Applied Intelligence最新文献

Accurately ranking search results based on user query relevance is a complex, multi-dimensional challenge in information retrieval systems, inherently subject to ambiguity and uncertainty. This inherent complexity stems from the ambiguity and uncertainty surrounding relevance judgments. Factors like imprecise user queries, expert disagreements on relevance, and complex relationships between features of documents and queries all contribute to this. Traditional learning-to-rank algorithms often struggle to handle these uncertainties. This paper proposes a novel approach that leverages Sugeno and Choquet fuzzy integrals to model the uncertainty of features and their interactions. This allows our algorithm to make more nuanced ranking decisions. The proposed approach is extensively evaluated on major benchmark datasets like MSLR-Web10K, Istella LETOR, and WCL2R, demonstrating its effectiveness in outperforming baseline methods across standard criteria such as P@n, MAP, and NDCG@n. Notably, the proposed algorithm ranks top results, which are most crucial for user satisfaction. This practical improvement can benefit web search engines by providing users with more relevant information at the top of their search results.

In multi-label learning (MLL), label-specific feature (LSF) learning assumes that labels are determined by their inherent characteristics. However, in multi-view multi-label learning (MVMLL), the heterogeneity problem persists within the feature space. The views with varying dimensions can result in different dimensions of extracted LSF. Existing algorithms extract the LSF for each view separately, suffering the inadequate communication of the LSF and poor classification accuracy. The subspace learning method can address the dimension-inconsistency problem in multi-views by extracting extract the shared subspace for each view by substituting the original view feature space. However, the individual subspaces contain relatively homogeneous information. Based on this analysis, the GLocal Shared Subspace Learning (GLSSL) algorithm was proposed for multi-view multi-label learning to access more informative subspaces. First, the label groups were obtained through spectral clustering, entirely considering the correlation between the label groups and features to identify the specific relevant view features corresponding to each label group. Subsequently, the global shared subspace (global subspace) and local shared subspace (local subspace) were extracted from the original feature space and feature sets, respectively. Finally, the local subspace was complemented with the global subspace for LSF learning. The proposed algorithm was validated through comparative experiments with several state-of-the-art algorithms on multiple benchmark multi-view multi-label datasets.

With the rise of social networks, there has been a marked increase in offensive content targeting women, ranging from overt acts of hatred to subtler, often overlooked forms of sexism. The EXIST (sEXism Identification in Social neTworks) competition, initiated in 2021, aimed to advance research in automatically identifying these forms of online sexism. However, the results revealed the multifaceted nature of sexism and emphasized the need for robust systems to detect and classify such content. In this study, we provide an extensive analysis of sexism, highlighting the characteristics and diverse manifestations of sexism across multiple languages on social networks. To achieve this objective, we conducted a detailed analysis of the EXIST dataset to evaluate its capacity to represent various types of sexism. Moreover, we analyzed the systems submitted to the EXIST competition to identify the most effective methodologies and resources for the automated detection of sexism. We employed statistical methods to discern textual patterns related to different categories of sexism, such as stereotyping, misogyny, and sexual violence. Additionally, we investigated linguistic variations in categories of sexism across different languages and platforms. Our results suggest that the EXIST dataset covers a broad spectrum of sexist expressions, from the explicit to the subtle. We observe significant differences in the portrayal of sexism across languages; English texts predominantly feature sexual connotations, whereas Spanish texts tend to reflect neosexism. Across both languages, objectification and misogyny prove to be the most challenging to detect, which is attributable to the varied vocabulary associated with these forms of sexism. Additionally, we demonstrate that models trained on platforms like Twitter can effectively identify sexist content on less-regulated platforms such as Gab. Building on these insights, we introduce a transformer-based system with data augmentation techniques that outperforms competition benchmarks. Our work contributes to the field by enhancing the understanding of online sexism and advancing the technological capabilities for its detection.

Sepsis patients in the ICU face heightened mortality risks. There still exist challenges that hinder the development of mortality risk prediction models for sepsis patients. In the ensemble model, the differences between base classifier performance can affect the model accuracy and efficiency, and overlapping sample training will lead to repetitive learning, which reduces the model generalization. To tackle these challenges, we propose an Adaptive Weighted Stacking based on Optimal Weights Selection (AWS-OWS) model. A random sampling without replacement is employed to prevent repetitive learning in base classifiers. Additionally, a weighted function and the gradient descent algorithm is adopted to select optimal weights for base classifiers, enhancing the performance of stacking model. The MIMIC-IV dataset is used for model training and internal testing, and the independent samples from MIMIC-III are used for external validation. The results show that AWS-OWS achieves the best AUC of 0.88 in the internal test, with a threefold reduction in computation time compared to standard stacking. In external validation, it also demonstrates good model generalization. AWS-OWS significantly improves the prediction performance and model efficiency, facilitates the identification of high-risk patients with sepsis and supports clinicians in determining appropriate management and treatment strategies.

Aerial-view geo-localization aims to determine locations of interest to drones by matching drone-view images against a satellite database with geo-tagging. The key underpinning of this task is to mine discriminative features to form a view-invariant representation of the same target location. To achieve this purpose, existing methods usually focus on extracting fine-grained information from the final feature map while neglecting the importance of middle-layer outputs. In this work, we propose a Transformer-based network, named Multi-layer Local Pattern Cross Attention Network (MLPCAN). Particularly, we employ the cross-attention block (CAB) to establish correlations between information of feature maps from different layers when images are fed into the network. Then, we apply the square-ring partition strategy to divide feature maps from different layers and acquire multiple local pattern blocks. For the information misalignment within multi-layer features, we propose the multi-layer aggregation block (MAB) to aggregate the high-association feature blocks obtained by the division. Extensive experiments on two public datasets, i.e., University-1652 and SUES-200, show that the proposed model significantly improves the accuracy of geo-localization and achieves competitive results.

The recent emergence of the hybrid quantum-classical neural network (HQCNN) architecture has garnered considerable attention because of the potential advantages associated with integrating quantum principles to enhance various facets of machine learning algorithms and computations. However, the current investigated serial structure of HQCNN, wherein information sequentially passes from one network to another, often imposes limitations on the trainability and expressivity of the network. In this study, we introduce a novel architecture termed parallel proportional fusion of spiking and quantum neural networks (PPF-SQNN). The dataset information is simultaneously fed into both the spiking neural network and the variational quantum circuits, with the outputs amalgamated in proportion to their individual contributions. We systematically assess the impact of diverse PPF-SQNN parameters on network performance for image classification, aiming to identify the optimal configuration. On three datasets for image classification tasks, the final classification accuracy reached 98.2%, 99.198%, and 97.921%, respectively, with loss values all below 0.2, outperforming the compared serial networks. In noise testing, it also demonstrates good classification performance even under noise intensities of 0.9 Gaussian and uniform noise. This study introduces a novel and effective amalgamation approach for HQCNN, laying the groundwork for the advancement and application of quantum advantages in artificial intelligence computations.

Human gaze is a crucial cue used in various applications such as human-robot interaction, autonomous driving, and virtual reality. Recently, convolution neural network (CNN) approaches have made notable progress in predicting gaze angels. However, estimating accurate gaze direction in-the-wild is still a challenging problem due to the difficulty of obtaining the most crucial gaze information that exists in the eye area which constitutes a small part of the face images. In this paper, we introduce a novel two-branch CNN architecture with a multi-loss approach to estimate gaze angles (pitch and yaw) from face images. Our approach utilizes separate fully connected layers for each gaze angle prediction, allowing explicit learning of discriminative features and emphasizing the distinct information associated with each gaze angle. Moreover, we adopt a multi-loss approach, incorporating both classification and regression losses. This allows for joint optimization of the combined loss for each gaze angle, resulting in improved overall gaze performance. To evaluate our model, we conduct experiments on three popular datasets collected under unconstrained settings: MPIIFaceGaze, Gaze360, and RT-GENE. Our proposed model surpasses current state-of-the-art methods and achieves state-of-the-art performance on all three datasets, showcasing its superior capability in gaze estimation.

Distant supervised relation extraction has been increasingly popular in recent years, which generates datasets automatically without human intervention. However, the distant supervised assumption has the limitation that the generated datasets have inevitable labeling errors. This paper proposes the method of Hierarchical Symmetric Cross Entropy for Distant Supervised Relation Extraction (HSCERE) to alleviate the impact of the noisy labels. Specifically, HSCERE simultaneously utilizes two extractors with the same network structure for collaborative learning. This collaborative learning process guides the optimization of the extractor through a joint loss function, namely Hierarchical Symmetric Cross Entropy (HSCE). Within the HSCE loss, the predicted probability distribution of the extractors serves as the supervisory signal, guiding the optimization of the extractors on two levels to reduce the impact of noisy labels. The two levels include the internal optimization within each extractor and the collaborative optimization between extractors. Experiments on generally used datasets show that HSCERE can effectively handle noisy labels and can be incorporated into various methods to enhance their performance.

Real-world information is often characterized by uncertainty and partial reliability, which led Zadeh to introduce the concept of Z-numbers as a more appropriate formal structure for describing such information. However, the computation of Z-numbers requires solving highly complex optimization problems, limiting their practical application. Although linguistic Z-numbers have been explored for their computational straightforwardness, they lack theoretical support from Z-number theory and exhibit certain limitations. To address these issues and provide theoretical support from Z-numbers, we propose a Z-number linguistic term set to facilitate more efficient processing of Z-number-based information. Specifically, we redefine linguistic Z-numbers as Z-number linguistic terms. By analyzing the hidden probability density functions of these terms, we identify patterns for ranking them. These patterns are used to define the Z-number linguistic term set, which includes all Z-number linguistic terms sorted in order. We also discuss the basic operators between these terms. Furthermore, we develop a multi-criteria group decision-making (MCGDM) model based on the Z-number linguistic term set. Applying our method to predict the acceptance of academic papers, we demonstrate its effectiveness and superiority. We compare the performance of our MCGDM method with five existing Z-number-based MCGDM methods and eight traditional machine learning clustering algorithms. Our results show that the proposed method outperforms others in terms of accuracy and time consumption, highlighting the potential of Z-number linguistic terms for enhancing Z-number computation and extending the application of Z-number-based information to real-world problems.

The purpose of task-based dialogue systems is to help users achieve their dialogue needs using as few dialogue rounds as possible. As the demand increases, the dialogue tasks gradually involve multiple domains and develop in the direction of complexity and diversity. Achieving high performance with low computational effort has become an essential metric for multi-domain task-based dialogue systems. This paper proposes a new approach to guided dialogue policy. The method introduces a conditional diffusion model in the reinforcement learning Q-learning algorithm to regularise the policy in a diffusion Q-learning manner. The conditional diffusion model is used to learn the action value function, regulate the actions using regularisation, sample the actions, use the sampled actions in the policy update process, and additionally add a loss term that maximizes the value of the actions in the policy update process to improve the learning efficiency. Our proposed method is based on a conditional diffusion model, combined with the reinforcement learning TD3 algorithm as a dialogue policy and an inverse reinforcement learning approach to construct a reward estimator to provide rewards for policy updates as a way of completing a multi-domain dialogue task.