一种基于RoBERTa模型的文本搜索排序方法  

Text Retrieval Sorting Method Based on RoBERTa Model

作　　者：唐伟广[1] 陈勇[1] 姚剑[1] TANG Weiguang;CHEN Yong;YAO Jian(The 54th Research Institute of CETC,Shijiazhuang 050081,China)

机构地区：[1]中国电子科技集团公司第五十四研究所,河北石家庄050081

出　　处：《计算机与网络》2024年第5期448-455,共8页Computer & Network

摘　　要：针对日益增长的资料快速检索共享需求,利用鲁棒性优化的BERT方法(Robustly optimized BERT approach,RoBERTa)预训练模型对现有资料进行训练,基于Transformer自注意力机制的语言学习模型,生成文本嵌入向量,将文本向量作为全文本的上下文表征。通过将关键搜索词向量化,使用欧氏距离计算向量与其他向量之间的距离,并使用快速排序算法,以找到最相似的向量输出显示,解决基于内容和上下文语义搜索的应用需求。In response to the growing demand for rapid retrieval and sharing of information,the per-trained model of Robustly optimized BERT approach(RoBERTa)is used to train the existing data.Based on the language learning model of Transformer self-attention mechanism,the text embedding vector is generated,and the text vector is used as the context representation of the full text.The key search words are vectorized,and the distance between the vector and other vectors is calculated by Euclidean distance.And quick sort is used to find the most similar vector output display,so as to solve the application requirements of content-based and contextual semantic search.

关 键 词：TRANSFORMER 文本搜索 注意力机制 嵌入向量 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]