity is achieved by natural gamma, density and other relevant parameters. The performance of ISFLA_BP model is compared with the standard three-layer BPNN and four BPNN parameter optimization methods based on swarm intelligence algorithms. Experimental results show that the proposed model has higher training accuracy, stability and faster convergence speed, with a mean square error of 0.02, and its prediction accuracy for porosity is higher than that of the other five methods.