Named entity recognition method based on joint entity boundary detection

To solve the problem that traditional named entity recognition methods cannot effectively utilize entity boundary information, a named entity recognition method based on joint entity boundary detection was proposed. The method took entity boundary detection as an auxiliary task, so that the model can enhance the ability of entity boundary recognition, and then improve the effect of entity recognition. Firstly, the Bert pretraining language model was used to embed the features of the original text to obtain word vectors, and the self-attention mechanism was introduced to enrich the context features of words. Secondly, on the basis of named entity recognition task, an auxiliary entity boundary detection task was added to enhance the recognition ability of the model to the entity boundaries. Thirdly, the effectiveness of the named entity recognition method and the baseline method was compared, and the test results were from ablation experiments. Finally, the influence of loss weight β on entity boundary detection was analyzed by examples. The experimental results show that on the English social media dataset Twitter-2015, the named entity recognition method combined with entity boundary detection achieves higher accuracy, recall rate and F1 value than the baseline model, of which the F1 value can reach 7357%. In addition, the boundary detection auxiliary task has a certain improvement effect on the baseline method. The proposed method can effectively utilize entity boundary information to obtain better entity recognition effect, and promote the development of human-computer interaction system, which is of great significance for downstream tasks of natural language processing.