Finding the privileged few: supporting privilege review for e-discovery

As typically conceived, the goal of information retrieval is to help people find what they seek. However, when sensitive content is intermixed with what is sought, it can be equally important to help protect what should not be seen. This paper focuses on one such case, the protection of content that is subject to a claim of attorney-client privilege when sharing evidence incident to civil litigation, a process called e-discovery. Although lawyers have been quick to embrace text classification techniques for finding relevant evidence, the stakes involved in inadvertent disclosure of privileged content create reluctance to trust any fully automated technique to accurately recognize content that can properly be withheld. Rather than starting with privilege classification as a goal, this paper proposes a modest first step for building tools that can help lawyers to make faster and more accurate privilege judgments by scoring the importance of specific email addresses to determine their propensity to engage in privileged communication. Both recursive and heuristic techniques are used to estimate that propensity score, ultimately resulting a coverage of 94% of the email addresses.