International audience

ABSTRACT Aiming at the problems of slow convergence speed, susceptibility to local optima, and insufficient diversity when the multi‐objective firefly algorithm (MOFA) is leveraged to solve the microgrid dispatching optimization problem, the research proposes an improved multi‐objective firefly algorithm (IMOFA) and applies it to microgrid dispatching. An enhanced attraction factor is designed and incorporated into the movement formula of dominant fireflies to strengthen attractiveness and accelerate convergence. For non‐dominant fireflies, an external archive guidance term is introduced into their movement formula, which optimizes their movement pattern, enhances global exploration capability, and avoids local optima. A hierarchical penalty function is proposed. By dynamically adjusting the penalty factor and designing different indicator function differences of equality constraints and inequality constraints to penalize terms, the constraint handling effect is improved. An environmental selection mechanism based on adjacent distance is constructed. The adjacent distance is leveraged to measure the crowding degree of solutions to maintain the diversity and uniform distribution of offspring solutions. Based on IMOFA, a microgrid dispatching optimization model (IMOFA‐MG) with operation cost and pollutant gas treatment cost as objective functions is established. To verify the optimization performance of IMOFA and the dispatching efficiency of IMOFA‐MG, comparison tests involving IMOFA and 5 other algorithms are carried out, and dispatching simulation experiments are conducted for the IMOFA‐MG model. The results show that on benchmark test problems, IMOFA has better convergence performance and can solve optimization problems; in dispatching simulation experiments, the IMOFA‐MG model can obtain a set of dispatching scheme Pareto frontiers with better convergence and distribution. The effectiveness of IMOFA‐MG in dispatching optimization is verified.

Conventional opinion dynamics models primarily focus on interpersonal interactions within closed environments, often overlooking the substantial influence of external events. However, in open social networks, public opinion is highly dynamic, frequently shifting in response to rapidly evolving and unpredictable external stimuli. To address these limitations, this paper proposesOpenNOD(Open Network Opinion Dynamics), a novel framework for modeling event-driven opinion dynamics in open social environments. Grounded in real-world social media data, OpenNOD introduces a group-level opinion dynamics model that captures both intergroup influences among opinion communities and the impact of external events. To support this framework, two key technical modules are introduced: (1) As online opinions are primarily expressed through natural language, an in-context learning-based method is designed for individual opinion classification. This approach enhances opinion extraction accuracy by leveraging class-aware contrastive similarity to select optimal demonstration examples for large language models (LLMs); (2) To quantify the influence of external events over time, a temporal event influence measurement module is designed to integrate both the intensity of events and their directional impact on public sentiment. By jointly modeling internal and external factors, OpenNOD enables more realistic modeling of how real-world events trigger rapid shifts in public opinion and reshape influence structures. Extensive experiments on two real-world datasets demonstrate the framework's adaptability across diverse topics and its effectiveness in capturing nonlinear opinion dynamics. These findings highlight OpenNOD's potential to bridge the gap between traditional modeling and the complex behaviors observed on modern online platforms, offering a robust foundation for future studies.

This paper addresses the important yet underexplored task of multi-class sentiment analysis (MCSA), which remains challenging due to the subtle semantic differences between adjacent sentiment categories and the scarcity of high-quality annotated data.To tackle these challenges, we propose RD-MCSA (Rationales and Demonstrations-based Multi-Class Sentiment Analysis), an In-Context Learning (ICL) framework designed to enhance MCSA performance under limited supervision by integrating classification rationales with adaptively selected demonstrations.First, semantically grounded classification rationales are generated from a representative, class-balanced subset of annotated samples selected using a tailored balanced coreset algorithm.These rationales are then paired with demonstrations chosen through a similaritybased mechanism powered by a multi-kernel Gaussian process (MK-GP), enabling large language models (LLMs) to more effectively capture fine-grained sentiment distinctions.Experiments on five benchmark datasets demonstrate that RD-MCSA consistently outperforms both supervised baselines and standard ICL methods across various evaluation metrics.

Stance detection—the task of inferring users’ implicit attitudes from textual statements—plays a crucial role in sociological research by enabling systematic analysis of public opinion on complex societal issues. However, existing methods struggle with macro topics—broad, abstract subjects (e.g., ‘globalization’ or ‘AI development’)—where individuals often express views indirectly or avoid explicit statements. Traditional approaches, which rely on overt linguistic markers, fail to capture the nuanced discourse characteristic of such topics. To address this challenge, we propose STS4MTS, a dual-layer framework that integrates latent topic modeling with hierarchical semantic analysis. First, a discourse-driven Discriminative Latent Dirichlet Allocation (Disc-LDA) model decomposes macro topics into contextually relevant sub-topics (e.g., ‘labor market impacts’ under ‘globalization’), preserving thematic nuances. Next, stance detection is performed hierarchically: (1) at the sub-topic level, large language models (LLMs) conduct fine-grained semantic analysis of stances expressed in text, and (2) at the macro topic level, a feature-based classifier aggregates stance signals across sub-topics to infer implicit attitudes—even when the macro topic itself remains unmentioned. Experiments on social media and financial discourse datasets show that STS4MTS outperforms state-of-the-art methods by 12–18% in detecting indirect stances on macro topics. These findings underscore the effectiveness of combining topic decomposition with LLM-driven analysis, offering a scalable approach for advancing sociological research on abstract and contentious societal issues.

Hyperspectral images (HSIs) and light detection and ranging (LiDAR) data provide complementary spectral-spatial and elevation information, respectively, whose fusion can significantly improve classification accuracy. However, their inherent heterogeneity challenges effective spectral-geospatial integration. Although convolutional neural networks (CNNs), recurrent neural networks (RNNs), and Transformer models have advanced multimodal remote sensing classification, each shows distinct limitations. CNNs excel in spatial feature aggregation but lack global context, whereas RNNs and Transformers, despite capturing long-range spectral features, face issues such as computational inefficiency. To address these limitations, we propose a dual-feature attention hybrid graph convolutional network (GCN) Mamba network (DAHGMN) for joint HSI and LiDAR classification. Specifically, multimodal image cubes are first extracted by a CNN to obtain initial features. Subsequently, a dual-feature attention (DA) module is introduced to adaptively recalibrate spectral and spatial feature weights, enhancing discriminability. Furthermore, we propose a hybrid GCN Mamba (HGM) module with both low parameter complexity and time complexity, which combines the local geometric modeling capability of GCNs with the global long-range dependency modeling of Mamba’s state-space model (SSM). A probability-based decision fusion strategy is employed to integrate multi-level classification results, achieving an efficient combination of the spatial-spectral contextual features. Extensive experiments on three benchmark HSI-LiDAR datasets demonstrate that DAHGMN achieves superior classification accuracy while significantly reducing parameter complexity compared to state-of-the-art methods. The implementation code is publicly available at https://github.com/RogsXie/DAHGMN.

The world is evolving in a pace faster than ever, in terms of the ways in which human can interact with the rest of the world virtually or physically. Our exploration together with Carl along the road involves many models and methods that help better understand users' needs, desires and intentions, specify clearer system requirements and know-how, describe simpler service interface to find a match between requests and capabilities. There are implicit relations between these research topics, namely, building smarter software systems that can adapt to the needs and intentions of users, as well as the concrete operational context and environment. This paper may look retro-futurism in a sense, as we are looking forward to fulfillment the early AI and SE visions, by reviewing viable fundamental software requirements engineering principles and service system thinking methods, and at the same time by making sense of the on-going development of smart applications with the latest intelligent computing technologies.

Chinese Grammatical Error Diagnosis (CGED) suffers the problems of numerous types of grammatical errors and insufficiency of training data. In this paper, we propose a string editing based CGED model that requires less training data by using a unified workflow to handle various types of grammatical errors. Two measures are proposed in our model to enhance the performance of CGED. First, the detection and correction of grammatical errors are divided into different stages. In the stage of error detection, the model only outputs the types of grammatical errors so that the tag vocabulary size is significantly reduced compared with other string editing based models. Secondly, the correction of some grammatical errors is converted to the task of masked character inference, which has plenty of training data and mature solutions. Experiments on datasets of NLPTEA-CGED demonstrate that our model outperforms other CGED models in many aspects.

Lossy compression will inevitably introduce image artifacts in the decoded image and degrade the image quality. In recent years, convolutional neural network (CNN) has been exploited for removing compression artifacts with great success. However, most existing CNN-based methods only utilize image's local prior without considering the global prior on the training of their networks. In this letter, a novel CNN, called the local and global priors network (LGPNet), is proposed that simultaneously learns both the local and the global priors for removing compression image artifacts. To achieve this goal, a dual-attention unit (DAU) is developed and incorporated into the well-known U-Net architecture for learning a better local prior. Meanwhile, the global prior is also learned from the entire image via our proposed global prior network. Extensive experimental results have clearly demonstrated that our proposed LGPNet is able to effectively remove image artifacts and greatly improve the image quality of JPEG-compressed images.

Lossy compression brings artifacts into the compressed image and degrades the visual quality. In recent years, many compression artifacts removal methods based on convolutional neural network (CNN) have been developed with great success. However, these methods usually train a model based on one specific value or a small range of quality factors. Obviously, if the test images quality factor does not match to the assumed value range, then degraded performance will be resulted. With this motivation and further consideration of practical usage, a highly robust compression artifacts removal network is proposed in this paper. Our proposed network is a single model approach that can be trained for handling a wide range of quality factors while consistently delivering superior or comparable image artifacts removal performance. To demonstrate, we focus on the JPEG compression with quality factors, ranging from 1 to 60. Note that a turnkey success of our proposed network lies in the novel utilization of the quantization tables as part of the training data. Furthermore, it has two branches in parallel-i.e., the restoration branch and the global branch. The former effectively removes the local artifacts, such as ringing artifacts removal. On the other hand, the latter extracts the global features of the entire image that provides highly instrumental image quality improvement, especially effective on dealing with the global artifacts, such as blocking, color shifting. Extensive experimental results performed on color and grayscale images have clearly demonstrated the effectiveness and efficacy of our proposed single-model approach on the removal of compression artifacts from the decoded image.

User's new requirements on a system are critical factors for driving software service evolution. New requirements usually arise when users are not satisfied with the existing system as often reflected in users' divergent behaviors that can be detected through comparison with the known behavior patterns. In this paper, we propose a methodology that applies Conditional Random Fields (CRF) as the mathematical foundation for inferring users' desires based on a peculiar form of observations and further exploring their new intentions that often imply new requirements. The potential new intentions detected by the CRF model will be verified and analyzed to elicit users' new requirements for the system. As a result, the system should evolve through modifications or acquiring new functionalities to satisfy the new requirements. An experiment on a research library system is conducted to demonstrate how the new intentions are detected using the CRF method and used to drive system evolution.

The capability to accurately recognize users' new intentions is critical for service evolution, because software systems are designed and implemented to assist their users. In sensor-rich context-aware environments, it is possible to recognize users' intentions from what users do (actions) and their interactions with and effects on the environment (environmental contexts). However, in order to establish the computational model and implement the intention recognition mechanism, it is essential that intentions and the critical situations used to define human intentions be clearly specified using observable contexts and events. In this paper, we present a set of definitions that enable systems to capture these situations and recognize users' intentions to support service evolution. An algorithm for intention recognition and the implementation of this algorithm to simulate the process of recognizing intentions at runtime are also presented. A case study of ATM cash withdraw is used to demonstrate the intention specification process.