Semantics, knowledge and advanced cyber-infrastructure represent three communities of practice that have significantly influenced the development of computing technologies and systems. The rapid development of Artificial Intelligence and Data Science challenges the existing research on semantics, knowledge, and cyber-infrastructure. Related research concerns the following topics: (1) semantics, specifically semantics modeling and semantic link networks including construction, discovery, analysis and management; (2) knowledge, specifically knowledge discovery in big data, methods for transforming big data into knowledge, Knowledge modeling and management, Knowledge Graph, and advanced computing infrastructures; (3) advanced cyber-infrastructure, specifically new data models, software tools and methodologies as well as advanced computing infrastructures for big data management; and, (4) applications based on semantics, knowledge and advanced computing infrastructure, including smart cities, healthcare, industries and advanced information services such as question answering, recommendation and summarization on multimedia integrating texts, pictures, and videos.

This special issue is to report research development of the above areas, promote cross-area research and accelerate cross-area research. It includes four papers selected from open submissions and the selected papers presented at the 15th International Conference on Semantics, Knowledge, and Grids. Conference papers were significantly extended and reviewed as normal submissions. All accepted papers were published online first.

The first paper introduces a link prediction algorithm that predicts relations between links based on network structure. It uses common neighbors and degree distribution of nodes to estimate the possibility of the presence of a link between nodes within network.¹ It is an interesting research issue if research on link prediction can be extended to semantic link prediction on semantic link network.

The second paper introduces a neural-based method for generating captions for images on news automatically. A multimodal pointer mechanism is proposed by using both textual attention and visual attention to compute pointer distributions. A multimodal coverage mechanism is defined to reduce repetitions of attentions or repetitions of pointer distributions. The proposed model outperforms the original pointer-generator network, the generic image captioning method, the extractive news image captioning method, and the LDA-based method.² As other deep learning methods, the proposed method faces some challenges including interpretability.

The third paper surveys two branches of research for combining graphics, uncertainty and semantics to develop a stronger modeling method: one is to combine graphics and uncertainty as probabilistic graphical models and then incorporate semantics, and the other is to combine graphics and semantics and then incorporate probability to handle uncertainty. The models and methods involved in these efforts including their expressiveness, pros, and cons are discussed.³ This survey helps establish links between different research directions.

The fourth paper surveys the state-of-the-art chatbots for psychotherapy including a series of tasks necessary for future research. Techniques include retrieval-based methods and cognitive behavior therapy. The assessments show that chatbots can preliminarily recognize specific kinds of negative emotions and give relatively appropriate responses. The randomized controlled trials show that psychotherapy chatbots are useful for some people with mental health problem but more effective and safer psychotherapy chatbots are needed.⁴ The techniques for developing Chatbots can be used to develop intelligent agents in future cyber-physical-social environment.

We have witnessed a great development of online communities and requirements of advanced information technologies during the unprecedented pandemic. One of the insightful visions is the cyber-physical society that integrate cyberspace that follows cyber principles, physical space that follows physical principles, and social space that follows socioeconomic principles.⁵ The development of digital twins in various domains provides the experience and technical basis for implementing the vision.

A new research paradigm on science and engineering is emerging with the development of the fundamental research on semantics, knowledge and cyberinfrastructure that underpins various advanced intelligent applications.⁶

语义、知识和先进的网络基础设施代表了对计算技术和系统的发展产生重大影响的三个实践社区。人工智能和数据科学的快速发展对语义、知识和网络基础设施的现有研究提出了挑战。相关研究涉及以下主题：（1）语义，特别是语义建模和语义链接网络，包括构建、发现、分析和管理； （2）知识，特别是大数据中的知识发现、大数据转化为知识的方法、知识建模与管理、知识图谱和先进计算基础设施； （3）先进的网络基础设施，特别是新的数据模型、软件工具和方法以及用于大数据管理的先进计算基础设施； （四）基于语义、知识和先进计算基础设施的应用，包括智慧城市、医疗、行业以及文本、图片、视频相结合的多媒体问答、推荐、摘要等先进信息服务。

本期特刊旨在报道上述领域的研究进展，推动跨领域研究，加速跨领域研究。它包括从公开提交的论文中选出的四篇论文以及在第 15 届语义、知识和网格国际会议上发表的精选论文。会议论文与正常提交的论文一样得到了显着扩展和审查。所有被录用的论文均首先在线发表。

第一篇论文介绍了一种基于网络结构预测链接之间关系的链接预测算法。它使用公共邻居和节点的度分布来估计网络内节点之间存在链接的可能性。¹如果链接预测的研究能够扩展到语义链接网络上的语义链接预测，这是一个有趣的研究问题。

第二篇论文介绍了一种基于神经网络的方法，用于自动生成新闻图像的标题。通过使用文本注意力和视觉注意力来计算指针分布，提出了多模式指针机制。定义多模态覆盖机制来减少注意力的重复或指针分配的重复。所提出的模型优于原始指针生成器网络、通用图像字幕方法、提取新闻图像字幕方法和基于 LDA 的方法。²与其他深度学习方法一样，所提出的方法面临着一些挑战，包括可解释性。

第三篇论文概述了将图形、不确定性和语义结合起来以开发更强的建模方法的两个研究分支：一是将图形和不确定性结合为概率图模型，然后结合语义，二是将图形和语义结合，然后结合语义。处理不确定性的概率。讨论了这些工作中涉及的模型和方法，包括它们的表现力、优点和缺点。³这项调查有助于建立不同研究方向之间的联系。

第四篇论文调查了用于心理治疗的最先进的聊天机器人，包括未来研究所需的一系列任务。技术包括基于检索的方法和认知行为疗法。评估表明，聊天机器人可以初步识别特定类型的负面情绪并给出相对合适的反应。随机对照试验表明，心理治疗聊天机器人对一些有心理健康问题的人有用，但需要更有效、更安全的心理治疗聊天机器人。⁴开发聊天机器人的技术可用于开发未来网络物理社会环境中的智能代理。

在这场前所未有的疫情期间，我们目睹了网络社区的巨大发展和对先进信息技术的需求。其中一个富有洞察力的愿景是网络物理社会，将遵循网络原则的网络空间、遵循物理原则的物理空间和遵循社会经济原则的社会空间整合在一起。⁵数字孪生在各领域的发展为实现愿景提供了经验和技术基础。

随着支撑各种先进智能应用的语义、知识和网络基础设施基础研究的发展，一种新的科学和工程研究范式正在出现。⁶