面向协作知识建构会话的智能观点分类研究——基于深度神经网络与会话分析的融合方法  

Research on Intelligent Viewpoint Classification for Collaborative Knowledge Construction Discourse--Based on the Fusion Method of Deep Neural Network and Discourse Analysis

作　　者：马志强 王文秋 MA Zhi-qiang;WANG Wen-qiu(Jiangsu Research Base for“Internet Plus Education”,Jiangnan University,Wuxi,Jiangsu,China 214100)

机构地区：[1]江南大学江苏“互联网+教育”研究基地,江苏无锡214100

出　　处：《现代教育技术》2022年第6期5-14,共10页Modern Educational Technology

摘　　要：在面向知识建构的会话智能分析研究中,如何改善原有互动行为分析视角的局限,从会话观点的角度精准描述会话的语义特征以实现对会话进行自动分类,是研究者关注的核心问题。基于此,文章融合深度神经网络与会话分析方法,构建了包含相关度、纵深度、聚敛度三种会话分类特征的面向协作知识建构会话的智能观点分类框架,并设计包含六个环节的智能观点分类流程,引入到自然语言处理领域的BERT、TextCNN、Fasttext模型,从精确率、召回率、F1值、准确率四个指标对三种模型的会话分类特征进行比较,发现BERT模型在整体语义特征、单一会话类型的分类性能上均拥有更高的准确率。文章探索数智融合的会话分析框架与分析路径,证明了深度神经网络在协作知识建构会话智能量化分析中的可行性,有助于改善智能会话分析的质量与效率。In the research of discourse intelligence analysis for knowledge construction,how to improve the limitations of the original interactive behavior analysis perspective,and accurately describe the semantic features of discourse from the discourse perspective,so as to realize automatic classification of discourse is the core concern of researchers.Based on this,fused with the deep neural network and discourse analysis method,an intelligent viewpoint classification framework for collaborative knowledge construction discourse was constructed,which contained three discourse classification features of relevance,depth,and convergence.Meanwhile,an intelligent viewpoint classification process consisting of six steps was designed,and further the process was introduced into the BERT,TextCNN,and Fasttext models of natural language processing,and the discourse classification characteristics of the three models were compared in terms of four indicators,namely precision,recall,F1-score,and accuracy.It was found that the BERT model has higher accuracy in overall semantic features and classification performance of single discourse type.This paper explored the analysis framework and analysis path of discourse analysis of the fusion of data and intelligence,which proved the feasibility of the deep neural network in the intelligence quantitative analysis of collaborative knowledge construction discourse,and was helpful for improving the quality and efficiency of intelligent discourse analysis.

关 键 词：人工智能教育 学习分析 会话分析 知识建构 观点改进 

分 类 号：G40-057[文化科学—教育学原理]