基于双流Transformer与多层级特征融合的金融命名实体识别方法

打开文本图片集

中图分类号：TP391.1 文件标识码：A 文章编号：2096-4706（2025）16-0063-07

Abstract： Chinese financial Named Entity Recognition （NER）aims to extract entities from unstructured financial texts. Firstly，toaddress thelack offinancial domaindatasets，adatasetcontainingseven categoriessuchaspersonnames，titles， companies isconstructedSecondly，totackleissuessuchassingle featureusageandambiguous boundaries inChinesetext， an entityrecognition model that integrates dictionary features and Chinese character structural features is proposed.This modelicorpratesdictionaryiformationintocharacterrepresentatiostoenhanceentityboundariesandusesaual-stream Transformer architecture to fuse Chinese character shapes andradical features tofurther improve model performance.Finalythe experimentalresults show that the modelperforms wellonboth various datasets and theself-constructed financial dataset.

Keywords：Named Entity Recognition; feature fusion;dual-stream Transformer;financial entityrecognition

0 引言

命名实体识别旨在从非结构化文本中识别有价值的实体，如人名、地名、组织机构名等，作为自然语言处理领域一项基本任务，它在信息检索、智能问答、自动摘要、知识图谱等下游子任务中具有重要研究意义[1]。（剩余10796字）