摘要
针对传统教学系统在交互过程中存在问答质量和问答效率低的问题,提出一种基于A-GCN网络和知识图谱技术相结合的问答交互生成方法。首先,采用A-GCN网络+微调T5-Pegasus模型对交互中的问题摘要文本进行生成;然后结合知识图谱生成答案;最后对两种生成方法进行验证,并将其部署在教学系统中进行测试应用。实验结果表明,本摘要生成方法的ROUGE-N、ROUGE-L和BLEU指标分别为94.44、92.81和60.35,比传统的PGN方法和Seq2Seq方法的摘要生成质量更好。对比于讯火星火认知大模型,基于知识图谱的答案生成方法生成答案更接近参考答案。通过这两种方法可提升系统问答交互效果。由此证明,本方法可在教学系统的问句摘要和答案生成任务中进行实际应用。
In view of the problems of low quality and low efficiency of the traditional teaching system,the combination of A-GCN network and knowledge graph technology are proposed.First,A-GCN network+fine-tuning T5-Pegasus model is used to generate the problem summary text in the interaction;then generate answers with the knowledge graph;finally,the two generation methods are verified and tested in the teaching system.Experimental results show that the ROUGE-N,ROUGE-L and BLEU indexes are 94.44,92.81 and 60.35,respectively,which are of better quality than the traditional PGN and Seq2Seq methods.Compared with the cognitive model,the answer generation method based on knowledge graph is closer to the reference answer.Through these two methods,the interaction effect of the system can be improved.This proves that this method can be applied in the question summary and answer generation task of the teaching system.
作者
赵妍
张晶
苏红梅
ZHAO Yan;ZHANG Jing;SU Hongmei(Xi’an Siyuan University,Xi’an 710000,China)
出处
《自动化与仪器仪表》
2025年第8期138-142,共5页
Automation & Instrumentation
基金
西安思源学院2024-2025学年校长基金重点项目《AI赋能“知识-能力-问题”图谱建构:校本一流课程教学模式创新实践研究》(XASYA24ZD02)。
作者简介
赵妍(1990-),女,甘肃白银人,硕士研究生,讲师。
