基于对抗学习的查新检索式自动生成  

Novelty Retrieval Expression Automatic Generation Based on Adversarial Learning

作　　者：曾立英[1] 王亭亭 刘耀[2] 王晓燕 ZENG Liying;WANG Tingting;LIU Yao;WANG Xiaoyan(College of International Education,Minzu University of China,Beijing 100081,China;Institute of Scientific and Technology Information of China,Beijing 100038,China;Library,Minzu University of China,Beijing 100081,China)

机构地区：[1]中央民族大学国际教育学院,北京100081 [2]中国科学技术信息研究所,北京100038 [3]中央民族大学图书馆,北京100081

出　　处：《郑州大学学报（理学版）》2024年第6期70-76,共7页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家社会科学基金项目(21BTQ011);中央民族大学博士研究生自主科研项目(BZKY2023099)。

摘　　要：科技查新是科研人员获取前沿信息的重要途径,但伴随着信息量的剧增,传统查新检索式的构建方法存在效率低、关键词提取不全面、一词多义等问题,因此提出了融合基于Transformer的双向编码器表达与SequenceGAN的查新检索式自动构建模型BSGAN。通过BiLSTM-CRF构建领域词表及概念同义词词表,解决了查新检索式构建过程中关键词不够全面的问题;采用基于Transformer的双向编码器表达模型中多头注意力机制,解决了检索式中一词多义问题;使用BSGAN检索式自动构建模型,实现了查新检索式的自动生成与逻辑构建,解决了传统方法中专家手工构建检索式效率低的问题。最后,通过万方中文数据库中的检索结果来评价检索式,实验结果表明,自动构建模型BSGAN生成的查新检索式在医药、化工、计算机等领域均达到了较高的查准率与查全率。Scientific and technological novelty retrieval was an important way for researchers to obtain frontline information.But with the blooming of information,the traditional construction method of novelty retrieval expression had some problems,including low efficiency,incomplete keywords extraction,polysemy,etc.Regarding the issues above,a new model called BSGAN was proposed that could combine BERT and SeqGAN for automatic construction of novelty retrieval expression.The method solved the issue that keywords were not comprehensive enough in the construction process of novelty retrieval expression by building domain vocabulary and concept synonym vocabulary through BiLSTM-CRF.At the same time,the issue of polysemy in retrieval expression was solved by using the Multi-headed Self-attention mechanism in Bert.In addition,BSGAN was used to implement the automatic generation and logical construction of novelty retrieval expression,which could solve the low efficiency of experts′traditional manual construction methods.Finally,the retrieval expression was evaluated by the retrieval results in Wanfang Chinese database.The experiment outcome showed that the novelty retrieval expression automatically generated by BSGAN achieved high precision and recall in the fields of medicine,chemical engineering,computer,etc.

关 键 词：查新检索式 对抗学习 BiLSTM-CRF TRANSFORMER 

分 类 号：G252.7[文化科学—图书馆学]