Knowledge graphs play a pivotal role in structuring human knowledge within artificial intelligence systems. Nonetheless, knowledge distribution is markedly uneven across languages, and linguistic community activity can hinder the performance and scale. Cross-lingual transfer learning emerges as a predominant effective strategy to surmount linguistic barriers, facilitating knowledge transfer across natural languages. This paper reviews cross-lingual knowledge acquisition for knowledge graphs, offering the first systematic integration of cross-lingual transfer paradigms and resources in this field. It critically examines the state of research across subtasks (including named entity recognition, relation extraction, coreference resolution and entity linking). Despite the advancements facilitated by multilingual word embeddings, pre-trained language models and large language models, persistent challenges such as language bias-induced alignment difficulties and low transfer efficiency continue to impede progress. Enhancing model effectiveness through both paradigms and resources will benefit the future construction of multilingual or minor-language knowledge graphs.

Cuckoo Search (CS) has demonstrated competitive performance in single- and multi-objective optimization. However, for many-objective optimization problems (MaOPs) with irregular Pareto fronts (PFs), the proliferation of dominance-resistant solutions severely weakens Pareto dominance-based selection pressure, hindering simultaneous convergence and diversity. To address this, a collaborative many-objective CS framework based on a new α-dominance mechanism, termed Nα-CMaOCS, is proposed. First, a collaborative co-evolution strategy is embedded into the CS search mechanism via a candidate-sharing pool covering the entire exploration–exploitation process. Specifically, the global update incorporates a Cauchy operator and a convergence-aware weighting factor to enhance exploration while guiding convergence. Conversely, the local update employs a convex-combination factor and a differential-vector strategy to improve exploitation precision on irregular PFs. Furthermore, a modified operator evolution mode adjusts the behavior of both global and local operators to improve robustness. Second, the proposed Nα-dominance mechanism adaptively expands or contracts the dominance region to reinforce selection pressure, and its theoretical validity is rigorously established via three lemmas and three theorems, which substantiate the strict monotonicity property, hierarchical implication theorems, and asymptotic convergence guarantee. A two-stage environmental selection strategy combining local–global principles is also developed, utilizing perpendicular Euclidean distance for subspace prioritization to mitigate overcrowding. Additionally, a fitness-based evaluation criterion preserves high-quality individuals. Extensive experiments on 15 irregular-PF benchmarks and four engineering problems, using eight state-of-the-art MaO algorithms, demonstrate that Nα-CMaOCS achieves superior convergence and robustness, indicating strong competitiveness for complex MaOPs.

Feature selection plays a fundamental role in machine learning by reducing dimensionality, improving interpretability, and enhancing computational efficiency, with wide applications in domains such as bioinformatics, text mining, and intelligent systems. However, the rapid growth of large-scale data, deep learning architectures, and distributed learning environments has introduced new challenges that are not fully addressed in existing surveys. From an artificial intelligence perspective, this paper provides a comprehensive and critical synthesis of feature selection methodologies. We systematically analyze the main feature selection paradigms, including filter, wrapper, embedded, and ensemble approaches, and further examine recent advances in heuristic optimization-based feature selection. Beyond algorithm description, we present a comparative analysis framework that reveals how different feature selection strategies behave under varying data characteristics, such as high dimensionality and feature dependency. From an engineering perspective, this review discusses how these methods impact practical applications, highlighting challenges in scalability, stability, and model selection in real-world scenarios, including high-dimensional biomedical data and streaming environments. Furthermore, emerging directions such as explainability-guided feature selection, federated and distributed feature selection, and deep learning–integrated feature selection are analyzed in terms of their practical implications. This work aims to bridge methodological advances and application needs, providing structured insights into the feature selection research landscape while identifying key challenges and future research opportunities.

Coverage optimization in WSNs is critical for disaster warning and industrial monitoring, but is challenging due to multi-modality and high-dimensionality. Particle swarm optimization (PSO) offers fast convergence and simple parameter tuning, yet suffers from premature convergence and parameter sensitivity in multi-peak problems. To address these issues, we propose a Quad-module Ring-Competitive PSO (QRC-PSO). It comprises four heterogeneous subgroups with distinct parameter configurations for global exploration, local exploitation, balanced search, and perturbation enhancement. Subgroups evolve independently but exchange elite particles via a ring-topology migration strategy: every 20 iterations, the best three particles of each subgroup move clockwise to the next subgroup and replace its three worst ones, enabling high-quality solution diffusion while preserving diversity. Simulations on a 100 m × 100 m field with 20 and 30 nodes show that QRC-PSO achieves coverage rates of 84.69% and 98.62%, outperforming GA, standard PSO, APSO, LPSO, ALPSO, GWO, and DE. Tests on a 500 m × 500 m area with 500 and 750 nodes further confirm its superiority. These results demonstrate that the proposed subgroup structure and competitive mechanism effectively overcome traditional PSO weaknesses, making QRC-PSO an efficient and reliable solution for WSN coverage optimization.

Engineering constrained optimization problems (e-COPs) present computational challenges for conventional metaheuristic algorithms, particularly regarding population diversity preservation within constrained feasible domains, wherein algorithms frequently encounter premature convergence. This paper introduces a dual-nest pigeon-inspired optimization (DNPIO) algorithm with three key enhancements: a dual-nest operator using an exponential-sorting selection mechanism that balances global and local search strategies; Latin hypercube sampling for population initialization; and a damping boundary condition to explore constraint boundaries. DNPIO was validated using the CEC2017 benchmark dataset. Statistical analysis confirms its superiority, with an average Friedman rank of 1.10 across 29 test functions, with pairwise Wilcoxon tests demonstrating statistically significant advantages in 20 instances. To demonstrate practical applicability, DNPIO was also tested on three complex engineering design problems from CEC2020, achieving optimal solutions for the Welded Beam, Gear Reducer, and Multi-Plate Clutch Brake problems.

The Pigeon-Inspired Optimization (PIO) algorithm constitutes a metaheuristic method derived from the homing behaviour of pigeons. Initially formulated for three-dimensional path planning in unmanned aerial vehicles (UAVs), the algorithm has attracted considerable academic and industrial interest owing to its effective balance between exploration and exploitation, coupled with advantages in real-time performance and robustness. Nevertheless, as applications have diversified, limitations in convergence precision and a tendency toward premature convergence have become increasingly evident, highlighting a need for improvement. This review systematically outlines the developmental trajectory of the PIO algorithm, with a particular focus on its core applications in UAV navigation, multi-objective formulations, and a spectrum of variant models that have emerged in recent years. It offers a structured analysis of the foundational principles underlying the PIO. It conducts a comparative assessment of various performance-enhanced versions, including hybrid models that integrate mechanisms from other optimization paradigms. Additionally, the strengths and weaknesses of distinct PIO variants are critically examined from multiple perspectives, including intrinsic algorithmic characteristics, suitability for specific application scenarios, objective function design, and the rigor of the statistical evaluation methodologies employed in empirical studies. Finally, this paper identifies principal challenges within current PIO research and proposes several prospective research directions. Future work should focus on mitigating premature convergence by refining the two-phase search structure and adjusting the exponential decrease of individual numbers during the landmark operator. Enhancing parameter adaptation strategies, potentially using reinforcement learning for dynamic tuning, and advancing theoretical analyses on convergence and complexity are also critical. Further applications should be explored in constrained path planning, Neural Architecture Search (NAS), and other real-world multi-objective problems. For Multi-objective PIO (MPIO), key improvements include controlling the growth of the external archive and designing more effective selection mechanisms to maintain convergence efficiency. These efforts are expected to strengthen both the theoretical foundation and practical versatility of PIO and its variants.

Weighted fuzzy production rules (WFPRs) provide superior expressiveness and interpretability in knowledge engineering area. However, manual construction of WFPRs is labor-intensive, time-consuming, and inherently subjective, which greatly restricts their practical application. The back propagation neural network (BPNN) has been widely adopted for automatic WFPR extraction. Nevertheless, its high sensitivity to initial weight configurations frequently results in premature convergence to local optima, generating redundant, poorly interpretable rule sets that compromise the inherent interpretability advantage of WFPRs. This paper proposes an elite dynamic scout-guided grey wolf optimizer (EDSG-GWO) and integrates it into a BPNN-based WFPR extraction framework to optimize network initial weights. The EDSG-GWO incorporates a nonlinear convergence factor, dynamic weighted position updating, an elite opposition-based learning mechanism, and an adaptive scout bee perturbation strategy to effectively balance global exploration and local exploitation. Unlike existing GWO variants, the EDSG-GWO achieves the synergistic integration of the four above strategies, collectively enhancing convergence accuracy and exploration capability. Numerical experiments are conducted on twelve benchmark functions. The results demonstrate that the EDSG-GWO delivers competitive optimization accuracy and convergence speed. Validated on the PIMA Indians Diabetes Database, the optimized BPNN attains a test accuracy of 72.92%, which is comparable to other metaheuristic-based approaches. More notably, the extracted WFPRs reach an accuracy of 77.08%, outperforming the baseline method by a notable margin. The four extracted rules involve merely six core diagnostic features, whose weight distributions are highly consistent with established medical knowledge. This contributes to a concise, clinically plausible, and highly interpretable rule set for auxiliary diabetes diagnosis. Further validation on the Breast Cancer Wisconsin dataset yields a WFPRs testing accuracy of 94.15%, confirming the framework’s generalizability.

Weighted Fuzzy Production Rules (WFPRs) are vital for Clinical Decision Support Systems (CDSSs), significantly impacting diagnostic accuracy and bridging the gap between data-driven insights and actionable clinical decisions through knowledge engineering. This paper proposes an integrated approach combining the Dynamic Dimension Adjustment Harmony Search (DDA-HS) Algorithm and Back Propagation Neural Networks (BPNNs) to enhance WFPR extraction accuracy. DDA-HS dynamically adjusts search space dimensions through fitness evaluations, optimizing initial weights in BPNNs and leveraging an absorbing Markov chain to enhance transition probabilities, supporting exploration and avoiding local optima in high-dimensional spaces. Evaluated against existing optimization methods including Harmony Search (HS), Cuckoo Search (CS), Adaptive Global Optimal Harmony Search (AGOHS), and Harmony Search with Cuckoo Search (HSCS) Algorithms, DDA-HS achieves 74.48% accuracy for BPNN classification and 77.08% for WFPR classification on the PIMA dataset, representing improvements of 3.6% and 6.5%, respectively. WFPR extraction enhances BPNN interpretability by revealing feature influences on decision-making, improving both accuracy and transparency. The proposed method offers a robust framework for reliable and interpretable CDSSs in healthcare.

The Wind Driven Optimization (WDO) algorithm is a metaheuristic technique inspired by atmospheric flow dynamics. Although structurally simple, WDO suffers from static search behavior and diversity loss caused by global wind-speed-based sorting. It also lacks fine-grained local search capabilities, which limit its performance in complex, high-dimensional, and constrained tasks. To address these issues, this study proposes an improved optimization model by integrating WDO, Beluga Whale Optimization (BWO), and the Golden Sine Algorithm (GSA). WDO removes wind-speed sorting to preserve diversity and enable multi-strategy fusion. Subsequently, WDO is integrated with BWO to enhance global search capabilities, while GSA is adopted for fine-grained local exploitation. Consequently, the integrated WDO-BWO-GSA approach is applied to estimate the optimal hardware-software (HW-SW) partitioning. Experiments on 24 benchmark functions including both ablation and comparative studies were conducted to evaluate the effectiveness of the proposed algorithm. The results confirmed that the integrated WDO-BWO-GSA significantly enhanced global exploration and local exploitation. Comparative results further demonstrated that the proposed integrated WDO-BWO-GSA outperformed WDO, BWO, Grey Wolf Optimizer (GWO), Whale Optimization Algorithm (WOA), Particle Swarm Optimization (PSO), and GSA, achieving average convergence speed improvements of 62.07

Harmony search algorithm (HSA) is extensively utilized in engineering optimization. Nevertheless, it encounters problems of slow convergence and reduced accuracy, which hinder its capability to escape local optima. This paper proposes HSA-DELF, a novel hybrid algorithm that combines differential evolution (DE) and Lévy flight (LF) techniques to enhance the performance of HSA. HSA-DELF leverages multi-mutation strategies of DE and LF random walk combined with weighted individuals to improve exploration and exploitation based on population fitness standard deviation comparison, and adopts pairwise iterative updates of the population to achieve faster convergence and higher solution quality. Extensive experiments were conducted to validate performance on 23 classic benchmark functions and 12 CEC 2022 benchmark functions, followed by comprehensive testing on 7 engineering problems, demonstrating the superiority of HSA-DELF. Comparative analysis with 5 well-known algorithms (HSA, DE, CSA, GA, and PSO) and 4 HSA variants (IHS, MHSA, IHSDE, and IMGHSA) confirmed the robustness of HSA-DELF. Statistical results, including best, mean, standard deviation, and worst values, consistently highlight the superior performance of HSA-DELF in terms of convergence speed, solution quality, and robustness. The Wilcoxon signed-rank test further corroborated these significant advantages. HSA-DELF showed notable improvements in 6 out of 7 engineering problems, achieving an accuracy of 85.71%. This study establishes HSA-DELF as an effective and reliable method for solving complex engineering optimization problems.

Coverage optimization in wireless sensor networks (WSNs) is critical due to two key challenges: (1) high deployment costs arising from redundant sensor placement to compensate for blind zones, and (2) ineffective coverage caused by uneven node distribution or environmental obstacles. Cuckoo Search (CS), as a type of Swarm Intelligence (SI) algorithm, has garnered significant attention from researchers due to its strong global search capability enabled by the Lévy flight mechanism. This makes it well-suited for solving such complex optimization problems. Based on this, this study proposes an improved Cuckoo Search algorithm with multi-strategies (ICS-MS), motivated by the ‘no free lunch’ theorem’s implication that no single optimization strategy universally dominates. This is achieved by analyzing the standard CS through Markov chain theory, which helps identify areas for enhancement after characterizing the WSN and its coverage issues. Subsequently, the strategies that constitute ICS-MS are individually explained. The evaluation of the proposed ICS-MS is carried out in two phases. First, a numerical comparison is provided, a numerical comparison is presented by contrasting the performance of ICS-MS with the standard CS and its variations employing different strategies in terms of function optimization results. Second, a series of coverage optimization experiments are conducted under various scenarios. The experimental results demonstrate that ICS-MS exhibits significant improvements in both test function optimization and WSN coverage applications. In high-dimensional optimization problems, all enhancement strategies of ICS-MS prove independently effective, showing strong robustness, faster convergence speed, and higher solution accuracy. For WSN coverage optimization, the ICS-MS algorithm outperforms comparative algorithms. At 200 iterations, it achieves an average coverage increase of 2.32-22.17% for 20-node deployments and 2.75-22.21% for 30-node deployments. At 1000 iterations, coverage improves by 1.78-21.65% for 20-node deployments and 1.23-20.99% for 30-node deployments. Additionally, the algorithm demonstrates enhanced stability, more uniform node distribution, and reduced optimization randomness. These improvements collectively elevate coverage rates while lowering deployment costs.

Signal-to-noise ratio (SNR) fluctuations significantly affect spectrum sensing performance in wireless communications. Traditional convolutional neural network (CNN) exhibits limited feature extraction capabilities and inefficient feature utilization at low SNR levels, leading to suboptimal spectrum sensing performance. This paper proposes a spectrum sensing method based on a multi-scale feature fusion network (MSFFNet) to address this issue. First, the proposed method employs a multi-scale feature extraction block (MSFEB) to capture multi-scale information from the input data comprehensively. Next, an adaptive feature screening strategy (AFSS) highlights key features while suppressing redundant information. Finally, a multi-level feature fusion mechanism (MLFFM) optimizes and integrates features across scales and levels, enhancing spectrum sensing performance. Simulation results demonstrate that compared to other methods, the proposed approach achieves superior performance in low-SNR communication scenarios. At an SNR of -14 dB, the detection probability Pd reaches 0.936, while the false alarm probability Pfa is only 0.1. Furthermore, this paper constructs a multi-level mixed-SNR dataset to simulate real communication environments and enhance the robustness of spectrum sensing.

With the exponential growth of data volume, information overload has become a problem for users, and recommender systems were employed to solve this problem. The purpose of the recommender system is to mine the information of interest to users from massive amounts of data. However, some problems have not been well addressed in recommender systems, e.g., the sparse interaction data between users and items and the cold -start problems when making recommendations to new users. In recent years, incorporating knowledge graphs as side information to recommender systems by knowledge graph embedding techniques has attracted considerable interest, because the rich information contained in the knowledge graph can effectively solve the above problems. This paper provides a systematic review of recommender systems based on knowledge graph embedding in terms of methods and applications. Specifically, some basic notions of recommender systems and knowledge graphs are first briefly introduced, followed by a detailed description of how existing methods associate knowledge graph embedding and recommender systems. In addition, a series of related recommendation application scenarios are summarized along with information and statistics on related datasets.

Source code representation has garnered significant attention owing to its critical role in solving software engineering problems. There are various methodologies for representing code, and the abstract syntax tree (AST) is one of the most widely used techniques. However, current AST-based approaches need to pay more attention to the impact of AST generation, resulting in high duplication of the tree structure. Additionally, most AST-based methods struggle to balance homophily and structural equivalence in sequence sampling while failing to express semantic information in node feature learning. These limitations lead to poor source code representation and reduced performance in solving software engineering problems. To address these challenges, we propose a novel model that combines a deduplication algorithm, a hybrid sampling strategy, and an optimized skip-gram feature learning approach to represent source code. To evaluate the efficacy of our proposed model, we conduct experiments on OJ datasets for program classification. Our results demonstrate that the proposed model reduces the duplication of the tree structure and improves the representation performance of source code compared with other methods.

With the exponential growth of data volume, information overload has become a problem for users, and recommender systems were employed to solve this problem. The purpose of the recommender system is to mine the information of interest to users from massive amounts of data. However, some problems have not been well addressed in recommender systems, e.g., the sparse interaction data between users and items and the cold-start problems when making recommendations to new users. In recent years, incorporating knowledge graphs as side information to recommender systems by knowledge graph embedding techniques has attracted considerable interest, because the rich information contained in the knowledge graph can effectively solve the above problems. This paper provides a systematic review of recommender systems based on knowledge graph embedding in terms of methods and applications. Specifically, some basic notions of recommender systems and knowledge graphs are first briefly introduced, followed by a detailed description of how existing methods associate knowledge graph embedding and recommender systems. In addition, a series of related recommendation application scenarios are summarized along with information and statistics on related datasets.

The issue of increasing the range covered by a wireless sensor network with restricted sensors is addressed utilizing improved CS employing the PSO algorithm and opposition-based learning (ICS-PSO-OBL). At first, the iteration is carried out by updating the old solution dimension by dimension to achieve independent updating across the dimensions in the high-dimensional optimization problem. The PSO operator is then incorporated to lessen the preference random walk stage's imbalance between exploration and exploitation ability. Exceptional individuals are selected from the population using OBL to boost the chance of finding the optimal solution based on the fitness value. The ICS-PSO-OBL is used to maximize coverage in WSN by converting regional monitoring into point monitoring utilizing the discretization method in WSN. In the experiments, the ICS-PSO-OBL with the standard CS and three CS variants (MACS, ICS-2, and ICS) are utilized to execute the simulation experiment under different numbers of nodes (20 and 30, respectively). The experimental results reveal that the optimized coverage of ICS-PSO-OBL is 18.36%, 7.894%, 15%, and 9.02% higher than that of standard CS, MACS, ICS-2, and ICS when the number of nodes is 20. Moreover, it is 16.94%, 9.61%, 12.27%, and 7.75% higher when the quantity of nodes is 30, the convergence speed of ICS-PSO-OBL, and the distribution of nodes is superior to others.

Wind Driven Optimization (WDO) Algorithm is a novel metaheuristic algorithm inspired by the continuous flow of air resulting from differences in air pressure until the air reaches a state of pressure balance. Owing to its simple structure, few parameters, intuitive nature, and straightforward programming, WDO has garnered increasing attention from scholars since its inception. WDO’s novelty lies in its utilization of aerodynamic principles to orchestrate the search process, WDO draws on the dynamics of wind and atmospheric pressure differences to propel the search for optimal solutions. However, WDO has limitations such as sensitivity to algorithm parameters and premature convergence. Consequently, various WDO variants have been proposed to overcome the limitations of the original WDO. To identify potential avenues for further research and to develop WDO for future investigation. This article systematically reviews WDO and its variants from multiple perspectives. Initially, the principle of WDO is outlined. Subsequently, the impact of modifications to the WDO on its overall effectiveness is investigated. Furthermore, the distinctive characteristics of WDO variants and their practical applications are analyzed. Moreover, the conclusions of the review are summarized, and future research directions for WDO variants and their applications are described.

The state space explosion, a challenge analogous to that encountered in a Petri net (PN), has constrained the extensive study of fuzzy Petri nets (FPNs). Current reasoning algorithms employing FPNs, which operate through forward, backward, and bidirectional mechanisms, are examined. These algorithms streamline the inference process by eliminating irrelevant components of the FPN. However, as the scale of the FPN grows, the complexity of these algorithms escalates sharply, posing a significant challenge for practical applications. To address the state explosion issue, this work introduces a parallel bidirectional reasoning algorithm for an FPN that utilizes reverse and decomposition strategies to optimize the implementation process. The algorithm involves hierarchically dividing a large-scale FPN into two sub-FPNs, followed by a converse operation to generate the reversal sub-FPN for the right-sub-FPN. The detailed mapping between the original and reversed FPNs is thoroughly discussed. Parallel reasoning operations are then conducted on the left-sub-FPN and the resulting reversal right-sub-FPN, with the final result derived by computing the Euclidean distance between the outcomes from the output places of the two sub-FPNs. A case study is presented to illustrate the implementation process, demonstrating the algorithm’s significant enhancement of inference efficiency and substantial reduction in execution time.