基于时间卷积网络改进的自动数学应用题解算器  

Improved automatic math word solver based on temporal convolutional networks

作　　者：杨波 张珑 罗琨杰 孙华志[1] YANG Bo;ZHANG Long;LUO Kunjie;SUN Huazhi(College of Computer and Information Engineering,Tianjin Normal University,Tianjin 300387,China)

机构地区：[1]天津师范大学计算机与信息工程学院,天津300387

出　　处：《天津师范大学学报（自然科学版）》2021年第5期69-74,共6页Journal of Tianjin Normal University：Natural Science Edition

基　　金：国家自然科学基金面上项目(61771173);天津市自然科学基金重点项目(20JCZDJC00400).

摘　　要：在传统的seq2seq模型的基础上,加入时间卷积网络TCN设计了一种自动数学应用题解算器模型seq2seq+TCN.在数据预处理部分采用FoolNLTK分词工具和ELMo词嵌入方法,以保证较高质量的数据输入;利用LSTM和GRU的不同组合构成seq2seq模型的编码器和解码器;在seq2seq模型中加入TCN,以扩大解码器的感受野,保证应用题文本的时序性.在数据集Math23K和AI2上进行实验,结果表明,由BiLSTM-LSTM构成的seq2seq+TCN模型性能最优,答案准确率分别达到70.3%和87.5%,高于现有的2个集成seq2seq的解算器.An automatic math word solver is designed on the basis of the traditional seq2seq model by adding temporal convolutional network TCN.In the data preprocessing section,word segmentation tool FoolNLTK and word embedding method ELMo are adopted to ensure high-quality data input.Different combinations of LSTM and GRU are used to construct the encoder and decoder of seq2seq model.TCN in the seq2seq model can expand receptive field of the decoder and guarantee the timing of the text of the math word problems.The model is trained on databases Math23K and AI2.The results show that the performance of seq2seq+TCN model composed of BiLSTM-LSTM is best,whose accuracies of answers reach 70.3%and 87.5%,respectively,higher than those of two existing solvers integrated with seq2seq.

关 键 词：数学应用题解算器 深度学习 seq2seq模型 TCN LSTM GRU 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]