A review platform can provide a fantastic tool to review a very large volume of documents, allowing interaction between legal teams from around the world, and retaining a full history of who marked which document and when.
Despite the complex task review platforms perform, the basic concept behind a standard review platform is quite simple:
Store files, allow the files to be searched and filtered, keep track of what happens.
How does a review platform work?
A review platform provides a web interface for document review and therefore primarily consists of a web server (for the human interaction with the documents), a file server (to store the files) and a database (to track and monitor the documents).
What data is sent to a review platform?
Currently it is not possible to simply send PST files, or a CD full of word documents to a review platform and just load them into the system and expect it to present them to lawyers in an reviewable manner. Though this is certainly the future direction of the technology. Currently the data has to be processed by ediscovery tools before it can be loaded into a review platform.
The processing of data takes native files and conducts the following actions:
- Extracts out the metadata from native files. Examples of metadata extracted from a document include: Date created, author, last printed, date sent, document sent.
- Extracts out the text from the document, the actual body of a document
- Creates a CSV, XML file (or other similar listing) of the native files, and their metadata, this is known as a load file.
The text from a document, the load file and the native files are then sent to a review platform and loaded into the system. The text from the documents are used to create an index of the data to allow rapid keyword searching of the files, the load file keeps track of which document belongs to which text file and all of the assocated metadata, and the native files are what is shown to a reviewer when they want to see the actual document.
The load file
The load file is really the key to the review process. It is this which is loaded into the database and allows the reviewer to sort, filter, and track documents. For example if a reviewer wants to see all of the documents in a particular date range, they conduct a search trough the web interface which will then create a querry in the SQL database to search the load file and indentify all documents in the given data range. In many ways this is no more complex than applying a filter to a standard spreadsheet. Once a search is done a list of documents, which is simply a fitered version of the loaded file, will be presented through the web interface to the reviewer, on each line off the list will be the name of the document, and the metdata about the document. There will normally be a hyper link on each line which will allow the reviewer to click and view the native document.
If the reviewer conducts a keyword search of the documents a simialr process is conducted. The search is actually conducted of the text files, which are indexed (often by DTSearch). When the relevant text files are found the load file identifies which document this relates to and, as with the filtering, brings back a list of documents for the reviewer.
When documents of relevance are found the reviewer can mark/tag/catergorize these. Normally with a check box, pull down list, or other choice option. Choices normally available to reviewers include; relevant/responsive, not-relevant/non-responsive, priviledge, hot doc/important document, document relates to issue1/issue2/etc.
These marks/tags are entries in the database and track the status of each document. This allows the review team to collate together documents of a particular group, e.g all “hot docs” or all “relevant documents”.
The granularity of this information will vary between review tools and the needs of the review team. For example some review teams may only need to know that a document has been marked relevant, others lawyers may need to know who marked it relevant, when, and has its status ever been changed.