融合关键信息的科技文献创新点生成方法  

Method of generating scientific and technological literature innovation points combining key information

作　　者：李宝安 刘翔 王宗辉 吕学强[1] 游新冬 LI Bao-an;LIU Xiang;WANG Zong-hui;LYU Xue-qiang;YOU Xin-dong(Beijing Key Laboratory of Internet Culture and Digital Dissemination Research,Beijing Information Science and Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100101

出　　处：《计算机工程与设计》2023年第4期1267-1273,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(62171043);北京市自然科学基金项目(4212020);国防科技重点实验室基金项目(6412006200404);北京信息科技大学“勤信人才”培育计划基金项目(QXTCP B201908);北京市市教委科研计划基金项目(KM202111232001)。

摘　　要：相较传统的阅读方式,利用摘要生成等技术生成文献的创新点存在正确性低、事实准确性差、与原文的关键信息有偏差等问题。据此,提出一种融合关键信息的科技文献创新点生成方法,在该领域构建一个中文长文本-生成创新点语料库,提出一个领域知识抽取方法,用实体知识指导原文内容的选择,利用指针生成网络生成最终的创新点。实验结果表明,该方法在Rouge分数上均优于主流方法,生成的创新点具有更好的正确性和事实准确性。Compared with the traditional reading methods, the innovation of using summary generation and other technologies to generate documents has some problems, such as low correctness, poor factual accuracy, deviation from the key information of the original text and so on. Based on this, an innovation point generation method of scientific and technological literature integrating key information was proposed, a Chinese long text generation innovation point corpus in this field was constructed, a domain knowledge extraction method was proposed, the entity knowledge was used to guide the selection of original content, and pointer generation network was used to generate the final innovation point. Experimental results show that the proposed method is better than the mainstream method in Rouge score, and the generated innovation points have better correctness and factual accuracy.

关 键 词：创新点生成 摘要生成 知识抽取 强化学习 指针网络 事实准确性 语料库构建 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]