Postoperative risk prediction natural language data enhancement model and method

Abstract

The invention discloses a postoperative risk prediction natural language data enhancement model and method, and the method comprises the steps: enabling natural language data to obtain a pre-training model MedBERT through the training of a data set in the medical field, and converting the pre-training model MedBERT into a vector through the mode; discrete variables in table data are also converted into vectors in an entity embedding mode, and for different types of data of the table data and the discrete variables, a multi-head self-attention mode is selected to fuse the discrete variables and the vectors. The attention mechanism algorithm is used for extracting the relevance between the features and screening out important features for prediction, so that key information in the natural language data can be associated with key information of the table data, and the purpose of fusing multiple types of information is achieved; according to the method, the natural language data is brought into the task of postoperative risk prediction for the first time.

Description

technical field [0001] The invention relates to the technical field of information processing, in particular to a postoperative risk prediction natural language data enhancement model and method. Background technique [0002] Postoperative risk estimation is often viewed as a binary task. Statistical machine learning models are widely used to solve this problem, such as logistic regression (Logistic Regression, session, LR) and extreme gradient boosting (eXtreme Gradient Boosting, XGBoost). The vector-based LR method standardizes both discrete and continuous variables and inputs them into the model, and the tree-based XGBoost model directly uses structured data for training. [0003] In recent research work, many researchers have begun to use deep learning to solve the problem of predicting patients' postoperative risk because of its own complex feature expression ability and predictive performance. In these studies, perioperative tabular data were the main data source, wh...

AI Technical Summary

Problems solved by technology

The directly spliced ​​vectors will have the problem of information redundancy. When the vectors containing irrelevant information have high latitude, and the vectors containing important information have low latitude, splicing them will make the redundant information occupy the majority, resulting in the truly critical important information is ignored

Images