基金项目摘要的语步识别系统设计与实现

Design and Implementation of the Move Recognition System for Fund Project Abstract

[目的/意义]设计中文基金项目摘要的语步识别系统,实现基金项目摘要的自动结构化输出。[方法/过程]重点解决语步识别系统建设中的3个关键技术难点:(1)基于规则和深度学习方法构建基金项目摘要语步识别训练数据集,为系统提供数据支撑;(2)通过嵌入摘要中句子的位置信息来改进模型输入,实现语步结构的精准识别;(3)设计开放接口以实现系统的开放调用。[结果/结论]该系统已初步实现基金项目摘要的自动语步识别功能,并部署在多个平台网站上供科研人员试用。[局限]该系统目前只提供了基金项目申请摘要的语步识别服务,未来还将面向结题摘要进行语步分析与建设。

[Purpose/significance]This article intends to design a move recognition system for Chinese fund project abstracts to realize the automatic structured output of fund project abstracts.[Method/process]We focus on solving three key technical diffi-culties in the construction of the move recognition system:①Constructing fund project abstract move recognition training data based on rules and deep learning method to provide data support for the system.②Improving model input by embedding the position infor-mation of each sentence in the abstract to realize the precise recognition of sentences in each move.③Designing an open interface to support the open call of the system.[Result/conclusion]The system has realized the automatic move recognition function of fund project abstracts and has been deployed on multiple platform websites for scientific research personnel to try out.[Limitations]The system currently only provides the move recognition service for the abstracts of fund project applications,it will carry out the move analysis and construction for the concluding abstract in the future.