Post by Darla Jackson, OBA Practice Management Advisor

Darla JacksonAlgorithms greatly affect the results produced by legal research tools. While legal information professionals have understood this for some time, only recently has an independent study documented the variance in legal research results produced by the unique algorithms employed by different research platforms.  A report of the study, conducted by Susan Nevelow Mart, The Algorithm as a Human Artifact: Implications for Legal {Re}Search is available on SSRN (or an abbreviated comparison of results produced by the algorithms utilized by a number of research tools is available at https://www.aallnet.org/sections/rips/pdfs/21st-Legal-Research-Teach-In/Handouts-and-Guides/Algorithm-Comparison).

Despite the influence that algorithms have on legal research, most legal research tools have released minimal information about their proprietarily developed algorithms, which has been called a lack of “Algorithmic Accountability.” Understanding the Technical Bias of Westlaw, Lexis Advance, Fastcase, Google Scholar, and Casetext, Three Geeks and A Law Blog (12/12/16)

Not only have legal researchers had limited information about the algorithms used by these “black box” systems, the researchers have not been able to customize algorithms to contextualize their research. Fastcase 7 now allows users to perform a customization of the relevance algorithm.

On the Advanced Search display in Fastcase 7, if you check the option to "Customize Relevance Algorithm" the customization tool allows you can customize the relevance algorithm by adjusting the sliders for seven relevance factors.  These factors are the Search Relevance Score (this scores each document based on the numerosity, proximity, diversity, and density of your keywords); Large Document Relevance (This shows the relevance for larger documents); Small Document Relevance (this shows the relevance for very short documents, such as appeals that simply affirm the lower court); Authoritativeness (this scores judicial opinions by how many times they have been cited); Frequently Read (This machine learning tool prioritizes documents frequently read in the Fastcase service); Frequently Printed (this machine learning tool prioritizes documents frequently printed in Fastcase); and Frequently E-mailed (this machine learning tool prioritizes documents frequently e-mailed in Fastcase).

I must admit that in my brief non-scientific trial of adjusting the relevance algorithm, I did not observe any significant variation in the search results produced. However, I hope to engage with Fastcase to discuss the weighting and application of the customizable factors.  For now, some transparency and customization does help me at least feel somewhat more in control.

Fastcase algorithm