Design And Implementation Of Intelligent Question Answering System Based On Movie Knowledge Base And Knowledge Grap

With the advent of the digital economy,the people are paying more and more attention to the quality of cultural tourism consumption.The mobile internet has accelerated the rapid development of China’s film industry,and people’s demand for personalized and differentiated viewing continues to expand.However,movie information in mainstream movie websites and movie apps is generally scattered,and the search results are not concise and accurate enough.Moreover,traditional search engines are unable to obtain complex correlations between movie information,which is not friendly enough for users who want to understand movie knowledge in depth or have personalized query needs.To address these problems,this thesis designs and achieves an intelligent Q&A system for film knowledge base based on the knowledge graph by collecting massive film data and constructing a knowledge graph in the film field to provide users with a more user-friendly and convenient human-computer interaction experience.The main research of this thesis are as follows:1.In order to the problem of fragmentation of movie knowledge and the difficulty of utilizing,This thesis designs a BERT-Bi LSTM-CRF movie domain named entity recognition method based on web movie data to form a movie knowledge graph containing 18.3W entities and 25.8W associated relations.Firstly,a fine-grained film ontology library containing 9 types of entities and 9 types of relationships is constructed according to the practical application needs.Then a BERT-Bi LSTM-CRF model is designed to identify the movie entities contained in the unstructured text.The pre-training model BERT is introduced to give more semantic information to the text,thus solving the problem of multiple meanings of words in Chinese vocabulary.Through experimental validation,the F1 values of the model for named entity recognition on two movie domain datasets are 92.41% and 95.84% respectively,the F1 value is 7.65% and 2.16%higher than the mainstream entity recognition.Finally,the construction of the film knowledge graph was completed in the Neo4 j graph database.2.In order to solve the problem of variable forms of film interrogations and difficulties in machine understanding,this thesis proposes the Dense Net-ATT intention recognition algorithm based on the attention mechanism.The Dense Net-ATT algorithm uses a multi-layer convolution neural network to extract multi-granularity text features.A dense connection channel is constructed between the upstream and downstream convolution blocks in order to increase the feature granularity layer by layer aggregation.By enhancing the flow of text features between the convolution blocks,text feature utilization is increased and feature loss is reduced.To further reduce the effect of noise,an attention mechanism is introduced to assign adaptive weights to text features of different granularity,focusing on the representation of important granularity features.The experimental results show that the Dense Net-ATT algorithm achieves an F1 value of 94.4%,which is a 6.2% increase in F1 value compared to the baseline model.In addition,to address the shortage of intention recognition datasets in the film domain,the open source dataset was expanded using a template generation approach.A dataset of 9,533 interrogative sentences with 9 questioning intentions was constructed.3.An intelligent Q&A system based on knowledge graph is designed and implemented to solve the problem that the film knowledge fragmentation degree is strong and the search engine can’t mine the deep-seated relationship between the film knowledge.The system takes film knowledge graph as the underlying data support and can accept natural language as input.It can automatically recognize the film entities and questioning intent contained in the question,and return a concise and accurate answer to the query through knowledge query and reasoning on the knowledge graph,the system answered with 91% accuracy.At the same time,the system provides a visual display of the film knowledge graph and an entity query module,using the knowledge graph as the answer return form to visualize the deep-seated connections between film knowledge that are not easily perceived.The system uses natural language interaction to provide users with a convenient and comfortable query experience.In addition,the use of knowledge graph enhances the explain ability of answers,providing users with the possibility to further understand film knowledge and explore potential film interests.The system has practical application value.