Visual understanding and diagnosis method of interactive NL2SQL model

Abstract

The invention discloses a visual understanding and diagnosis method for an interactive NL2SQL model. The method comprises the following steps: 1) extracting NL2SQL model data; 2) carrying out WikiSQLdata feature extraction; 3) carrying out visual analysis of the NL2SQL model: (3-1) providing a statistical view for preliminary exploration; (3-2) providing detailed information and explored dimension reduction projection views; (3-3) displaying internal scores of the model and detail views of the original questions; (3-4) providing a control panel view for parameter selection and filtering; and4) visual diagnosis of the model based on data feature extraction. Model analysis, feature extraction and multi-dimensional exploration methods are fused, an interactive visual analysis system is designed, a user is allowed to explore the internal relation among an NL2SQL model, intermediate score data and model input and output data in an interactive mode, and a WikiSQL data set is used for conducting empirical research to analyze the effectiveness and efficiency of the system.

Description

technical field [0001] The invention relates to a visual understanding and diagnosis method of an interactive NL2SQL model. Background technique [0002] The external data of the machine learning model implies semantic information related to the model, but due to the complexity of the deep learning network used by the model, a large amount of intermediate data generated cannot be directly used for model interpretation and analysis. Moreover, the data volume in the middle of the model is relatively large, with high dimensions and many features, making it difficult to visualize. However, the existing NL2SQL model based on neural network mainly focuses on the optimization of the internal structure of the model, and rarely considers the relationship between the input and output of the model and the relationship between intermediate data and model results, which leads to the incomplete operation mechanism of the machine learning network. There are still the following main challe...

AI Technical Summary

Problems solved by technology

[0002] The external data of the machine learning model implies semantic information related to the model, but due to the complexity of the deep learning network used by the model, a large amount of intermediate data generated cannot be directly used for model interpretation and analysis

Moreover, the data volume in the middle of the model is relatively large, with high dimensions and many features, making it difficult to visualize

However, the existing NL2SQL model based on neural network mainly focuses on the optimization of the internal structure of the model, and rarely considers the relationship between the input and output of the model and the relationship between intermediate data and model results, which leads to the incomplete operation mechanism of the machine learning network. There are still the following main challenges: ①. The natural language model finds that there is a large amount of natural language semantic information, and it is difficult to visualize the expression and quickly discover the laws that humans can understand.

②. The intermediate results of the natural language model have the characteristics of high dimensionality and many features that are not easy to be understood by humans. The relationship between the external data of the model and the model results is not easy to analyze, and it also brings great challenges to interactive visual analysis.

Images