Research On Entity Recognition Based On Knowledge Enhancement And Convolutional Neural Network

Named entity recognition,a very fundamental task in the field of natural language processing,has received extensive research attention in recent years.The fundamental aim of this task is to identify the corresponding entities and their types from a piece of text described in natural language.Knowledge graphs,as external knowledge in the form of triples,have been used with good results in many natural language processing tasks.Most of the existing research approaches have neglected the external knowledge and the help of feature extraction for named entity recognition tasks.Based on the above-mentioned problems,the main points of this paper are as follows.(1)A method for incorporating external knowledge into named entity recognition models is investigated.To address the problem that most current methods simply combine pre-trained language models and prediction layers,ignoring the positive effect of external knowledge on the semantics of sentences resulting in insufficient learning of semantic information between individual words,this paper proposes an attention-aligned named entity recognition method based on knowledge graph enhancement.Firstly,the knowledge graph information is embedded through the embedding layer and attention mechanism to obtain the representation of the knowledge graph triad information;secondly,the contextual representation of the sentence is obtained through BERT-BiLSTM;then the representation of the knowledge graph information and the sentence information are fused through an attention alignment module assigning triad weights;finally,the predicted output of the fused representation vector is controlled through softmax,and thus obtain the labels of the entities.The method effectively avoids changing the semantic information of the original sentence due to the fusion of the knowledge graph,while also making the word vectors in the sentence rich in external knowledge.Experimental results confirm the improved effectiveness of the method compared to partial baseline models.(2)A method for applying convolutional neural networks to named entity recognition tasks is investigated.To address the problem that most of the current named entity recognition task models can only solve a single continuous entity recognition and perform poorly on nested entity recognition tasks,this paper investigates an encoding-decoding model incorporating a convolutional neural network to solve both discontinuous entity and nested entity recognition.The model consists of an encoding layer,a convolutional layer and a prediction layer.The encoding layer uses DeBERTa and GRU to extract sentence features;the convolutional layer uses multi-granularity convolution;and the prediction layer uses multi-headed attention mechanism and MLP mechanism to predict word-word relationships to solve the problem of nested entity recognition.The experiments also show that the results of this study are improved compared to most existing models for both continuous and nested entity recognition tasks.(3)A visual named entity recognition system was developed and designed for online recognition of entities in text.Users of the system can log in to the system platform by registering an account,and can use the services provided by the platform for both flat and nested entity recognition when using the system.At the same time,the administrator can view the user usage through the backend monitoring function after logging into the system platform,so that he can have an overview of the real-time status of the system.In summary,this paper proposes a named entity recognition model based on knowledge augmentation and convolutional neural networks for flat entity recognition and nested entity recognition respectively,and the experimental results demonstrate the improved effectiveness of the model proposed in this study compared to some existing baseline models.Finally,an online named entity recognition system based on these two models is designed to invoke these two pieces of functionality for users.