Research on the Construction of Topic Model in the Field of Digital Humanities Based on Sentence-BERT and LDA

In order to solve the problems that traditional methods cannot fully capture text semantic information in text topic recognition and sentence representation is not suitable for unsupervised learning, we use the Sentence-BERT model for sentence embedding tasks combined with the LDA model to construct a topic mining model, which provides new methods and ideas in the field of topic model construction methods. Firstly, the relevant literature was reviewed, then the text vectors were obtained by using Sentence-BERT and LDA models, and then the vectors of the two models were connected, and then the optimal number of clusters was judged, the clustering was carried out and the clustering effect was evaluated, and finally conduct comparative experiments. Through experiments on relevant literature data in the field of digital humanities, the model has a good effect on topic extraction, clear topic division, short calculation time, and it can effectively capture the features in the text.