Form classification method and device based on BERT and TCSMN

By constructing a table classification model based on BERT and TCSMN, the timing dependence and semantic structure information of the table are extracted, and the problems of low accuracy and poor applicability of table classification in the prior art are solved, and efficient table classification on the data set with or without labels is achieved.

CN119739795APending Publication Date: 2025-04-01INST OF COMPUTING TECH CHINESE ACAD OF SCI
0 Cites 0 Cited by

Patent Information

Application Number
CN202411799665.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-09
Publication Date
2025-04-01

AI Technical Summary

Technical Problem

The prior art has problems such as low classification accuracy, poor applicability and large training data requirements in table classification, especially poor performance when dealing with unlabeled data sets.

Method used

The table classification method based on BERT and TCSMN is adopted, and the classification model of the feature extraction layer, attention layer and normalization layer including BERT, TCSMN and MLP is constructed, and the timing dependence and semantic structure information of the table are extracted to classify table categories.

Benefits of technology

It improves the accuracy of table classification, enhances the ability to extract context and semantic features, and can effectively classify tables on labeled and unlabeled datasets.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a table classification method and device based on BERT and TCSMN. The method comprises the following steps: obtaining a to-be-classified table represented by a sequence; constructing a table classification model comprising a feature extraction layer, an attention layer and a normalization layer, wherein the structure of the feature extraction layer is a BERT network model, a TCSMN network model and an MLP network model which are connected in series; the feature extraction layer is used for extracting time sequence dependence among cells in the to-be-classified table, and the to-be-classified table is converted into the intermediate table according to the time sequence dependence and the MLP network model; respectively acquiring semantic structure information of each row and each column of data in the intermediate table by utilizing the attention layer; and classifying the to-be-classified table according to the semantic structure information and a plurality of preset table categories to obtain the table category of the to-be-classified table.
Need to check novelty before this filing date? Find Prior Art