Research On Intelligent Question Answering Technology For Ancient Chinese Poetry

With the development of deep learning technology in recent years,intelligent question answering has become widely used.However,research on intelligent question answering in ancient poetry is relatively scarce.It has been demonstrated that intelligent question answering for ancient poetry can improve the efficiency of obtaining knowledge related to ancient poetry and facilitate a more personalized search for information about ancient poetry.Ancient poetry emphasizes concise writing while pursuing the combination of rhythm and meter.The vocabulary and grammatical structure used in ancient poetry differ significantly from those used in modern Chinese.Therefore,it is difficult to extend the research on intelligent question answering directly to the field of ancient poetry intelligent question answering,and it is difficult to support the urgent needs of ancient poetry intelligent question answering.Ancient poetry intelligent question answering presents the following challenges: first,simply using ancient poetry corpus to fine-tune the pre-trained language model cannot align the semantic space between ancient poetry and modern Chinese,and it is difficult to accurately obtain the representation of ancient poetry.Second,ancient poetry intelligent question answering relies on ancient poetry knowledge graphs.It is important to note that at the present time,there is a relatively small number of ancient poetry knowledge graphs available,and they are not generally open-source.Based on the above challenges,this thesis uses the pre-training language model of ancient poetry to obtain a more accurate representation of ancient poetry,construct a high-quality knowledge graph of ancient poetry,and propose an intelligent question answering model of ancient poetry.The specific work is as follows:(1)A pre-trained language model for ancient Chinese poetry(CP-Chinese BERT)is proposed.In order to align the semantic space of ancient poetry and modern Chinese,this thesis reconstructs the lines of ancient poetry into a combination of ancient poetry and its modern Chinese translations,designs three reconstruction strategies for ancient poetry and translations,and proposes CP-Chinese BERT.Using the labeled data of ancient poetry and translations,the model can learn the semantic relationship between ancient poetry and translations,resulting in more accurate representations of ancient poetry.As a result of the reconstruction of ancient poetry and translations,the amount of training data has doubled.The CP-Chinese BERT can better align the semantic space of ancient poetry and modern Chinese through training in large-scale ancient poetry and translation annotation data.This thesis conducted experiments on ancient poetry emotion classification,ancient poetry title prediction,and ancient poetry recommendation tasks.According to the results,CPChinese BERT performed better in downstream tasks related to ancient poetry than baseline models.(2)This thesis proposes an intelligent question answering model for ancient poetry(KBQA-CP-Chinese BERT)based on the Tang poetry knowledge graph.First,this thesis constructs a knowledge graph(TPKG)of 300 Tang poems.TPKG focuses on poetry,poets,and time,introducing a three-level fine-grained framework for the hierarchical storage of poetic knowledge and introducing a time-chain framework for reasoning related to poetry and poets.Secondly,to obtain a more accurate representation of ancient poetry,the model uses CP-Chinese BERT to encode ancient poetry questions and knowledge graphs and uses graph attention networks to obtain the semantic representation of the ancient poetry knowledge graphs.Finally,considering that the intention and entity of the question are closely related to the semantics of the question expression,this thesis introduces two auxiliary tasks,entity recognition and intention recognition,that enable a more accurate response to the user’s question.According to the results of the experiment,KBQA-CPChinese BERT has achieved good results when answering ancient poetry questions.