Document categorization in legal electronic discovery: computer classification vs. manual review

Abstract

Abstract In litigation in the US, the parties are obligated to produce to one another, when requested, those documents that are potentially relevant to issues and facts of the litigation (called “discovery”). As the volume of electronic documents continues to grow, the expense of dealing with this obligation threatens to surpass the amounts at issue and the time to identify these relevant documents can delay a case for months or years. The same holds true for government investigations and third‐parties served with subpoenas. As a result, litigants are looking for ways to reduce the time and expense of discovery. One approach is to supplant or reduce the traditional means of having people, usually attorneys, read each document, with automated procedures that use information retrieval and machine categorization to identify the relevant documents. This study compared an original categorization, obtained as part of a response to a Department of Justice Request and produced by having one or more of 225 attorneys review each document with automated categorization systems provided by two legal service providers. The goal was to determine whether the automated systems could categorize documents at least as well as human reviewers could, thereby saving time and expense. The results support the idea that machine categorization is no less accurate at identifying relevant/responsive documents than employing a team of reviewers. Based on these results, it would appear that using machine categorization can be a reasonable substitute for human review.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 57 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.