Marty Kilmer, V.P. of Product Platforms, and Ian Koenig, Chief Architect, set out to give us an inside look at how the search algorithms deliver results within Lexis Advance. Kilmer also announced that the next major release of Lexis Advance is coming soon, including navigation improvements and adding source titles to the Word Wheel (a great addition in my opinion).
Before the discussion on the search algorithm, Koenig made two clarifications on searching issues:
Koenig then walked through the steps of a Lexis Advance search and how the search is analyzed by the algorithms in place.
The algorithm then ranks the results using the following criteria:
- The Word Wheel does not include associative retrieval (e.g., a query on "abortion" will not surface results for Roe v. Wade).
- Lexis Advance does perform word stemming but will not search alternative word forms. For example a query for "produce" (as in "produce documents") will not find instances of the term "production."
Unfortunately, there was no time for questions - and this seemed to be common with the Exhibitor Showcases that I attended. I think this is quite a shame and at future meetings I would like to see these sessions either lengthened or AALL should urge vendors to allow time for questions. Attendees were encouraged to visit their booths for follow-up questions, but I believe facing questions in front of an audience would have put their feet to the fire, so to speak, more so than talking with them one on one on their turf.
Specifically, I expected them to address a key issue that has come up in discussions with my fellow law librarians – that the Lexis Advance search algorithm - specifically as it applies to natural language searches - is simply not as strong as that of some of its competitors. It is interesting to know how it works, but I would have liked to know what they’re doing to make it work better.
- Phrase recognition
- Case names & citations
- Concentration of terms
- Coverage of terms
- Recentness (level of authority, validity)
- Document segment the search terms appear in
- Number of hits within the document of the search terms