innovation and emerging technologies

In industrial R&D, traditional technology scouting and solution discovery rely heavily on manual efforts, suffer from low efficiency, and face fragmented data sources (e.g., patent databases, product catalogs, competitive intelligence), leading to disjointed insights and delayed innovation. Method: This study introduces the first AI-driven intelligent platform for joint analysis of cross-domain patent and market data. It integrates large language models (LLMs) for semantic understanding, contextual reasoning, and cross-domain knowledge extraction, augmented by NLP techniques and clustering algorithms to parse unstructured patent text and integrate commercial intelligence. The platform features a novel co-evaluation framework for technical novelty and commercial feasibility, alongside a standardized technology taxonomy. Contribution/Results: Experiments demonstrate significant reductions in manual search effort and R&D cycle time, improved solution relevance and decision quality, and enhanced systemic innovation capability within complex technological ecosystems.

Current patent evaluation relies heavily on retrospective metrics or manual analysis, hindering efficient identification of high-value patents to support technology transfer. To address this, we propose a multi-stage hybrid intelligence framework. Methodologically, it integrates a “demand–seed” dual-agent system with a domain-specific core ontology, synergizing Learning-to-Rank (LTR), fine-tuned large language models (LLMs), patent semantic analysis, and market data mining—augmented by dynamic parameter weighting and human-in-the-loop validation. Our key contributions are: (1) the first integration of agent-based modeling and LTR for dynamic, context-aware patent-value matching; and (2) an interpretable, traceable, comprehensive assessment across 30+ legal and commercial dimensions. Empirical evaluation demonstrates substantial improvements in high-value patent identification accuracy, significantly enhancing both the strategic robustness and operational efficiency of technology transfer decision-making.

Chinese patent documents are linguistically complex and challenging to access and interpret, hindering innovation practices grounded in intellectual property among university faculty and students. To address this, we propose a novel system for creativity assessment and claim generation tailored to Chinese patents. Our method introduces the first domain-specific API integrating legal and technical semantics for Chinese patent texts; combines Chinese NLP-based parsing, domain-adaptive retrieval, structured claim generation, and a legal-knowledge-enhanced ranking model; and unifies novelty screening with automated claim drafting. The system significantly improves both accuracy and efficiency in novelty determination and enables rapid identification of technically improvable aspects. Empirical validation across multiple universities confirms its effectiveness in overcoming the longstanding bottlenecks of patent data—namely, difficulty in acquisition, comprehension, and application.

This study addresses the limitations of existing AI-assisted innovation systems, which often rely on a single methodology and struggle to integrate knowledge across diverse approaches or support traceable innovation reasoning, resulting in fragmented insights. To overcome this, the authors propose a multi-agent framework that synergistically combines TRIZ, design thinking, and SCAMPER methodologies through a persistent knowledge graph to co-generate and link patent claims. The core contributions include a graph-based cross-method convergence mechanism—formalized via CONVERGENT relationships—and an InnovationScore metric for comprehensive, quantitative assessment of claim quality. Experimental results in a legal technology context demonstrate that the approach significantly enhances the diversity and traceability of innovation candidates and enables the generation of structured patent drafts, outperforming single-method baselines.

Existing evaluation methods struggle to effectively assess the quality of patent specifications with respect to long-text coherence and legal compliance, particularly regarding enablement and written description requirements. This work proposes Pat-DEVAL, the first multidimensional evaluation framework tailored for patent specifications, which introduces a novel Chain-of-Legal-Thought (CoLT) mechanism that explicitly incorporates patent-law-specific constraints into the assessment process. By integrating legal-knowledge-guided reasoning and multidimensional scoring within an LLM-as-a-judge architecture, Pat-DEVAL establishes a new paradigm that balances technical rigor with legal adherence. Evaluated on the expert-annotated Pap2Pat-EvalGold dataset, the framework achieves a Pearson correlation coefficient of 0.69 overall and 0.73 on the legal compliance dimension, significantly outperforming existing baselines and general-purpose LLM evaluators.

This study addresses the challenge of early-stage technology opportunity identification, where ambiguous user needs and the absence of systematic integration of end-user values often lead to misalignment between technological potential and market demands. To bridge this gap, the authors propose a novel decision-support framework that integrates Technology Readiness Levels (TRL) with Schwartz’s theory of basic human values—introducing, for the first time, human values into the process of technology opportunity recognition. The framework defines two key metrics: “value breadth” and “vision gap.” Through qualitative analysis combining expert and consumer workshops in a case study at Sony CSL, the research demonstrates that successful technologies resonate across a broader spectrum of human values, and that experts articulate richer value dimensions than consumers. These findings validate the framework’s capacity to enhance technology–market fit through a value-driven approach.

Existing patent examination benchmarks are largely confined to static classification or information extraction, failing to capture the dynamic, iterative dialogue between examiners and applicants. This work proposes PatRe, the first interactive benchmark encompassing the full patent examination lifecycle, comprising 480 real-world cases. PatRe uniquely frames patent examination as a multi-turn argumentation and response task, enabling the generation of both examiner objections and applicant rebuttals. It introduces a dual evaluation paradigm grounded in authentic context and retrieval-augmented reasoning. Experiments reveal a substantial performance gap between closed- and open-source large language models in legal reasoning and novelty assessment, and uncover an asymmetry in task difficulty between the examiner and applicant perspectives, offering a new benchmark and critical insights for intelligent patent examination research.

Current approaches to patent novelty assessment often oversimplify the task into claim-level binary classification, which is susceptible to spurious correlations and lacks fine-grained analysis. This work reframes novelty evaluation as a feature-level retrieval and reasoning task by decomposing patent claims, aligning them with specific passages in prior art documents, and identifying the distinctive features that confer novelty. To support this paradigm, we introduce FiNE-Patents, the first dataset comprising 3,658 annotated claims with feature-level prior art references, and propose an interpretable evaluation framework based on large language models (LLMs). Experimental results demonstrate that our approach significantly outperforms embedding-based baselines in both passage retrieval and identification of novel features, while exhibiting greater robustness against spurious correlations in claim-level classification.