These properties affect how
LIKEP and some
queries are processed.
likeprowsrelevant documents are returned by a
LIKEPquery (default 100). This is an arbitrary cut-off beyond which most results would be increasingly useless. It also speeds up the query process, because fewer rows need to be sorted during ranking. By altering
likeprowsthis threshold can be changed, e.g. to return more results to the user. Setting this to 0 will return all relevant documents (no limit).
LIKEPqueries. This can be either 0, for early, or 1 for late. The default is 1, which is the correct setting for almost all cases. Does not apply to most Metamorph index searches.
LIKEPto only consider those documents containing all query terms as valid hits (i.e. just as
LIKEdoes). By default, since
LIKEPis a ranking operator it returns the best results even if only some of the set-logic terms (non-
-prefix) can be found. (Note that required terms - prefixed with a
+- are always required in a hit regardless of this setting.)
LIKEPto obey the intersects operator (
@) in queries. By default
LIKEPdoes not use it, because it is a ranking operator. Setting both
likepobeyintersectsto 1 will make
LIKEPrespect queries the same as
alintersectsmay have to be enabled in Vortex as well.)
LIKEPqueries: if the estimated number of matching rows for a set is greater than this, the set is considered infinitely-occurring. If all the search terms found in a given document are such infinite sets, the document is given an estimated rank. This saves time ranking irrelevant but often-occurring matches, at the possible expense of rank position. The default is 0, which means infinite (no infinite sets; rank all documents).
LIKE. After this many matches have been found, stop and return the results obtained so far, even if more hits exist. Typically this would be set to a high threshold (e.g. 100000): a query that returns more than that many hits is probably not specific enough to produce useful results, so save time and don't process the remaining hits. (It's also a good bet that something useful was already found in the initial results.) This helps keep such noisy queries from loading a server, by stopping processing on them early. A more specific query that returns fewer hits will fall under this threshold, so all matches will be considered for ranking.
Note that setting
likepindexthresh is a tradeoff between
speed and accuracy: the lower the setting, the faster queries can be
processed, but the more queries may be dropping potentially