基于BERT模型的教育技术学领域实体抽取

Named Entity Recognition Method in Educational Technology Field Based on BERT

网络环境下资源丰富导致教育技术学信息量大,使得学习者认知效率低、注意力无法集中,最终偏离学习的目标并且无法完成特定的学习任务。为了解决学习者在网络学习中遇到的这些问题,该文提出一种结合BERT-BiLSTM-CRF的教育技术学主干课程命名实体识别方法,以提高学习者学习效率为目的。首先构建教学技术学主干课程命名实体识别数据集,将文本转换成计算机可识别的形式,使用BERT语言模型进行文本特征提取获取字粒度向量矩阵;然后使用双向长短期记忆网络(Bi-directional Long Short-Term Memory,BiLSTM)提取输入语句与上下文之间字与字的关系;最后使用条件随机场(Conditional Random Field,CRF)模型,根据标签之间的依赖关系提取全局最优的输出标签序列;最终得到教育技术学主干课程命名实体。实验结果表明,该模型的识别效果优于CRF、BiLSTM-CRF,该模型的精确率、召回率和F1值均有提升,整体识别性能较高。

Abundant resources in the network environment lead to a large amount of information in educational technology,which makes learners’cognitive efficiency low,unable to concentrate,and eventually deviates from the learning goal and fails to complete specific learning tasks.In order to solve these problems encountered by learners in online learning,we propose a named entity recognition method for the main curriculum of educational technology combined with BERT-BiLSTM-CRF,with the purpose of improving the learning efficiency of learners.First,a named entity recognition data set is constructed for the main course of teaching technology to convert the text into a computer-recognizable form,and the BERT language model is used for text feature extraction to obtain the word granularity vector matrix.Then BiLSTM is applied to extract the words and words between the input sentence and the context.Finally,the CRF model is used to extract the global optimal output tag sequence according to the dependency relationship between tags,and the named entity of the main course of education technology is obtained.Experimental results show that the recognition effect of such model is better than that of CRF and BiLSTM-CRF.The accuracy,recall and F1 value of such model are improved,and the overall recognition performance is higher.