Research And System Inplementation Of Entity Relation Extraction Technology In Network Novels

As one of people’s pastimes,reading network novels is a pleasant choice for many people in their leisure time.A novel develops many different characters to advance the story.However,due to the long length of network novels and time-consuming reading,readers can not have a deep memory to connect the previous and subsequent text,which affects the reading perception.The main research content of this thesis is how to apply deep learning technology to transform the complex and changeable text structure into clear and exact structured information.To solve the above problems,the named entity recognition model and relationship extraction model are proposed,and the two are further combined to solve the problems of entity redundancy and relationship overlap,so as to realize entity relationship extraction of network novel text.The main research results are as follows:(1)Construct the named entity recognition model for network novels based on BERT-Bi LSTM-CRFIn order to extract the features of network novel text better,construct novel text corpus,and propose the named entity recognition model based on BERT-Bi LSTM-CRF.BERT pre-training model is used to generate dynamic word vectors to improve semantic extraction and word vector representation effect,and BERT Fine-tuning can adapt to task requirements in the domain.The obtained vector sequence is input into Bi LSTM model to extract global features of text sequence data,and the optimized label sequence is calculated by CRF model.Compared with other algorithm models,the effectiveness of this model in named entity recognition task is verified.(2)Construct the entity relation extraction model for network novels based on Bi GRU and attention mechanismAiming at the entity relation extraction task of Chinese network novels,an algorithm model combining Bi GRU and attention mechanism is proposed.Using Bi GRU model to extract context information can get deeper semantic understanding,solve the problem of long-term memory,and extract global features of text.The dual attention mechanism model integrating word level and sentence level extracts the correlation degree of different characters and context in sentences,thus solving the problem of long-distance dependence.Meanwhile,it automatically assigns the feature weight to different sentence categories,so as to improve the local feature extraction effect and successfully mine the effective information in sentences.The experimental results show that the proposed model is effective in the category of entity relation extraction in network novels.(3)Construct the entity relation joint extraction model for network novels based on feature enhancementCompared with the pipeline method,the joint learning method can use the close interaction information between entities and relations to extract the relations between entities and classify entity pairs,and alleviate the shortcomings of pipeline learning error propagation.Based on the idea of decomposition strategy,this thesis proposes a joint entity relation extraction model based on feature enhancement.The BERT pre-training model is used in the input layer,and the named entity features and part-of-speech tagging features are integrated to obtain the text features.The head entity coding vector obtained through the head entity recognition layer,and the text coding information obtained by the multi-head self-attention mechanism are fused to identify the tail entity and the overall relationship.Then obtain the triplet of entity relation extraction,and complete the joint extraction experiment of network novels.Compared with the traditional research methods,the performance of the model in this thesis is significantly improved.