Posts tagged fuzzy
Analyzing Match Scores in 'fuzzy searches' – explaining the shades of gray in a non-binary world
In a previous article – Not all has to be black and white in SQL Queries – , I discuss how not all searches are about perfect matches. The search for your lifelong partner, or your next job are fine examples of searches where a perfect match is hard to obtain. No single person or no single job will score one 100% on all your criteria. Unfortunately, such binary searches is exactly what SQL is good at. WHERE ALL CRITERIA ARE MET is the filter SQL applies to the records inspected. Of course you can use OR operators to leave some room for non-perfect matches and instead of using the equality (=) you can go for between or LIKE. But a structured approach to ‘fuzzy’ searches where the shades of gray most searches inevitably end up in is not readily available.
The prequel to this articles takes a few first steps in illustrating how scores against different criteria can be collected, how different weight factors for search criteria can be taken into account and how the first two search-match-patterns (number range, discrete values) started to emerge. One major question we ended that article with: how can we explain when the search is performed and the match scores for each record is More >
Not all has to be black and white in SQL Queries: returning match scores instead of only perfect fits
Recently I sat in on a very interesting presentation on an advanced search and matching engine called Elise. The power of this engine lies in the fact that it does not just query records, returning the set of records that satisfy the search criteria – although it has some pretty advanced search filters and operators, but is able to score all records according to the match criteria. Every record gets a score – which can be zero, meaning that not a single search criteria matched even the smallest way. Higher scores mean better matches on more, or on the most important, criteria.
Many search operations do not necessarily need perfect matches. They need possibly multiple results that are all matches to a certain degree. Google for example returns web pages that meet your search criteria to a certain degree. The most relevant matches are at the top of the search results – or at least that is the intention. Searching for interesting job vacancies, holiday destinations, new real estate, electronic devices, dating partners on a e-dating site, books on Oracle technology are other examples of searches that look for best matches given criteria, not necessarily perfects fits for all criteria More >
Recent Comments