BERT-LDA model-based text topic recognition method

• Main Authors: MIAO YONGMING, ZHANG YAN, LI YAOKUANG
• Format: Patent
• Language: Chinese; English
• Published: 05.01.2024
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a text theme recognition method based on a BERT-LDA model, and relates to the field of theme recognition. According to the technical scheme provided by the invention, a semantic word vector extracted by a BERT model is connected with a subject word vector extracted by an LDA model by means of the BERT model, and the connected word vectors are clustered by using a K-means algorithm. According to the method, the correlation between combined word vectors can be effectively analyzed, and the semantics and the importance degree of vocabularies can be fully mined, so that the aim of precisely recognizing themes is fulfilled. The key topic recognition method has the core advantages that context semantic information can be fully combined through the connected word vectors, the disadvantage of an LDA topic model is made up, better topic vectors are trained, and the key topic recognition effect with better fine granularity and clustering accuracy is obtained. 本发明提供了一种基于BERT-LDA模型的文本主题识别方法，涉及主题识别领