Systems and methods for censoring text inline

A method for censoring text-based data, the method comprising:

• receiving a plurality of target pattern types;
• obtaining a plurality of machine learning models for identifying target data patterns, the machine learning models corresponding to target pattern types, wherein at least one of the machine learning models is configured to perform operations comprising:
• identifying, within text-based data, a first target data pattern corresponding to a first one of the target pattern types;
• identifying first target characters within the first target data pattern;
• and
• censoring the first target characters;
• receiving text-based data;
• identifying within the received text-based data, the first target data pattern corresponding to the first target pattern type using the selected machine learning models;
• retrieving a profile associated with a receiving party, the profile containing a first one of a plurality of permission levels for the first target pattern type;
• accessing a token corresponding to the first target data pattern, the token associating the first target data pattern with:
• the permission levels;
• and
• context-based alternative user information, the context-based alternative user information corresponding to the permission levels and having a plurality of relative data sensitivities;
• retrieving a first part of the context based alternative user information, the first part of the context-based alternative user information corresponding to:
• the first target characters;
• and
• the first permission level;
• and
• modifying the text-based data by replacing the first target characters with the first part of the context-based alternative user information.

A non-transitory computer-readable medium storing instructions that, when executed by at least one a processor, cause the processor to perform operations comprising:

• receiving text-based data;
• receiving a target pattern type;
• generating training data, the training data comprising text-based data containing a plurality of target data patterns including target characters, the target data patterns corresponding to the target pattern type;
• training a computer-based model using the training data, the model identifying:
• a first one of the target data patterns, the first target data pattern corresponding to the target pattern type;
• first target characters within the first target data pattern;
• a second one of the target data patterns, the second target data pattern corresponding to the target pattern type;
• and
• second target characters within the second target data pattern;
• validating the model by:
• accessing validation data comprising a plurality of target probability values corresponding to third target characters;
• calculating an accuracy score based on a difference between a plurality of probability values generated by the model for the third target characters and the target probability values;
• and
• determining that the accuracy score exceeds a threshold score;
• and
• retrieving a profile associated with a receiving party, the profile containing a first one of a plurality of permission levels corresponding to the target pattern type;
• accessing a token corresponding to one of the first or second target data pattern, the token associating the one of the first or second target data pattern with:
• the plurality of permission levels;
• and
• context-based alternative user information corresponding to the permission levels;
• retrieving a first part of the context-based alternative user information associated with the token and corresponding to:
• the one of the first or second target characters;
• and
• the first permission level;
• censoring the text-based data by replacing the one of the first or second target characters, within the first and the second identified target data patterns, with the first part of the alternative user information;
• and
• transmitting the censored text-based data.