Design And Implementation Of Phrase Recognition Algorithm In Natural Language Processing Domain

Phrase recognition plays an important role in semantic comprehension.Phrase recognition technology can improve the computer's understanding of natural language by automatically dividing phrases that have been correctly segmented and marked with parts of speech,and can also be of great help to the text intention and emotion analysis.In order to obtain a phrase recognition model with high accuracy,the user input data is cleaned,the meaningless symbols are filtered,and the single sentence is segmented according to different types of symbols.Based on the CBOW model,word2 vec Chinese word vector is trained and the training part of speech vector is randomly initialized.Several bilstm-based models are designed and implemented as feature extraction networks to extract sentence context features.A phrase annotation system is designed to allow the existence of nested structures.For the extracted sentence features,the fully connected network is used to distinguish whether the window is a phrase window or a background,and further regression correction is made for the judged phrase window.The loss function of phrase window prediction has two parts: sigmoid function is used to calculate the classification loss of window,and RMSE is used to calculate the regression loss of phrase window bias correction.Finally,the softmax classifier is used to classify each candidate window.After completing the design and implementation of the phrase recognition algorithm,different feature extraction network models were tested and compared.In the task of phrase recognition algorithm,the F1 value of the model reached 87.5%.Finally,the algorithm is tested comprehensively in terms of function and performance,and it is found that each module achieves the desired effect.