An Explanation of Match Scores

To fully understand how to use the TIBCO Patterns servers' search technology, it's important to have a basic understanding of what the match score returned by the search engine does and does not mean.

What the match score is not is a percentage of the query found in the record. In other words, a match score of 0.2 does not mean that only 20% of the text in the query string was found in the record text.

The match score is an estimate of the likelihood (or probability) that a given record matches the query, based solely on the record and the query.

Note: This is not to be confused with probabilistic matching techniques used by other search engines. Probabilistic matching produces a match score based on a relative probability distribution function built from the database contents, meaning that the score of a single record also depends on the contents of all other records. The TIBCO Patterns servers uses an absolute probability distribution function which does not change as database contents change.

 

This means that it is not a good idea to arbitrarily cut off search results at a fixed match score value as is commonly done with percent-of-query match scores. A match score of 0.2 means that it is 20% likely that the record relates to the query, which is still a fairly high probability and records such as this should not be discarded unless there are large numbers of records with higher match scores.

This is why the TIBCO Patterns servers returns a fixed number of matches by default rather than returning matches above a static score threshold. There are several dynamic score cutoffs that are used while searching. See Dynamic Score Cutoffs for more information.