tris/recoll
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

when search includes composite spans + other terms, increase slack instead of switching to word split

Browse Source
This commit is contained in:
dockes 2007-10-04 12:26:04 +00:00
parent 90e378333e
commit 9b5de1a4ac
1 changed files with 17 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
src/rcldb/searchdata.cpp
View File

@ -1,5 +1,5 @@
#ifndef lint #ifndef lint
static char rcsid[] = "@(#$Id: searchdata.cpp,v 1.18 2007-09-20 08:43:12 dockes Exp $ (C) 2006 J.F.Dockes"; static char rcsid[] = "@(#$Id: searchdata.cpp,v 1.19 2007-10-04 12:26:04 dockes Exp $ (C) 2006 J.F.Dockes";
#endif #endif
/* /*
* This program is free software; you can redistribute it and/or modify * This program is free software; you can redistribute it and/or modify
@ -367,16 +367,19 @@ bool StringToXapianQ::processUserString(const string &iq,
LOGDEB(("strToXapianQ: phrase or word: [%s]\n", it->c_str())); LOGDEB(("strToXapianQ: phrase or word: [%s]\n", it->c_str()));
// If there are multiple spans in this element, including // If there are multiple spans in this element, including
// at least one composite, we need to use a word split, // at least one composite, we have to do something
// else a phrase query including a span would fail. // else a phrase query including a span would fail.
// (other possible solution: adjust slack to account for the // Ex: "term0@term1 term2" is onlyspans-split as:
// additional position increase?) // 0 term0@term1 0 12
// Ex: "term0@term01 term1" is onlyspans-split as: // 2 term2 13 18
// 0 term0@term01 0 12 // The position of term1 is 2, not 1, so a phrase search
// 2 term1 13 18 // would fail.
// The position of term1 is 2, not 1, so the phrase search would // We used to do word split, searching for
// fail. We search for "term0 term01 term1" instead, which may // "term0 term01 term1" instead, which may have worse
// have worse performance, but will succeed. // performance, but will succeed.
// We now adjust the phrase/near slack by the term count
// difference (this is mainly better for cjk where this is a very
// common occurrence because of the ngrams thing.
wsQData splitDataS(stops), splitDataW(stops); wsQData splitDataS(stops), splitDataW(stops);
TextSplit splitterS(&splitDataS, TextSplit splitterS(&splitDataS,
TextSplit::Flags(TextSplit::TXTS_ONLYSPANS | TextSplit::Flags(TextSplit::TXTS_ONLYSPANS |
@ -388,8 +391,10 @@ bool StringToXapianQ::processUserString(const string &iq,
splitterW.text_to_words(*it); splitterW.text_to_words(*it);
wsQData *splitData = &splitDataS; wsQData *splitData = &splitDataS;
if (splitDataS.terms.size() > 1 && if (splitDataS.terms.size() > 1 &&
splitDataS.terms.size() != splitDataW.terms.size()) splitDataS.terms.size() != splitDataW.terms.size()) {
splitData = &splitDataW; slack += splitDataW.terms.size() - splitDataS.terms.size();
// used to: splitData = &splitDataW;
}
LOGDEB(("strToXapianQ: splitter term count: %d\n", LOGDEB(("strToXapianQ: splitter term count: %d\n",
splitData->terms.size())); splitData->terms.size()));