Applied Intelligence最新文献

The combined use of the Analytical Hierarchy Process (AHP) and Fuzzy Inference Systems (FISs) can significantly enhance the effectiveness of transformative projects in organizations by better managing their complexities and uncertainties. This work develops a novel multicriteria model that integrates both methodologies to assist organizations in these projects. To demonstrate the value of the proposed approach, we present an illustrative example focused on the implementation of Industry 4.0 in SMEs. First, through a review of relevant literature, we identify the key barriers to improving SMEs' capability to implement Industry 4.0 effectively. Subsequently, the AHP, enhanced through Dong and Saaty’s methodology, establishes a consensus-based assessment of the importance of these barriers, using the judgments of five experts. Next, a FIS is utilized, with rule bases automatically derived from the preceding weights, eliminating the need for another round of expert input. This paper shows and discusses how SMEs can use this model to self-assess their adaptability to the Industry 4.0 landscape and formulate improvement strategies to achieve deeper alignment with this transformative paradigm.

Subspace clustering typically clusters data by performing spectral clustering to an affinity matrix constructed in some deterministic ways of self-representation coefficient matrix. Therefore, the quality of the affinity matrix is vital to their performance. However, traditional deterministic ways only provide a feasible affinity matrix but not the most suitable one for showing data structures. Besides, post-processing commonly on the coefficient matrix also affects the affinity matrix’s quality. Furthermore, constructing the affinity matrix is separate from optimizing the coefficient matrix and performing spectral clustering, which can not guarantee the optimal overall result. To this end, we propose a new method, affinity adaptive sparse subspace clustering (AASSC), by adding Laplacian rank constraint into a subspace sparse-representation model to adaptively learn a high-quality affinity matrix having accurate p-connected components from a sparse coefficient matrix without post-processing, where p represents categories. In addition, by relaxing the Laplacian rank constraint into a trace minimization, AASSC naturally combines the operations of the coefficient matrix, affinity matrix, and spectral clustering into a unified optimization, guaranteeing the overall optimal result. Extensive experimental results verify the proposed method to be effective and superior.

In recent years, the growing interest in environmental sustainability has led to Electric Vehicle Routing Problems (EVRPs) attracting more and more attention. EVRPs involve the use of electric vehicles, which have additional constraints, such as range and recharging time, compared to conventional Vehicle Routing Problems (VRPs). The complexity and dynamic nature of solving VRPs often lead to the introduction of Routing Policies (RPs), simple heuristics that incrementally build routes. However, manually designing efficient RPs proves to be a challenging and time-consuming task. Therefore, there is a pressing need to explore the application of hyper-heuristics, in particular Genetic Programming (GP), to automatically generate new RPs. Since this method has not yet been investigated in the literature in the context of EVRPs, this study explores the applicability of GP to automatically generate new RPs for EVRP. To this end, three RP variants (serial, semiparallel, and parallel) are introduced in this study, along with a set of domain-specific terminal nodes to optimise three criteria: the number of vehicles, energy consumption, and total tardiness. The experimental analysis shows that the serial variant performs best in terms of energy consumption and number of vehicles, while the parallel variant is most effective in minimising the total tardiness. A comprehensive analysis of the proposed method is conducted to determine its convergence properties and the impact of the proposed terminal nodes on performance and to describe several generated RPs. The results show that the automatically generated RPs perform commendably compared to traditional methods such as metaheuristics and exact methods, which usually require significantly more runtime. More specifically, depending on the scenario in which they are used, the generated RPs achieve results that are about 20%-37% worse compared to the best known results for the number of vehicles in almost negligible time, in just some milliseconds.

Temporal graphs have become an essential tool for analyzing complex dynamic systems with multiple agents. Detecting anomalies in temporal graphs is crucial for various applications, including identifying emerging trends, monitoring network security, understanding social dynamics, tracking disease outbreaks, and understanding financial dynamics. In this paper, we present a comprehensive benchmarking study that compares 12 data-driven methods for anomaly detection in temporal graphs. We conduct experiments on two temporal graphs extracted from Twitter and Facebook, aiming to identify anomalies in group interactions. Surprisingly, our study reveals an unclear pattern regarding the best method for such tasks, highlighting the complexity and challenges involved in anomaly emergence detection in large and dynamic systems. The results underscore the need for further research and innovative approaches to effectively detect emerging anomalies in dynamic systems represented as temporal graphs.

The goal of bundle recommendation is to offer users a set of items that match their preferences. Current methods mainly categorize user preferences into bundle and item levels, and then use graph neural networks to obtain representations of users and bundles at both levels. However, real-world interaction data often contains irrelevant and uninformative noise connections, leading to inaccurate representations of user interests and bundle content. In this paper, we introduce a Multi-view Denoising Contrastive Learning approach for Bundle Recommendation (MDCLBR), aiming to reduce the negative effects of noisy data on users’ and bundles’ representations. We use the original view, which includes bundle and item levels, to guide data augmentation for creating augmented views. Then, we apply the multi-view contrastive learning paradigm to enhance collaboration within the original view, the augmented views, and between them. This leads to more accurate representations of users and bundles, reducing the impact of noisy data. Our method outperforms previous approaches in extensive experiments on three real-world public datasets.

Advanced Persistent Threat (APT) attack is one of the most dangerous cyber-attack techniques nowadays. Therefore, the issue of detecting and predicting the spread of APT malware in the network is a very urgent issue to help the process of preventing this attack effectively. In this paper, we propose a new approach that is capable of predicting the spread of APT malware in the network based on the APT's own behaviors. Accordingly, to predict the spread of APT malicious code in the system, we propose to use a combination of two single Susceptible‐Infected‐Recovered (SIR) models. Specifically, the first SIR model was built to predict the spread of APT malicious code to devices and computers within the organization. These devices and computers are often used by APT malicious code as a basis to escalate privileges to devices or computers containing important and sensitive information of the organization. The second SIR model has the function of predicting the spread of APT malware to a group of computers containing sensitive information or potentially causing high risks to the organization. The two SIR models will provide information about infections between computer groups in the system to help accurately predict the spread of APT malware in the system. The proposal to combine two SIR models in the article is a new proposal based on the behavior of APT malware in practice. By combining two SIR models, the proposal in this article has opened up a new approach for a number of problems predicting the spread in the internet such as malicious code in wireless sensor networks or malicious information on the social network.

Offline Reinforcement Learning (Offline RL) is able to learn from pre-collected offline data without real-time interaction with the environment by policy regularization via distributional constraints or support set constraints. However, since the policy learned from offline data under the constrains of support set is usually similar to the behavioral policy due to the overly conservative constraints, offline RL confronts challenges in active behavioral exploration. Moreover, without online interaction, policy evaluation becomes prone to inaccuracy, and the learned policy may lack robustness in the presence of sub-optimal state-action pairs or noise in a dataset. In this paper, we propose an Offline-to-Online Reinforcement Learning Approach based on Multi-action Evaluation with Policy Extension(MAERL) for improving the ability of the policy exploration and the effective value evaluation of state-action in offline RL. In MAERL, we develop four modules: (1) in the policy extension module, we design a policy extension method, which uses the online policy to extend the offline policy; (2) in the multi-action evaluation module, we present an adaptive manner to merge the offline and online policies to generate an action of the agent; (3) in the action-oriented module, we learn the action trajectories of the agent from the dataset, mitigating the issue of actions deviating excessively during environmental exploration; (4) to maintain the consistency in the agent’s actions, we propose an action temporally-aligned representation learning method to maintain the trend of actions of agents. This approach ensures that the agent’s actions align with the learned trajectories, preventing significant deviations during exploration. Extensive experiments are conducted on 15 scenarios of the D4RL/mujoco environment. Results demonstrate that our proposed methods achieve the best performance in 12 scenarios and the second-best performance in 3 scenarios compared to state-of-the-art methods. The project’s code can be found at https://github.com/FrankGod111/Policy-Expansion.git

Large-scale Pre-trained Language Models (PLMs) have become the backbones of text classification due to their exceptional performance. However, they treat input documents as independent and uniformly distributed, thereby disregarding potential relationships among the documents. This limitation could lead to some miscalculations and inaccuracies in text classification. To address this issue, some recent work explores the integration of Graph Neural Networks (GNNs) with PLMs, as GNNs can effectively model document relationships. Yet, combining graph-based methods with PLMs is challenging due to the structural incompatibility between graphs and sequences. To tackle this challenge, we propose a graph-enhanced text mutual learning framework that integrates graph-based models with PLMs to boost classification performance. Our approach separates graph-based methods and language models into two independent channels and allows them to approximate each other through mutual learning of probability distributions. This probability-distribution-guided approach simplifies the adaptation of graph-based models to PLMs and enables seamless end-to-end training of the entire architecture. Moreover, we introduce Asymmetrical Learning, a strategy that enhances the learning process, and incorporate Uncertainty Weighting loss to achieve smoother probability distribution learning. These enhancements significantly improve the performance of mutual learning. The practical value of our research lies in its potential applications in various industries, such as social network analysis, information retrieval, and recommendation systems, where understanding and leveraging document relationships are crucial. Importantly, our method can be easily combined with different PLMs and consistently achieves state-of-the-art results on multiple public datasets.