Predictive Coding
Definition:
Predictive coding is a technology that learns, and makes specific decisions based on general instructions. Software with predictive coding is used in legal review, to allow computers rather than people to identify "responsive" and "privileged" documents. Without predictive coding, human being would be given instructions and then manually read documents to identify the right documents.
Every program that uses predictive coding works differently, but the process is essentially simple.
You provide data to the software to inform it of what you are looking for, and parameters for the case. You give it the documents for the case, and the software then creates various metrics and develops a model for the quality control stage. Either you give it examples of the documents you are looking for, or the software present with a real or "dummy" example document of what it believes you are looking for. You then look at the documents that it presents to you, and makes any necessary comments. After a few rounds of back and forth the accuracy of the search increases, resulting in the "responsive" documents you are looking for.
A powerful advantage of predictive coding (as part of a Technology Assisted Review or TAR) is that once the initial review is completed, if new conditions arise, the additional factors can be entered into the software and in a matter of minutes, the review will reflect the new search parameters. In a linear (or manual ) review, if would be necessary to start the review almost from scratch, re-reviewing every document that was not found to be responsive in the initial review (typically 99.9% of documents are non-responsive).
Predictive coding is a technology that learns, and makes specific decisions based on general instructions. Software with predictive coding is used in legal review, to allow computers rather than people to identify "responsive" and "privileged" documents. Without predictive coding, human being would be given instructions and then manually read documents to identify the right documents.
Every program that uses predictive coding works differently, but the process is essentially simple.
You provide data to the software to inform it of what you are looking for, and parameters for the case. You give it the documents for the case, and the software then creates various metrics and develops a model for the quality control stage. Either you give it examples of the documents you are looking for, or the software present with a real or "dummy" example document of what it believes you are looking for. You then look at the documents that it presents to you, and makes any necessary comments. After a few rounds of back and forth the accuracy of the search increases, resulting in the "responsive" documents you are looking for.
A powerful advantage of predictive coding (as part of a Technology Assisted Review or TAR) is that once the initial review is completed, if new conditions arise, the additional factors can be entered into the software and in a matter of minutes, the review will reflect the new search parameters. In a linear (or manual ) review, if would be necessary to start the review almost from scratch, re-reviewing every document that was not found to be responsive in the initial review (typically 99.9% of documents are non-responsive).
Source...