Applied Intelligence最新文献

Predicting review helpfulness (RH) to ensure that consumers make effective purchasing decisions is a significant area of study. Many scholars have attempted to develop accurate review helpfulness prediction (RHP) methodologies. However, most previous studies have mainly focused on predictions using product review texts, and few studies have used product satisfaction as indicated by star ratings, particularly the consistency between review texts and star ratings. This study proposes a novel model called BHelP-CoRT (Bidirectional Encoder Representations from Transformers based RHP model utilizing consistency of ratings and texts) to predict RH. The proposed model consists of a review text encoder, star rating encoder, and text-rating interaction. The review text encoder was developed by applying the BERT model to extract contextual semantic features embedded in review texts. The star rating encoder was designed to embed star ratings into feature vectors. The text-rating interaction was constructed by applying an attention mechanism to extract the text-rating interaction and introduce consistency into the RHP tasks. This study conducted extensive experiments to demonstrate the effectiveness of the proposed model from multiple perspectives using real-world online reviews collected from Amazon. The experimental results show that the proposed model outperforms the state-of-the-art models, indicating that it can improve the RHP performance. Specifically, this effectiveness is reflected in the processing of reviews containing inconsistent information. This study supports the marketing efforts of the e-commerce industry by providing an RHP service to address consumer information overload.

Graphical abstract

The field of RGB-D semantic segmentation has attracted considerable interest in recent times. The challenge is to develop an effective method for combining RGB images, which capture colour variations, with depth images, which provide robust information about object geometry regardless of lighting conditions. Treating both image types equally through the same convolution operator fails to take into account their inherent differences. Thus, in this paper, we propose a novel approach that combines a geometry-aware convolution (GAConv) module and a multiscale fusion module (MFM) with the aim of enhancing the performance of RGB-D image segmentation. The GAConv module effectively captures fine-grained geometric details from depth images, while the MFM module enables efficient integration of multi-scale features, allowing the network to utilise both spatial and semantic information. Extensive experimentation was conducted on the NYUv2 and SUN RGB-D datasets, wherein our model demonstrated consistent superiority over existing state-of-the-art methods in terms of pixel accuracy and mean intersection over union (mIoU).

With the widespread application of time series data, the study of classification techniques has become an important topic. Although existing multivariate time series classification (MTSC) methods have made progress, they often rely on one-dimensional (1D) time series, which limits their ability to capture complex temporal dynamics and multiscale features. To address these challenges, a Spectral Convolutional Network (SCNet) is introduced in this work. SCNet effectively transforms 1D time series data into the frequency domain using an enhanced Discrete Fourier Transform (enhanced_DFT), revealing periodicity and key frequency components while reshaping the data into a two-dimensional (2D) time series for better representation. Furthermore, it uses a Spectral Energy Prioritization method to optimize frequency domain energy distribution and a multiscale convolutional module to capture features at different scales, improving the model’s ability to analyze short-term and long-term trends. To validate the effectiveness and superiority, we conducted extensive experiments on 10 sub-datasets from the well-known UEA dataset. The results show that our proposed SCNet achieved the highest average accuracy of 74.3%, which is 2.2% higher than the current state-of-the-art models, demonstrating its potential for practical application and efficiency in MTSC task.

Knowledge distillation (KD) has become a pivotal technique in deep learning, facilitating model compression and regularization by transferring knowledge from one neural network to another, enhancing its capabilities for downstream tasks such as classification. However, real-world datasets often suffer from noisy label problems, significantly hindering neural network learning in supervised tasks. Recent advancements in KD aim to improve noise-robustness and regularization in deep neural networks through different learning paradigms. Yet, prevalent approaches often exhibit noise-prone behaviors as the student network heavily relies on the teacher’s learning. To address this challenge, we propose a robust knowledge transfer method, NoRD: a Noise-Resilient Self-Distillation framework. This approach leverages relative self-supervision combined with decision matching to minimize noise susceptibility during the knowledge transfer process. Our study evaluates this technique on CIFAR-10, CIFAR-100, and MNIST datasets with synthetic label noise. Results showcase that our method achieves 8-10% higher test accuracy compared to state-of-the-art noise-robust loss functions at noise rates exceeding 50%, surpassing well-known KD methods by 4-5% in top-1 test accuracy. The code is available at https://github.com/philsaurabh/NoRD_Applied-Intelligence.

The efficiency of multi-objective soft subspace clustering algorithms (MSSCAs) can be low when applied to large-scale datasets. This inefficiency arises because the multi-objective evolutionary algorithms (MOEAs) utilized in MSSCAs often require a large number of soft subspace clustering objective function evaluations due to their population-based nature. Moreover, relying solely on negative Shannon entropy to constrain feature weights is inadequate for soft subspace clustering algorithms. To address these issues, a knowledge-guided classification and regression surrogates co-assisted multi-objective soft subspace clustering (KCRS-MOSSC) algorithm is presented. First, an inter-cluster feature weight dissimilarity function is designed to further constrain the feature weights. Furthermore, a novel surrogate-based optimization framework called the knowledge-guided classification and regression surrogates co-assisted multi-objective evolutionary framework (KCRS-MOEF) is proposed to efficiently optimize the proposed inter-cluster feature weight dissimilarity function, intra-cluster compactness function, inter-cluster separation function, and negative Shannon entropy function. In KCRS-MOEF, a classification decision tree is utilized as the classification surrogate model to help generate a set of promising offspring, while a radial basis function (RBF) model is employed as the regression surrogate model to assist in the infill criterion by predicting the objective function values of the offspring. Furthermore, to fully leverage the knowledge of the evolutionary process, an infill criterion guided by dynamic process knowledge of elite individuals is designed to enhance the convergence and diversity of the population. Finally, a clustering ensemble strategy based on knee point guidance is proposed to generate a final solution from a set of non-dominated individuals. KCRS-MOEF outperforms state-of-the-art counterparts in terms of convergence, diversity, and time efficiency, as demonstrated in four experiments conducted on the DTLZ benchmark. Furthermore, experiments on various datasets show that the clustering performance and time efficiency of KCRS-MOSSC exceed those of comparison algorithms.

This paper takes into account a general two-player zero-sum Markov game scenario in which our agent faces multi-type opponents with multiple policies. To enhance our agent’s return against opponent’s diverse policies, a novel Decision-making Framework based on Opponent Distinguishing and Policy Judgment (DF-ODPJ) is proposed. On the basis of the pre-trained Nash equilibrium strategies, DF-ODPJ can distinguish the opponent’s type by sampling from the interaction trajectory. Then a fast criterion is proposed to judge the opponent’s policy which is proven to minimize the misjudgment probability with optimal threshold calculated. According to the identification results, appropriate policies are generated to enhance the return. The proposed DF-ODPJ is more flexible since it is orthogonal to existing Nash equilibrium algorithms and single-agent reinforcement learning algorithms. The experimental results on grid world, video games, and UAV aerial combat environments illustrate the effectiveness of DF-ODPJ. The code is available at https://github.com/ChenXJ295/DF-ODPJ.

Stock prices are influenced by numerous factors, including social media, news, and financial reports, serving as indicators of financial market dynamics. However, harnessing diverse information from different sources and structures to predict price trends remains challenging. In this paper, we propose a dual-stage deep learning model based on the Mixture-of-Expert (MoE) mechanism. In stage one, three distinct expert networks encode information about price movements, financial news, and investor sentiments through multi-source interaction attention. In stage two, a gated network dynamically fuses outputs, capturing temporal relationships in windowed data. Experimental results on the Chinese stock market demonstrate our model outperforms existing ones in forecasting tasks.

Code completion is a crucial aspect of contemporary integrated development environments (IDEs), as it not only streamlines the software development process but also bolsters the quality of software products. By leveraging large-scale codes to learn the probability distribution among code token units, deep learning methods have demonstrated significant improvements in the accuracy of token unit recommendations. However, the efficacy of code completion employing deep learning is often compromised by information loss. To mitigate this issue, we introduce a novel code language model that incorporates both the pointer network and the Transformer-XL architecture to surpass the constraints of current approaches in code completion. Our proposed model accepts as input the original code snippet and its corresponding abstract syntax tree (AST), utilizing the Transformer-XL model as the foundational architecture for capturing long-term dependencies. Additionally, we incorporate a pointer network as an adjunct component to forecast Out-of-Vocabulary (OoV) words. Our approach has been rigorously evaluated on the authentic PY150 and JS150 datasets. The comparative experimental results demonstrate the effectiveness of our model in improving the accuracy of the code completion task at the token unit level.

In visual simultaneous localization and mapping (SLAM), loop closure detection plays an irreplaceable role in eliminating cumulative errors, optimizing robot poses, and ensuring map consistency. Most loop closure detection algorithms adopt single feature or feature fusion to detect loop closures, which makes it difficult to ensure accuracy in environments with changing lighting or high-similarity scenes. In this work, image features and motion trajectories are combined to improve the effectiveness of loop closure detection via a staged detection method. First, histogram equalization is used to reduce the algorithm’s sensitivity to lighting. Then, LBP features are used to divide keyframes into multiple sequences, and the sequence where the loop closure candidate frame is located is determined according to the image feature matching results. Then, the most matched keyframe is searched in the sequence as a candidate loop closure. Finally, the true loop closure is confirmed by comparing the motion trajectory similarity to improve the algorithm’s adaptability to high-similarity scenes. The experimental results show that in different application scenarios, the proposed method can achieve good results in terms of precision, recall, area under the curve (AUC), and recall when the precision is 100%.

Dynamics and uncertainty are the fundamental reasons for the difficulty in accurately predicting traffic flow. In recent years, graph convolutional networks have been widely used in traffic flow prediction because of their excellent dynamic feature mapping ability. However, the existing models usually overlook the correlations among the nodes and the complex impact of external factors on traffic flow, which make it challenging to explore the complex spatial-temporal features. To overcome these shortcomings, we propose a novel Spatial-temporal Clustering enhanced Multi-Graph Convolutional Network (SCM-GCN) for traffic flow prediction. First, a Spatial-Temporal Clustering (STS) module based on the improved adjacency matrix DBSCAN clustering algorithm is constructed, this module divides traffic nodes into multiple highly correlated clusters, each of which consists of multi-graph features and time-varying features. Then, a Multi-Graph Spatial Feature Extraction (MGSFE) module that integrates the graph convolution operation and attention mechanism is designed to extract dynamic spatial features of multi-graph and time-varying features. Next, the Time-Varying Feature Extraction (TVFE) module based on the dilated convolution and gated attention mechanism is constructed. It integrates the output of the MGSFE module to extract dynamic temporal features of time-varying features and output the predicted values. Finally, the comparison and ablation experiments on four datasets show that the proposed model performs better than state-of-the-art models. The key source code and data are available at https://github.com/Bounger2/SCMGCN.