Research And Implementation Of Multi-label News Text Classification System Based On Deep Learning

With the refinement of user needs and the diversification of network news text content in the era of big data,a news text instance can be divided into multiple categories.Now,the task of news text classification has changed into multi-label text classification task.However,in the existing multi-label text classification research,The semantic relationship between fine-grained text content and tags is not considered deeply,and there are some defects in various algorithms for unbalanced multi-label data sets.Therefore,this paper proposes a multi-label news text classification system based on deep learning,which is used to accurately screen out the news content that meets users’ needs from the massive network news text.Firstly,in order to reduce the imbalance ratio of multi-label data sets and improve the results of unbalanced data sets in classification tasks,this paper proposes a FAA-MLSMOTE algorithm based on attention mechanism.Based on the constructed tag forest,the algorithm extends the sample hierarchically,then the linear relationship between the old and new samples is analyzed and the tag set of samples is generated by combining the attention mechanism.Compared with other oversampling algorithms,the expanded dataset has a certain improvement effect on BR,CLR and MLk NN classifiers.Then,the expanded data set is applied to the WTL-BERT model proposed in this paper,which adds a WTL encoder on the basis of the traditional Bert model.The encoder analyzes the semantic connection between fine-grained text content and tags by WTL attentional mechanism using the word feature vector output by Transformer encoder and tag feature vector.The model is equipped with 12-layer Transformer encoder and WTL encoder to deeply mine the correlation between words and tags.Finally,the CLS vector of the last layer of the model is input into the classification network to obtain the classification result.Experimental results show that WTL-Bert model has good performance in several evaluation indexes.Finally,this paper applies the above innovative methods to the multi-label news text classification system and tests the main functions of the system.The test results show that a multi-label news text classification system based on deep learning has certain reliability.