关键词: 关键词: 古籍智能整理, 片段抽取原型网络, 提示学习, 自动断句标点, 深度学习

Abstract: Abstract: In view of the phenomenon that automatic sentence segmentation and punctuation task in ancient book information processing rely on large-scale annotated corpora, and considering that high-quality, large-scale training samples are expensive and difficult to obtain, a learning method for ancient text sentence segmentation and punctuation based on span-extracted prototypical network was proposed. Firstly, structured information was incorporated into the support set, forming an effective template to improve the model's learning efficiency. Then, combined with a punctuation position extractor and a prototype network classifier, the misjudgment impact and interference from non-punctuation labels in traditional sequence labeling method were effectively reduced. Experimental results show that on the "Shiji" dataset, the F1 score is 2.5 percentage points higher than with the BERT-BiGRU-CRF(Bidirectional Encoder Representation from Transformer-Bidirectional Gated Recurrent Unit -Conditional Random Field) method. In addition, on the public multi-domain ancient text data set CCLUE, the precision rate and F1 score of this method reach 91.60% and 93.12% respectively. This indicates that the method can effectively and automatically segment punctuation in multi-domain ancient text by using a small number of training samples. Therefore, the proposed method offers a new approach and thought for conducting in-depth research on automatic sentence segmentation and punctuation in multi-domain ancient text, as well as for enhancing the model's learning efficiency.

Key words: Keywords: intelligent collation of ancient books, span-extracted prototypical network, learning, automatic sentence segmentation and punctuation, deep learning