文本蕴含关系识别与知识获取研究进展及展望

发布时间：2018-03-22 21:14

本文选题：文本蕴含　切入点：知识获取　出处：《计算机学报》2017年04期 　论文类型：期刊论文

【摘要】：文本蕴含关系是广泛分布于自然语言文本中的单向推理关系,文本蕴含相关研究是自然语言处理领域的一项基础性研究,它可以辅助其他自然语言处理任务的进行,并且具有丰富的应用场景.文中首先界定了文本蕴含研究的范畴.作为一种二元关系,文本蕴含有3个基本研究任务——关系识别、知识获取和蕴含对生成.其中,关系识别有两个核心问题——语义表示与推理机制;知识获取也有两个核心问题——知识表示与知识来源;蕴含对生成研究进展比较缓慢,文中细致地分析了其内因和外因.文中围绕语义表示与推理机制这两个核心问题梳理了关系识别的研究进展,围绕知识表示与知识来源梳理了知识获取的研究进展,并指出了各类方法的可取之处与不足之处.文本蕴含研究的进展离不开相关国际评测,文中也对这些国际评测和数据集进行了归纳总结.大数据时代的到来和深度学习理论的不断发展,为文本蕴含相关研究提供了丰富的知识来源和有力的研究工具,同时也带来了许多崭新的研究课题.文中立足当前研究形势,展望了未来研究方向,并从理论上探讨了其可行性.
[Abstract]:The text contains the relationship is one-way relationship reasoning is widely distributed in the natural language text of the text contains related research is a fundamental research field of Natural Language Processing, it can help the other Natural Language Processing tasks, and has rich application scenarios. This paper first defines the text contains the research category. As a two yuan, the text contains 3 basic research tasks -- the relationship between knowledge acquisition and recognition, the implication of generation. Among them, the relationship between identification of two core issues: semantic representation and reasoning mechanism; knowledge acquisition has two core issues: knowledge representation and knowledge sources; the implication of generation research progress is relatively slow, the paper analyzes the internal and external factors. In this paper the semantic representation and inference mechanism of the two core issues reviews the research progress of the relationship between recognition, based on the knowledge representation With the knowledge sources, reviews the knowledge acquisition, and points out the merits and shortcomings of various methods. The research progress of the text contains cannot do without the relevant international evaluation, the international evaluation and these data sets are summarized. The big data era and deep learning theory of continuous development, provide a rich source of knowledge and a powerful tool for studying the text contains related research, but also brought many new research topic. In this paper, based on the current research situation and prospects of future research, and discusses its feasibility.

【作者单位】：哈尔滨工业大学计算机科学与技术学院社会计算与信息检索研究中心;
【基金】：国家自然科学基金(61472105,61472107) 国家“八六三”高技术研究发展计划项目基金(2015AA015407)资助~~
【分类号】：TP391.1

【相似文献】