Research On Knowledge-Based Question Answering Dialogue System

Knowledge-based dialogue question answering is an important way for human-machine interaction.With the development of human-machine dialogue,it is of great significance for the computer to accurately understand the user's query intention.This thesis studies two main parts of the existing knowledge-based dialogue question answering task:intention recogni-tion and knowledge-based question representation.Intention recognition aims at judging the user's intention in human machine dialogue and improves the accuracy and naturalness of the human machine dialogue system.This thesis studies the existing corpus in the field of inten-tion recognition and proposes a hybrid model to effectively improve the system performance This thesis further studies the influence of modeling abstract meaning representation struc-ture on knowledge-base question answering on English datasets.By introducing abstract meaning representation and modeling,the system can be assisted to obtain additional infor-mation and improve overall performance.A general knowledge-based question answering system is studied by introducing pre-training language models on Chinese datasets(1)Hybrid Models for Human-machine Dialogue Intention ClassificationThis thesis first analyzes the advantages and disadvantages of multiple classification models in the intention classification task.On this basis,this thesis proposes a novel hybrid neural network model to comprehensively utilize the diversity outputs of multiple deep net-work models.To further improve the performance,the language model embedding is used in the input feature preprocessing and the semantic mining ability possessed for the hybrid network which can effectively improve the expression ability of the model.The proposed model achieves 2.95%and 3.85%performance improvement on the two data sets respec-tively compared to the best benchmark model.Our model achieves optimal performance on both datasets(2)Modeling AMR Structures for Knowledge-Based Question AnsweringIn previous studies on knowledge based question answering(KBQA),staged query graph generation is a widely used approach to obtaining representation of input questions.The representation achieved this way generally considers token sequences and syntactic in-formation,but without deep semantic information.To the best of our knowledge,there are no previous works that utilize staged query graph generation and semantic parsing together to obtain question representation.In this thesis we for the first time propose to model Ab-stract Meaning Representation(AMR)structures of questions into KBQA systems.AMR is a representative semantic formalism which encodes semantics of a sentence into a directed and typed graph.We design two different methods to encode AMR graphs of questions into dense vectors which are used together with the representation from query graph generation.Experimental results on a benchmark dataset show that integrating AMR parsing helps to achieve new state-of-the-art performance as compared to the previous studies.(3)Research on Chinese Knowledge-Based Question Answering SystemIn this thesis,a question answering system including subject entity identification,entity link and attribute link is built for large-scale Chinese knowledge based question answering.After named entity recognition based on language model and BiLSTM-CRF model compo-nents,we propose two simple strategies to complete entity linking,and finally use semantic similarity calculation based on language model to complete attribute linking.Our system achieves excellent performance on a public test dataset.