Multi-label Text Adaptive Classification Method Fused by Semantic Similarity and BERT Model
The text search requirements are difficult to judge.and the text is difficult to classify,hence,a multi-label text adap-tive classification method based on the fusion of semantic similarity and bidirectional encoder representations from transformers(BERT),bidirectional language encoder model is studied.It preprocesses the text and determines the text representation,ex-tracts and reduces the text features based on the information gain theory,calculates the similarity between texts based on the semantic similarity theory,introduces the BERT model to build a multi-label text adaptive classification framework,and obtains the model through adversarial training.The best parameters are inputted,the text to be classified is inputted into the trained text classification BERT model,then the adaptive classification of multi-label text can be realized.The experimental data show that the F1 parameter obtained by the proposed method is greater than the given minimum limit,and the Hamming loss parame-ter is less than the given maximum limit,which fully confirms that the proposed method has better text classification effect.
万方数据
维普
