Improved Document Search
Document search results are based partly on whether a document contains terms that are “conceptual matches” to the search query. “Conceptual match term” was previously defined as any of the twenty most-related terms to the query. In this release, we have changed the definition to additionally require that conceptual match terms have an association score of at least 0.6 with the search query, in order to exclude extraneous terms in cases where a query simply does not have many highly-related terms. Many terms will have the same conceptual matches as before, but some will have fewer or none.This new algorithm increases the accuracy of our document search (and corresponding match counts), providing better metrics for Share-of-Voice and other document-dependent analyses.
This change impacts old projects, new projects, and new dashboards. Old dashboard snapshots (published) are not impacted.
Luminoso Analytics and Luminoso Compass
Improved detection of multi-word phrases including capitalization
Previously, a group of words would be considered a more likely multi-word phrase if all of its component words were capitalized, e.g. "Kindle Fire" was a more likely multi-word phrase than "kindle fire," but "Arm and Hammer" was no more likely than "arm and hammer" because of the presence of the lowercased "and". Now, the presence of certain lowercased, insignificant words such as "and", "the", or "of" will no longer penalize the likelihood of a group of words being a multi-word phrase, e.g. "Arm and Hammer" will be considered just as likely to be a multi-word phrase as "Arm And Hammer". This only impacts new projects and recalculated projects.
Previously when detecting multi-word phrases, quotation marks would be treated as phrase boundaries. Now quotation marks can be included within a phrase. For example, if power button is a phrase, "power" button will also count as an instance of that phrase.This doesn't re-analyze documents in already-calculated projects, but it does happen in searches for all projects (old and new). For example, if power button is a phrase in an old project, nothing is going to go through all that project's documents and find instances of "power" button and call them phrases (until you recalculate), but if you do a search for "power" button, it will match documents containing power button.