Artificial Intelligence Review最新文献

The rapid advancement of AI technology has led to increasingly complex and opaque systems, creating a critical need for explainable AI (XAI) that enhances transparency and user trust. Despite extensive research on XAI methods and applications, the user-centered design (UCD) process for XAI remains fragmented and unclear. This systematic review analyzes 27 studies from 2020 to 2024 to develop a comprehensive framework for user-centered XAI system design (UCXAISD). We identify five key stages: contextual inquiry, explanation needs identification, XAI method selection, user interface design, and evaluation and refinement. Our framework aligns with traditional UCD processes while incorporating specialized elements for XAI, including user-centricity, transparency, and actionability. For each stage, we provide evidence-based guidelines derived from the literature. The framework serves as a blueprint for developing XAI systems that balance technical sophistication with user needs.

This systematic literature review (SLR) integrates Geographic Information Systems (GIS), deep learning, and Multi-Criteria Decision Making (MCDM) to enhance road route optimization, crucial for global infrastructure development. This SLR aims to identify existing research trends, methodologies, research gaps and propose a generalized framework for streamlining the road route optimization process. The review addresses three key research questions: RQ-1. The application of deep learning for Land Use and Land Cover (LULC) classification, RQ-2. The use of MCDM techniques in road route alignment and RQ-3. Techniques for optimizing road route alignment. Utilizing PRISMA, we assessed 370 papers, selected 132 through full-text evaluation, and added 25 via. snowball sampling, totalling 157 records for analysis. The results reveal trends in current research, geographical distribution and the evolution of methodologies. It is found that Deep learning techniques significantly improve LULC classification accuracy, while MCDM techniques enable a holistic approach to road route alignment by incorporating diverse factors. The proposed generalized framework outlines a systematic approach encompassing problem definition, criteria selection, data preparation, deep learning-based LULC classification, MCDM and Least Cost Path analysis for road route alignment. This work uniquely identifies research trends, methodologies, and gaps in road route optimization, addressing three specific research questions (RQ-1 to RQ-3) on deep learning (LULC classification), MCDM techniques, and route alignment optimization. This work also highlights the scope for integrating emerging technologies, enhancing MCDM approaches, promoting cross-disciplinary collaboration, addressing data availability and quality, conducting case studies, emphasizing sustainability, resilience and focusing on global and regional contexts. This SLR will surely contribute to the development of efficient, sustainable and equitable road route optimization strategies for better infrastructure planning and worldwide development.

Given the large size and volumes of contracts and their underlying inherent complexity, manual reviews become inefficient and prone to errors, creating a clear need for automation. Automatic Legal Contract Classification (LCC) revolutionizes the way legal contracts are analyzed, offering substantial improvements in speed, accuracy, and accessibility. This survey delves into the challenges of automatic LCC and a detailed examination of key tasks, datasets, and methodologies. We identify seven classification tasks within LCC, and review fourteen datasets related to English-language contracts, including public, proprietary, and non-public sources. We also introduce a methodology taxonomy for LCC, categorized into Traditional Machine Learning, Deep Learning, and Transformer-based approaches. Additionally, the survey discusses evaluation techniques and highlights the best-performing results from the reviewed studies. By providing a thorough overview of current methods and their limitations, this survey suggests future research directions to improve the efficiency, accuracy, and scalability of LCC. As the first comprehensive survey on LCC, it aims to support legal NLP researchers and practitioners in improving legal processes, making legal information more accessible, and promoting a more informed and equitable society.

Human action recognition (HAR) encompasses the task of monitoring human activities across various domains, including but not limited to medical, educational, entertainment, visual surveillance, video retrieval, and the identification of anomalous activities. Over the past decade, the field of HAR has witnessed substantial progress by leveraging convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to effectively extract and comprehend intricate information, thereby enhancing the overall performance of HAR systems. Recently, the domain of computer vision has witnessed the emergence of Vision Transformers (ViTs) as a potent solution. The efficacy of Transformer architecture has been validated beyond the confines of image analysis, extending their applicability to diverse video-related tasks. Notably, within this landscape, the research community has shown keen interest in HAR, acknowledging its manifold utility and widespread adoption across various domains. However, HAR remains a challenging task due to variations in human motion, occlusions, viewpoint differences, background clutter, and the need for efficient spatio-temporal feature extraction. Additionally, the trade-off between computational efficiency and recognition accuracy remains a significant obstacle, particularly with the adoption of deep learning models requiring extensive training data and resources. This article aims to present an encompassing survey that focuses on CNNs and the evolution of RNNs to ViTs given their importance in the domain of HAR. By conducting a thorough examination of existing literature and exploring emerging trends, this study undertakes a critical analysis and synthesis of the accumulated knowledge in this field. Additionally, it investigates the ongoing efforts to develop hybrid approaches. Following this direction, this article presents a novel hybrid model that seeks to integrate the inherent strengths of CNNs and ViTs.

Deep learning, driven by big data and graphic processing units, has garnered significant attention across various domains. The flexibility of network architectures, combined with their diverse components, has allowed deep learning techniques to be applied across a wide range of domains, expanding from low- and high-level computer vision tasks to encompass video processing, natural language processing (NLP), and 3D data processing. However, there has been relatively little effort to systematically summarise these works from principles to applications in terms of deep learning fundamentals. The present study aims to address this gap in the literature by presenting components of deep networks for image applications, and describing several classical deep networks for image applications. The study then introduces principles, relations, ranges, and applications of deep networks across an expanded scope, covering low-level vision tasks, high-level vision tasks, video processing, NLP, and 3D data processing. The study then compares the performance of different networks across these diverse tasks. Finally, it summarises potential focuses and challenges of deep learning research for these applications with concluding remarks.

In the burgeoning field of Large Language Models (LLMs), developing a robust safety mechanism, colloquially known as “safeguards” or “guardrails”, has become imperative to ensure the ethical use of LLMs within prescribed boundaries. This article provides a systematic literature review on the current status of this critical mechanism. It discusses its major challenges and how it can be enhanced into a comprehensive mechanism dealing with ethical issues in various contexts. First, the paper elucidates the current landscape of safeguarding mechanisms that major LLM service providers and the open-source community employ. This is followed by the techniques to evaluate, analyze, and enhance some (un)desirable properties that a guardrail might want to enforce, such as hallucinations, fairness, privacy, and so on. Based on them, we review techniques to circumvent these controls (i.e., attacks), to defend the attacks, and to reinforce the guardrails. While the techniques mentioned above represent the current status and the active research trends, we also discuss several challenges that cannot be easily dealt with by the methods and present our vision on how to implement a comprehensive guardrail through the full consideration of multi-disciplinary approach, neural-symbolic method, and systems development lifecycle.

There is a strong correlation between linguistics and artificial intelligence (AI), best manifested by deep learning language models. This study provides a thorough scientometric analysis of this correlation, synthesizing the intellectual production over 51 years, from 1974 to 2024. Web of Science Core Collection (WoSCC) database was the data source. The data collected were analyzed by two powerful software, viz., CiteSpace and VOSviewer, through which mapping visualizations of the intellectual landscape, trending issues and (re)emerging hotspots were generated. The results indicate that in the 1980s and 1990s, linguistics and AI (AIL) research was not robust, characterized by unstable publication over time. It has, however, witnessed a remarkable increase of publication since then, reaching 1478 articles in 2023, and 546 articles in January-March timespan in 2024, involving emerging issues including Natural language processing, Cross-sectional study, Using bidirectional encoder representation, and Using ChatGPT and hotspots such as Novice programmer, Prioritization, and Artificial intelligence, addressing new horizons, new topics, and launching new applications and powerful deep learning language models including ChatGPT. It concludes that linguistics and AI correlation is established at several levels, research centers, journals, and countries shaping AIL knowledge production and reshaping its future frontiers.

This study presents a systematic literature review on the explainability and interpretability of machine learning models within the context of predictive process monitoring. Given the rapid advancement and increasing opacity of artificial intelligence systems, understanding the "black-box" nature of these technologies has become critical, particularly for models trained on complex operational and business process data. Using the PRISMA framework, this review systematically analyzes and synthesizes the literature of the past decade, including recent and forthcoming works from 2025, to provide a timely and comprehensive overview of the field. We differentiate between intrinsically interpretable models and more complex systems that require post-hoc explanation techniques, offering a structured panorama of current methodologies and their real-world applications. Through this rigorous bibliographic analysis, our research provides a detailed synthesis of the state of explainability in predictive process mining, identifying key trends, persistent challenges and a clear agenda for future research. Ultimately, our findings aim to equip researchers and practitioners with a deeper understanding of how to develop and implement more trustworthy, transparent and effective intelligent systems for predictive process analytics.

Stroke remains a significant global health concern, necessitating accurate and timely prognosis to optimize patient care and outcomes. In recent years, deep learning, a subset of artificial intelligence, has emerged as a promising tool for enhancing stroke prognosis by leveraging its capability to analyze complex clinical and imaging data. This advancement has sparked a significant increase in research publications in this domain. Therefore, our objective in this systematic literature review (SLR) is to: systematically review and analyze the existing body of literature to identify key deep learning architectures, evaluate their performance against conventional prognosis methods, explore the range of clinical and neuroimaging data sources employed, and investigate the potential impact of deep learning on personalized stroke management. Our findings reveal that deep learning holds considerable promise in improving stroke prognosis accuracy, offering opportunities for more precise clinical decision-making. However, challenges related to data heterogeneity, interpretability, and clinical integration persist. We discuss these challenges and propose future directions to facilitate the successful integration of deep learning into routine stroke care. As the demand for precise stroke prognosis intensifies, this review serves as a valuable resource for researchers, clinicians, and policymakers alike, offering insights into the current state of deep learning applications in stroke prognosis and guiding efforts toward leveraging artificial intelligence to alleviate the burden of stroke on individuals and healthcare systems.

Composting is a complex biological process, and due to the numerous variables affecting its course, it requires constant supervision and, depending on the needs, appropriate modifications. In particular, it is necessary to strive to ensure the quality of substrates, the elimination of possible contaminants, the efficient and inexpensive conduct of the process, and the fulfillment by the finished compost of the quality requirements allowing its use as a fertilizer or crop improvement agent. Therefore, new effective methods for composting optimization are needed. This paper reviews the state of the art on the use of artificial neural networks (ANN) in bio-waste composting with a special focus on applying machine learning tools. Artificial neural networks were characterized along with their division into different types, the basics of the composting process and legal requirements for bio-waste recycling were described. Different types of machine learning were compared with attention paid to the effectiveness of the tools used. Also, for further studies, the appropriate independent variables were proposed to be used in ANN designing. The presented examples of the application of ANN confirm the usefulness of this method, to solve the complexity of the composting issue, and the need for further research.