计算机科学 ›› 2026, Vol. 53 ›› Issue (4): 377-383.doi: 10.11896/jsjkx.250600032
吴荞睿1, 罗丽2, 赵才荣1
WU Qiaorui1, LUO Li2, ZHAO Cairong1
摘要： 针对专利摘要生成任务中的技术特征完整性缺失、合规性不足以及边缘部署效率受限等问题,提出一种基于循环一致性约束的大语言模型增强型语言模型训练框架(LACC)。LACC框架通过构建摘要生成与权利要求扩写的双向可逆任务结构,引入循环一致性约束机制,实现技术特征与法律表述的联合优化。在此基础上,LACC集成了由大语言模型驱动的可控数据增强策略,自动构建高质量专利样本对,并结合动态验证机制,提升了生成文本的技术准确性与合规稳定性。实验结果表明,LACC在中文专利数据集CPTD上的ROUGE-L得分达到56.74,较基线提升8.99个百分点,在循环一致性评分(RCS)方面亦取得显著优势。此外,该方法在边缘设备上的推理延迟控制在420 ms以内,单卡内存占用不超过4.5 GB,具备良好的工业部署潜力。综上所述,LACC框架为专利撰写辅助、法律文本生成与知识产权智能管理等下游任务提供了具备实用价值的技术路径,展现出推动知识产权全生命周期自动化处理的重要应用前景。
中图分类号:
| [1]SEMERIKOV S O,VAKALIUK T A,KANEVSKA O B,et al.LLM on the edge:the new frontier[C]//Proceedings of the 5th Edge Computing Workshop(doors 2025),2025:137-161. [2]FRIHA O,FERRAG M A,KANTARCI B,et al.LLM-basededge intelligence:A comprehensive survey on architectures,applications,security and trustworthiness[J].IEEE Open Journal of the Communications Society,2024,5:5799-5856. [3]LASKAR M T R,ALQAHTANI S,BARI M S,et al.A systematic survey and critical review on evaluating large language mo-dels:Challenges,limitations,and recommendations[C]//Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing.2024:13785-13816. [4]ZHU J Y,PARK T,ISOLA P,et al.Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks[C]//2017 IEEE International Conference on Computer Vision(ICCV).IEEE,2017:2242-2251. [5]KIM B,KIM J,LEE J G,et al.Unsupervised Deformable Image Registration Using Cycle-Consistent CNN[C]//Medical Image Computing and Computer Assisted Intervention-MICCAI 2019.2019. [6]SHEN T X,LEI T,BARZILAY R,et al.Style Transfer fromNon-Parallel Text by Cross-Alignment[C]//31st Conference on Neural Information Processing Systems(NIPS 2017).2017. [7]LIAN C,LI X,KONG L,et al.CoCycleReg:Collaborative Cycle-Consistency Method for Multi-Modal Medical Image Re-gistration[J].Neurocomputing,2022,500:799-808. [8]LI C Y,LI X M,KONG L K,et al.CoCycleReg:Collaborative Cycle-Consistency Method for Multi-Modal Medical Image Re-gistration[J].Neurocomputing,2022,500:799-808. [9]MA C,ZHANG W E,GUO M Y,et al.Multi-document Summarization Via Deep Learning Techniques:A Survey[J].ACM Computing Surveys,2023,55(5):1-37. [10]WANG Z,WANG P,LIU K,et al.A comprehensive survey on data augmentation[J].arXiv:2405.09591,2024. [11]ZHOU Y,GUO C,WANG X,et al.A survey on data augmentation in large model era[J].arXiv:2401.15422,2024. [12]SCHICK T,SCHÜTZE H.Few-Shot Text Generation with Pattern-Exploiting Training[J].arXiv:2012.11926,2020. [13]WEI J,ZOU K.EDA:Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Confe-rence on Natural Language Processing.2019:6382-6388. [14]DING B,QIN C,ZHAO R,et al.Data augmentation usingLLMs:Data perspectives,learning paradigms and challenges[C]//Findings of the Association for Computational Linguistics ACL 2024.2024:1679-1705. [15]ZELIKMAN E,WU Y H,MU J,et al.STaR:BootstrappingReasoning with Reasoning[C]//NEURIPS 2022.2022. [16]LOUIS D,VAN H,JAAP S,et al.Robust Deformable ImageRegistration Using Cycle-Consistent Implicit Representations[J].IEEE Transactions on Medical Imaging,2024,43(2):784-793. [17]RAMAKANTH P,ASLI C,MICHEL G,et al.Data Augmentation for Abstractive Query-Focused Multi-Document Summarization[J].Computing Research Repository,2021,35(15):13666-13674. [18]GOU J,YU B,MAYBANK S J,et al.Knowledge distillation:A survey[J].International Journal of Computer Vision,2021,129(6):1789-1819. [19]GEOFFREY H,ORIOL V,JEFF D.Distilling the Knowledge in a Neural Network[J].arXiv:1503.02531,2015. [20]KUMAR S,ALESSANDRO S,MRINMAYA S.Distilling Reasoning Capabilities into Smaller Language Models[J].arXiv:2212.00193,2022. [21]LIU Z Y,SOUMYA S,ISABELLE L,et al.Self-Contradictory Reasoning Evaluation and Detection[J].Association for Computational Linguistics,2024(11):3725-3742. [22]XU X,LI M,TAO C,et al.A survey on knowledge distillation of large language models[J].arXiv:2402.13116,2024. [23]ZHANG Y,KANG B,HOOI B,et al.Deep long-tailed learning:A survey[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(9):10795-10816. |
| [1] | 宋建华, 刘淳, 张龑. 基于结构化知识蒸馏的轻量级伪装目标检测模型 Lightweight Camouflaged Object Detection Model Based on Structured Knowledge Distillation 计算机科学, 2026, 53(4): 299-307. https://doi.org/10.11896/jsjkx.250100105 |
| [2] | 胡俊杰, 陈宇杰, 胡义坤, 文成, 曹嘉伦, 马智, 苏杰, 孙纬地, 田聪, 秦胜潮. 大模型驱动的形式化定理证明:综述与展望 Formal Theorem Proving Empowered by Large Language Model:Survey and Perspectives 计算机科学, 2026, 53(4): 1-23. https://doi.org/10.11896/jsjkx.251000067 |
| [3] | 刘苏熠, 刘淇, 高维博. Agent4Stu:基于大语言模型的学生作答行为高效模拟智能体 Agent4Stu:Efficient LLM-based Student Answer Behavior Simulation Agent 计算机科学, 2026, 53(4): 347-355. https://doi.org/10.11896/jsjkx.250800012 |
| [4] | 潘嘉豪, 冯翔, 虞慧群. 基于多任务强化学习的优先级加权软模块化方法:SM-PHT SM-PHT:Robust,Scalable,and Efficient Method for Multi-task Reinforcement Learning 计算机科学, 2026, 53(4): 366-376. https://doi.org/10.11896/jsjkx.250700198 |
| [5] | 王智彬, 李世鹏, 周宇航, 李雪, 张中辉, 蒋智威, 顾荣, 田臣, 陈贵海, 仲盛. 大语言模型服务系统服务级目标和系统级指标优化 Optimization of Service Level Objectives and System Level Metrics in Large Language ModelServing System 计算机科学, 2026, 53(3): 23-32. https://doi.org/10.11896/jsjkx.250900173 |
| [6] | 周悦媛, 卢冠泽, 向佳为, 章家维, 邵恩, 何鑫. 面向海光DCU基于自适应转置的大语言模型训练系统 Training System for Large Language Models Based on Adaptive Transpose on Hygon DCU 计算机科学, 2026, 53(3): 33-40. https://doi.org/10.11896/jsjkx.250600073 |
| [7] | 陈涵, 徐泽锋, 蒋究, 樊凡, 章军建, 何楚, 王文伟. 基于大语言模型和深度网络的认知评估量表自动诊断 Large Language Model and Deep Network Based Cognitive Assessment Automatic Diagnosis 计算机科学, 2026, 53(3): 41-51. https://doi.org/10.11896/jsjkx.250600034 |
| [8] | 钱清, 陈辉程, 崔允贺, 唐瑞雪, 付金玫. 多尺度聚合协同轴向语义引导的实体关系联合抽取方法 Joint Entity and Relation Extraction Method with Multi-scale Collaborative Aggregation and Axial-semantic Guidance 计算机科学, 2026, 53(3): 97-106. https://doi.org/10.11896/jsjkx.250500095 |
| [9] | 葛泽庆, 黄圣君. 针对多标记表格数据的半监督学习方法 Semi-supervised Learning Method for Multi-label Tabular Data 计算机科学, 2026, 53(3): 151-157. https://doi.org/10.11896/jsjkx.250600149 |
| [10] | 吴贤杰, 李彤亮, 李舟军. 表格问答研究综述 Survey of Table Question Answering Research 计算机科学, 2026, 53(3): 295-306. https://doi.org/10.11896/jsjkx.250900006 |
| [11] | 徐成, 刘宇轩, 王欣, 张铖, 姚登峰, 袁家政. 大语言模型驱动的言语障碍评估方法综述 Review of Speech Disorder Assessment Methods Driven by Large Language Models 计算机科学, 2026, 53(3): 307-320. https://doi.org/10.11896/jsjkx.250300125 |
| [12] | 陈昱胤, 李贯峰, 秦晶, 肖毓航. 知识图谱的复杂逻辑查询方法研究综述 Survey on Complex Logical Query Methods in Knowledge Graphs 计算机科学, 2026, 53(2): 273-288. https://doi.org/10.11896/jsjkx.250400033 |
| [13] | 孙明旭, 梁刚, 吴逸飞, 胡海馨. 融合仇恨对象特征与变体词还原机制的中文仇恨言论检测 Chinese Hate Speech Detection Incorporating Hate Object Features and Variant Word Restoration Mechanism 计算机科学, 2026, 53(2): 289-299. https://doi.org/10.11896/jsjkx.241200004 |
| [14] | 郭陆祥, 王越余, 李芊玥, 李莎莎, 刘晓东, 纪斌, 余杰. 大语言模型智能体操作系统研究综述 Comprehensive Survey of LLM-based Agent Operating Systems 计算机科学, 2026, 53(1): 1-11. https://doi.org/10.11896/jsjkx.250500002 |
| [15] | 刘利龙, 刘国明, 齐保元, 邓雪杉, 薛迪展, 钱胜胜. 实际应用场景中的大模型高效推理技术综述 Efficient Inference Techniques of Large Models in Real-world Applications:A Comprehensive Survey 计算机科学, 2026, 53(1): 12-28. https://doi.org/10.11896/jsjkx.250300030 |
|
||
首页