Research On Algorithm Of Neural Network Data Label Generation

With the development of Internet technology and the popularization of applications,the amount of information in the Internet is growing exponentially.It is difficult to find the text resources you need in such a huge amount of data.Classify massive resources and automatically generate different labels for resources to meet people's needs for resource management and use,which has important research value.In view of the above problems,this article studies the text multi-label classification algorithm,which mainly includes the following aspects.(1)Research on text feature extraction based on neural network.Use XML-CNN to extract local information features of text,and use Bi-GRU to extract global information features of text,and fuse different text features to obtain a more comprehensive representation of text feature information.And considering the importance of different words in the text,a Self-Attention mechanism is added to the Bi-GRU model to add weight to the information characteristics of different words to obtain more accurate global information representation.(2)Construction of multi-label classification algorithm.In order to fully consider the influence of text features on the results of multi-label classification,this topic fuses local feature vectors of text and global feature vectors of text,and uses the sigmoid function as a multi-label classifier.Based on XML-CNN and Bi-LSTM+Self-Attention the multi-label classification algorithm provides a theoretical basis for the realization of the text label generation scheme.(3)Design test and analysis of text multi-label generation scheme.The Tensorflow2 framework was used to build the model of the text multi-label generation algorithm for this topic,and the training of the model was completed,and four evaluation indicators of micro-Precision,micro-Recall,micro fl-score and Hamming-loss were used to compare the model's Excellence.(4)The experimental results show that:the effectiveness of this subject model on the cnki paper dataset,AAPD dataset and RCV1-V2 dataset has a certain improvement compared with the benchmark model.Among them,the micro F1-score on the cnki paper dataset The optimal value is 0.7926,and the label generation effect of the model is better.