徐雪松* **,肖刚**,刘星光***,张侠*,叶江涛**,程振波**.融合关联语义信息的工程领域标准规范表格检索方法[J].高技术通讯(中文),2025,35(7):724~733 |
融合关联语义信息的工程领域标准规范表格检索方法 |
Table retrieval method for engineering standard specifications with correlated semantic information |
|
DOI:10. 3772 / j. issn. 1002-0470. 2025. 07. 005 |
中文关键词: 表格检索;特征融合;自然语言处理;语义表示;标准规范 |
英文关键词: table retrieval,feature fusion,natural language processing,semantic representation,standard specification |
基金项目: |
作者 | 单位 | 徐雪松* ** | (* 浙江工业大学机械工程学院 杭州 310023)
(** 浙江工业大学计算机科学与技术学院 杭州 310023)
(*** 行吟信息科技(上海)有限公司 上海 200003) | 肖刚** | | 刘星光*** | | 张侠* | | 叶江涛** | | 程振波** | |
|
摘要点击次数: 348 |
全文下载次数: 525 |
中文摘要: |
根据工程标准规范中表格的特点,提出通过构建问句和表格多属性之间的关注联合表示,实现经由问句自动检索其关联表格的方法。 该方法先计算问句关键词与表格标题、表头以及单元内容等属性之间的预关注度,得到问句与表格的关注度向量。 然后,通过双向长短期记忆网络(bidirectional long short term memory network,Bi-LSTM)和多头自注意力机制(multi-headed self-attention,MHA)形成问句与表格的联合向量。 利用关注向量与联合向量的连接构成问句与表格的关注联合表示,并将其作为单层感知机的输入,得到问句与表格之间的相似度。 在公开的中文表格数据集和工程领域的表格数据集上进行实验,结果表明本文方法在检索准确率上具有显著优越性。 |
英文摘要: |
According to the characteristics of tables in engineering standard specifications, a method is proposed to achieve automatic retrieval of their associated tables via query sentences by constructing a concentration joint representation between query sentences and multiple attributes of tables. Firstly,the concentration vector is obtained by calculating the pre-concentration between the keywords of the query and the attributes of the table title,table header and cells. Subsequently,the joint vector of query and tables is obtained by the bidirectional long short term memory network (Bi-LSTM) and multi-headed self-attention (MHA). Then,the concentration-joint representation between query and tables is constructed by fusing the concentration vector with the joint vector. Finally,the concentration-joint vector is input to the single-layer perceptron to obtain the similarity between query and tables. Experimental verification is carried out on the publicly available Chinese table datasets and constructed table datasets of engineering fields,which shows that the proposed method has significant superiority in retrieval accuracy. |
查看全文
查看/发表评论 下载PDF阅读器 |
关闭 |
|
|
|