i had to create a autosuggest implementation not too long ago, originally i was using faceting, where i would match wildcards on a tokenized field and facet on an unaltered field, this had the advantage that i could do everything from one index, though it was also limited by the fact suggestions came though facets and scoring and highlighting went out the window
what i settled on was to create a separate core for suggest to use, i analyze the fields i want to match against with whitespace tokenizer and edgengram filter, this has multiple advantages: query is ran through text analysis where as with wildcarded terms they are not highlighter will highlight only the text matched not the expanded word scoring and boosts can be used to rank suggest results i tokenize on whitespace so i can match out of order tokens , ex q=family guy stewie and q=stewie family guy, etc, this is something that prefix based solutions wont be able to do, one small gotcha is that i recently submitted a patch to edgengram filter to fix highlighting behaviour, it has been comitted to lucenes trunk but its only available in versions 2.9.2 and up unless you patch it yourself On Wed, Feb 24, 2010 at 7:35 AM, Grant Ingersoll <gsing...@apache.org> wrote: > You might also look at http://issues.apache.org/jira/browse/SOLR-1316 > > On Feb 24, 2010, at 1:17 AM, Sachin wrote: > >> >> >> Hi All, >> >> I am trying to setup autosuggest using solr 1.4 for my site and needed some >> pointers on that. Basically, we provide autosuggest for user typed in >> characters in the searchbox. The autosuggest index is created with older >> user typed in search queries which returned > 0 results. We do some lazy >> writing to store this information into the db and then export it to solr on >> a nightly basis. As far as I know, there are 3 ways (apart from wild card >> search) of achieving autosuggest using solr 1.4: >> >> 1. Use EdgeNGrams >> 2. Use shingles and prefix query. >> 3. Use the new Terms component. >> >> I am for now more inclinded towards using the EdgeNGrams (no method to >> madness) and just wanted to know is there any recommended approach out of >> the 3 in terms of performance, since the user excepts the suggestions to be >> almost instantaneous? We do some heavy caching at our end to avoid hitting >> solr everytime but is any of these 3 approaches faster than the other? >> >> Also, I would also like to return the suggestion even if the user typed in >> query matches in between: for instance if I have the query "chicken pasta" >> in my index and the user types in "pasta", I would also like this query to >> be returned as part of the suggestion (ala Yahoo!). Below is my field >> definition: >> >> <fieldType name="suggest" class="solr.TextField" >> positionIncrementGap="100"> >> <analyzer type="index"> >> <tokenizer class="solr.KeywordTokenizerFactory"/> >> <filter class="solr.LowerCaseFilterFactory"/> >> <filter class="solr.EdgeNGramFilterFactory" minGramSize="2" >> maxGramSize="50" /> >> </analyzer> >> <analyzer type="query"> >> <tokenizer class="solr.KeywordTokenizerFactory"/> >> <filter class="solr.LowerCaseFilterFactory"/> >> </analyzer> >> </fieldType> >> >> >> I tried changing the KeywordTokenizerFactory with LetterTokenizerFactory, >> and though it works great for the above scenario (does a in-between match), >> it has the side-effect of removing everything which are not letters so if >> the user types in "123" he gets absolutely no suggestions. Is there anything >> that I'm missing in my configuration, is this even achievable by using >> EdgeNGrams or shall I look at using perhaps the TermsComponent after >> applying the regex patch from 1.5 and maybe do something like >> ".*user-typed-in-chars.*"? >> >> Thanks! >> >> >> > > >
