Chinese Book Information Extraction Based on Bert and Rule Matching

Extracting book-related information from length texts is of great importance as the basis for downstream tasks in the field of graphical intelligence. Book texts are characterized by prominent information features and simple sentences, but few studies have been conducted on the information extraction of book texts. Based on this, this paper proposes an information extraction model that combines entity extraction and rule matching using a BERT pretrained model. The BERT-wwm-ext+CRF model is used as the basis for extracting entities and formulating specific rules for book texts to match the relationships between entities so that the book information can be output as a result of a triplet. Through this method, there are 5 types of entities that can be finally recognized and 4 types of information extracted, which can match the common basic information of books and meet the basic requirements of book information processing. After experimental comparison, the method has good performance in extracting book information and can extract entities and identify entity relationships and attributes from text data more accurately. The final triplets obtained by the model can pave the way for tasks such as building knowledge graphs, thus serving smart libraries.