Sensitive Data Estimation Using NLP Features and Pretrained Models

The patent segments the data identification process into two distinct phases: (1) training phase where the system learns from manually labeled sensitive data to build a machine learning model, and (2) inference phase where the trained model automatically identifies sensitive data. This segmentation allows the system to achieve high accuracy through learning while maintaining high productivity during automatic identification, resolving the contradiction between measurement precision and productivity.

The patent applies preliminary action by performing manual labeling and model training in advance during the training phase. The machine learning model is pre-trained on a dataset with ground truth labels before deployment. This preliminary action creates a ready-to-use model that can rapidly identify sensitive data in the inference phase without requiring time-consuming manual scanning, thus reducing loss of time while maintaining measurement precision.

The patent introduces a machine learning model as an intermediary between the raw data and the sensitivity identification process. The model acts as a mediator that has been pre-trained on labeled data and can now automatically perform sensitivity classification. This intermediary approach simplifies the operational complexity during inference while achieving high productivity, as the complex learning process is confined to the training phase rather than requiring complex real-time processing infrastructure.