面向物流标书关键信息的自动提取方法  

Automatic Extraction Method of Key Information from Logistics Bidding Documents

作　　者：马静[1] 俞瑛[1,2] MA Jing;YU Ying(College of Economics and Management,Nanjing University of Aeronautics and Astronautics,Nanjing 210016;Keyi College of Zhejiang Sci-Tech University,Shaoxing 312300)

机构地区：[1]南京航空航天大学经济与管理学院,南京211106 [2]浙江理工大学科技与艺术学院,绍兴312300

出　　处：《计算机与数字工程》2024年第5期1400-1405,共6页Computer & Digital Engineering

基　　金：国家自然科学基金面上项目“基于多模态特征融合的网络舆情智能发现与检测及演化研究”(编号:72174086);国家社会科学基金重大招标项目“新时代军民融合创新资源配置效率测度与评价研究”(编号:20ZDA092);南京航空航天大学前瞻性发展策略研究基金项目“基于网络舆情大数据的我国跨境电商营商环境优化策略研究”(编号:NW2020001)资助。

摘　　要：物流标书信息识别是物流企业精准获取商业情报的一项重要手段。物流招标公告作为情报来源,存在文本不规范、领域标签不足、抽取难度大等问题,因此目前尚缺乏对物流标书领域命名实体识别的研究。为准确识别物流标书中的有效信息,该研究以中国物流招标网和物通网获取的物流招标公告作为实验数据集,构建了物流标书领域命名实体识别语料库,并提出一种融合BERT预训练的BiLSTM-CRF深度学习模型,以实现对招标公司、运输地点和运输项目三类实体的识别。实验结果表明,论文所提模型不但具有较优的实体识别准确度,同时还具备了较强的稳定性。Logistics bid information identification is an important means for logistics companies to accurately obtain business intelligence.As a source of information,logistics bidding announcements have problems such as irregular text,insufficient label-ing,and difficulty in extracting.Therefore,there is a lack of named entity recognition in the field of logistics bidding documents.To accurately identify the effective information,this study uses logistics bidding announcements obtained from China Logistics Bidding Network and Wutong Network as experimental dataset,then constructs a corpus about logistics bidding documents,and proposes a BERT-BiLSTM-CRF model to identify the tenderee,shipping location and shipping item three entities in the logistics bidding docu-ments.Experimental results show that the model proposed in this paper not only has better entity recognition accuracy but also has strong stability.

关 键 词：物流标书 命名实体识别 深度学习 BERT BiLSTM-CRF 

分 类 号：O141.4[理学—数学]