atlassian confluence

Current AI writing tools are predominantly designed for single users, hindering seamless integration into collaborative authoring workflows and resulting in contextual fragmentation and high integration overhead. This paper proposes deep integration of AI agents into collaborative writing environments—not as autonomous participants, but as shared, controllable resources. We introduce two shared state abstractions—“agent profiles” and “task objects”—to render AI behavior transparent, customizable, and aligned with team norms. Our approach combines collaborative editing system design, shared state management, and role-aware agent modeling, augmented by qualitative user studies. AI responses are embedded as inline comments to ensure frictionless interaction. A one-week field study with 14 teams demonstrates that AI-generated outputs become reusable shared assets, while teams retain full creative agency and control throughout the process. This work constitutes the first systematic redefinition of human-AI responsibility boundaries and technical interfaces in collaborative writing.

In the context of software engineering’s evolution toward SE 3.0, this study presents the first large-scale empirical investigation into the real-world impact of AI agents on documentation writing and their collaboration patterns with human developers. Leveraging the AIDev dataset, we analyze 1,997 documentation-related pull requests (PRs) through authorship attribution, PR categorization, and manual review behavior analysis. Our findings reveal that AI-generated documentation PRs significantly outnumber those authored by humans and are frequently merged without substantive modifications. This highlights a notable lack of scrutiny in current documentation review practices toward AI-generated content, exposing emerging challenges in ensuring documentation quality within human-AI collaborative workflows. The results offer critical insights for developing reliable and governable AI-assisted software development practices.

Official API documentation often fails to meet developers’ needs due to being outdated and incomplete. This work proposes an automated approach that, for the first time, integrates fine-grained API knowledge extraction, dense retrieval, and large language model–based summarization to generate structured documentation from community content such as Stack Overflow. By fine-tuning a dense retrieval model to identify seven categories of API knowledge and incorporating hallucination mitigation and redundancy reduction mechanisms, the method significantly enhances generation quality. Experimental results show that it improves accuracy by up to 77.7% over baseline methods, reduces redundant content by 9.5%, and recovers 34.4% of critical knowledge missing from official documentation. User studies further confirm its substantial advantages in comprehensiveness, conciseness, and practical utility.

Real-world API documentation is often unstructured and highly heterogeneous across tools, leading to high development costs and poor generalization in agent construction. Method: This paper proposes an end-to-end, scalable tool generation framework: (1) parsing raw API documentation to automatically extract interface semantics and parameter constraints; (2) generating executable Python tool functions; and (3) incorporating a code-agent-driven iterative feedback optimization mechanism to improve tool invocation accuracy. Contribution/Results: To our knowledge, this is the first approach achieving high validation rates (>92%) in fully automated tool generation directly from real-world API documentation—without human annotation or domain-specific adaptation. On the WebArena benchmark, it improves task success rate by 55% and reduces tool construction cost by 90%. Furthermore, it demonstrates strong cross-domain generalization, validated in complex vertical domains such as sugar materials science.

To address insufficient content controllability when compressing lengthy technical documents (e.g., research papers) into concise formats (e.g., blog posts), this paper proposes an interactive reverse-source outline mechanism. It explicitly models LLM-based summarization as an editable, traceable, structured outline—comprising hierarchical, semantically grounded nodes—enabling iterative user refinement of functional points to precisely govern information coverage. The method integrates large language models, dynamic content-to-outline mapping, and an interactive UI to realize an end-to-end, fine-grained summarization system. Empirical evaluation and real-world deployment demonstrate significant improvements: author satisfaction with content coverage increases markedly; information change per editing operation rises by 37%; and retention of critical research insights improves 2.1×. This work achieves, for the first time, bidirectional controllability—both selection and synthesis—over content in technical document summarization.

Current document and table modeling lacks a unified standard, with existing approaches typically modeling spatial or semantic structures in isolation—leading to format incompatibility and poor cross-task generalization. To address this, we propose S2Doc: a standardized document data model that jointly encodes spatial layout and semantic hierarchy. S2Doc employs a hierarchical structure that integrates coordinate information with semantic labels, enabling unified, multi-page document representation while natively supporting interoperable formats such as JSON. It is the first model to concurrently represent both spatial and semantic dimensions within a single, standardized structure—thereby filling a critical gap in document modeling standards. Empirical evaluation demonstrates that S2Doc significantly improves cross-task compatibility, model interoperability, and data exchange efficiency. It has been validated across core document understanding tasks, including OCR, information extraction, and table recognition.

Traditional document-centric structures hinder the structuring, updating, and reuse of knowledge, while existing formal methods struggle to gain widespread adoption due to their neglect of human–machine usability. This work proposes the MMM data model, which combines lightweight normative constraints with free-text tags to enable cross-disciplinary and cross-application knowledge interoperability without enforcing rigid semantic uniformity. By introducing a decentralized knowledge architecture that preserves expressive freedom, the model balances human readability with system interoperability. A reference implementation and pilot deployments across multiple disciplines demonstrate the approach’s feasibility and preliminary effectiveness, offering a novel paradigm for building interoperable, decentralized knowledge communities.

This study addresses the risk of information loss and disruption to the scholarly record posed by the absence of long-term preservation mechanisms for academic blogs, an emerging form of scholarly communication. Employing a convergent mixed-methods approach, the research integrates quantitative analysis of 866 German-language academic blogs, in-depth interviews with 13 bloggers, and open community-based participatory review to systematically synthesize multi-source evidence for the first time. The work proposes a comprehensive set of digital preservation requirements specifically tailored to academic blogs and develops an actionable implementation guide for library practitioners. This framework offers an innovative solution to effectively integrate academic blogs into scholarly information infrastructures, ensuring their long-term accessibility, reusability, and citability.

This work addresses the challenges of maintaining documentation in large codebases—namely, the lack of semantic structure in existing tool-generated content and difficulties in tracking changes—by proposing Repository Knowledge Graphs (RepoKG). RepoKG introduces a three-stage pipeline comprising code entity relation extraction, functional module clustering, and agent-driven documentation generation, establishing knowledge graphs as the semantic foundation for the entire documentation lifecycle. It incorporates modular hierarchical organization and a bidirectional semantic influence propagation mechanism to enable structured, cross-referable documentation with efficient incremental updates. Evaluated across 24 multilingual repositories, RepoKG improves API coverage by 32.5% and completeness by 10.4%, while accelerating generation by 3× and reducing token consumption by 85%. For incremental updates, it cuts update time by 73%, lowers token usage by 77%, and increases update recall by 10.2%.

Scientific posters, as vital vehicles for scholarly communication, have long suffered from low sharing rates, the absence of persistent identifiers, incomplete metadata, and a lack of established citation practices, all of which hinder their discoverability and reuse. This study presents the first large-scale empirical analysis of 86 global poster-sharing platforms, integrating platform surveys, metadata assessments, and usage metrics—including views, downloads, and citations—with a focus on major repositories such as Zenodo and Figshare. As of 2024, only approximately 150,000 posters are publicly shared, with most platforms failing to adhere to FAIR principles, exhibiting critical gaps in conference-related metadata and extremely low citation rates. The findings reveal a disconnect between platform capabilities and user behaviors, prompting the proposal of community guidelines to standardize poster sharing and reuse.