I'm new to SOLR and am implementing it to search our product catalog. I'm creating ngrams and edge ngrams on the brand name, display name and category fields.
I'm using edismax and have qf defined as displayname_nge displayname_ng category_nge category_ng brandname_nge brandname_ng.
When I search for 'vitamin c' (without the quotes) I get all of the vitamins. If I surround it with quotes then I only get vitamin c. The problem is that I can't always surround the query string with quotes because a person might enter 'chewable vitamin c', or 'vendor x vitamin c'. I've tried the mm parameter without luck. I've also tried applying different boost levels and still not getting the expected results.
Any suggestions would be greatly appreciated. Thank you
Was there a reason for using only ngrams fields for searching? I'm not sure this is the problem in your case, but you may want to look at your ngrams analysis configuration in schema.xml. One from one of my indexes looks like this:
<fieldType name="ngram" class="solr.TextField" >
<analyzer type="index">
<filter class="solr.WordDelimiterFilterFactory" generateWordParts="1" generateNumberParts="1" catenateWords="1" catenateNumbers="1" catenateAll="0" splitOnCaseChange="1"/>
<tokenizer class="solr.LowerCaseTokenizerFactory"/>
<filter class="solr.EdgeNGramFilterFactory" minGramSize="2" maxGramSize="15" side="front"/>
</analyzer>
<analyzer type="query">
<filter class="solr.WordDelimiterFilterFactory" generateWordParts="1" generateNumberParts="1" catenateWords="1" catenateNumbers="1" catenateAll="0" splitOnCaseChange="1"/>
<tokenizer class="solr.LowerCaseTokenizerFactory"/>
</analyzer>
</fieldType>
Though you can see this is actually using the safer EdgeNGramFilterFactory, the important thing to note here is minGramSize="2". This means that during the indexing process only grams of at least two characters will be created. The word 'c'? That doesn't get any grams at all. While you could set minGramSize="1" and rebuild your index, single character grams are a very bad idea, as your search for 'c' would match against any document with a word that starts with 'c' (or contains the letter 'c' with NGramFilterFactory).
If you're currently using NGrams with minGramSize="2", a search for 'ca' would find any documents with any words containing the letters 'ca' consecutively in that order. This may not be exactly what you want, either.
My top suggestion would be to drop the ngrams in favor of a more vanilla Text field. Whether you want to keep the edge-ngrams around for better truncation support is up to you, but I suspect you'll have better luck if the Text field is at least in the mix.
You could also take a look at this question on StackOverflow: "Can I protect short words from an n-gram filter in Solr?" if you want to pursue the ngrams further.
Also, you should consider using Solr's built-in analysis tool to figure out where your searches are failing. You choose a field or fieldType, and provide values for what was entered into the index and what is being searched. It will show you how the analysis works against both values so you can see how each string is broken down and why it does or doesn't create matching tokens. The URL for the tool depends on whether you're in a multi-core environment, but if you go to Solr's web interface you should be able to find the Analysis link on the left.
Update:
Now that I have a little more detail from you and am thinking about it again, the results you're getting are very explainable.
With minGramSize="1", your unquoted search for 'vitamin c' is looking for records with the word 'vitamin' (or a longer word containing 'vitamin'), and the word 'c' (or a longer word containing 'c'). Since most records are likely to have a 'c' somewhere, this is hardly a limiting factor and your results will be very close to or exactly the same as your results for just the word 'vitamin'.
In the quoted search for 'vitamin c', the 'c' now has to appear in a word immediately following vitamin, making it a much more useful search, but still not great. You should be able to test this by finding records that have a word following vitamin that isn't a vitamin designation. For example, a record mentioning "vitamin tablets" should be found when searching for "vitamin b" (because there's a 'b' in "tablets"). and a record mentioning "vitamin chart" or "vitamin deficiency" should be found when searching for "vitamin c".
The upshot of this is that I strongly recommend having a set of fields for searching separate from your fields for autocomplete. The NGrams with minGramSize="1" are just not going to give you reasonable results for the actual search step.
Other option is to use edismax - 'mm', there you can give matching %. if you give 100% it will give you accurate matching. 75% will give you list of vitamin... you can programatically handle % according to your need
You may consider to replace the query keyword this way: "'vitamin c' vitamin c". In such case, records matching 'vitamin c' can get higher score than those matching 'vitamin' and 'c' separately. Your search results will still return all matching records. Please see if this help, and feel free to comment.
Related
I created my own core on http://localhost:8983/solr and added some documents so I could query. But When I query something like"dog", I want those documents that contains "pooch" will be returned too. So I want to implement SVD algorithm to make some improvement on my results.
Since I am new to the search engine thing. All I know is that I can use Mahout to implement SVD, but it seems a little bit difficult coz I have to install Maven, Hadoop and Mahout.
Any suggestion will be appreciated.
You can use SynonymGraphFilterFactory
This filter maps single- or multi-token synonyms, producing a fully correct graph output. This filter is a replacement for the Synonym Filter, which produces incorrect graphs for multi-token synonyms.
If you use this filter during indexing, you must follow it with a Flatten Graph Filter to squash tokens on top of one another like the Synonym Filter.
Create a file i.e mysynonyms.txt in the directory your_collection/conf/ and put the synonyms with => sign
pooch,pup,fido => dog
huge,ginormous,humungous => large
And Example Schema will be :
<analyzer type="index">
<tokenizer class="solr.StandardTokenizerFactory"/>
<filter class="solr.SynonymGraphFilterFactory" synonyms="mysynonyms.txt"/>
<filter class="solr.FlattenGraphFilterFactory"/> <!-- required on index analyzers after graph filters -->
</analyzer>
<analyzer type="query">
<tokenizer class="solr.StandardTokenizerFactory"/>
<filter class="solr.SynonymGraphFilterFactory" synonyms="mysynonyms.txt"/>
</analyzer>
Source : https://cwiki.apache.org/confluence/display/solr/Filter+Descriptions
The is another way to augment your index with terms not in the content. Synonyms is good as #ashraful says. But there are 2 other problems you will run into:
words used but not in the synonym list
behavioral search: using other user behavior as a hint to what they are looking for
These require you to augment the index with terms learned from 1) other searches, and 2) user behavior. Mahout's Correlated Cross Occurrence algorithm can help with both. You can set it up to find terms that lead to people reading an item and (if you have something like purchase or other preference data) conversion items that correlate with items in the index. In the second case you would add user conversions to the search query to personalize the results.
A blog about the technique here: http://actionml.com/blog/personalized_search
The page on Mahout docs here: http://mahout.apache.org/users/algorithms/intro-cooccurrence-spark.html
You should also look at word2vec, which will (given the right training data) find that "dog" and "pooch" are synonyms regardless of the synonym list because it is learned from the data. I'm not sure how you add word2vec to Solr but it is integrated into Fusion, the closed source product of Lucid.
Folks,
We wanted to make a search on solr such that it will give a priority to partial match in the sentences.
Lets say for example :
Sentence is like "Have wonderful evening today here"
If user is supplying "today here" then it should match.
If user is supplying "wonderful evening" then it should match.
If user is supplying "Have wonderful" then it should match.
We want to give low priority to key word search compared to above.
keyword match could be : "today" "wonderful" "evening" etc.
Is there any way this can be achieve is solr since solr works on inverted index of words on a given sentence.
You can use a separate field with a SingleFilter defined - this will combine runs of tokens into separate tokens, so that "Have wonderful evening today here" can be indexed as "have wonderful", "wonderful evening", "evening today" and "today here".
Make hits in this field a higher priority than hits in your regular search field by using qf=shinglefield^<boostvalue> - what the exact boost value needs to be depends on the scoring profile of your index and if you're doing other boosts.
<analyzer>
<tokenizer class="solr.WhitespaceTokenizerFactory"/>
<filter class="solr.LowerCaseFilterFactory"/>
<filter class="solr.ShingleFilterFactory" maxShingleSize="2" outputUnigrams="false"/>
</analyzer>
I am new to the solr i have a document indexed in solr e.g
{
"foodType": "basicFood",
"fulltext": [
"basicFood",
"3.718625",
"1 tbsp",
"Butter, salted"
],
"slims": "3.718625",
"displayText": "1 tbsp",
"displayName": "Butter, salted"
},
when i search for butter the result is null but it works fine for the query butter, how to make it working for butter also?
Add the following filter to your analyzer for both index and query phase.
<filter class="solr.PatternReplaceFilterFactory" pattern="([^A-Za-z0-9])" replacement="" replace="all"/>
This is will replace all the letters from token except a-z, A-Z and 0-9. To test this you might need to re-index your data, because your indexed data will have punctuations. or you can try in solr admin UI, analysis section.
Other approach, using a different tokenizer instead of StandardTokenizerFactory in the analyzer phase. You can use LetterTokenizerFactory which creates tokens consisting of strings of contiguous letters. Any non-letter characters will be discarded. But this can create many extra token which you might not want. Please check before you do.
Example: "I can't" ==> "I", "can", "t"
Update
If you need to use WordDelimiterFilter filter. Try the below configuration:
<filter class="solr.WordDelimiterFilter" generateWordParts="1" catenateWords="1" splitOnCaseChange="0" generateNumberParts="0" splitOnCaseChange="0"/>
This will splits words at delimiters and concatenate them again. All the other splits are turned off like numeric strings, camel-case change and transitions from alpha to numeric. If required you can turn on by providing non-zero value.
I have a solr setup(1.4) having a text field with ebook data. The params while hitting solr are -
"hl.fragsize":"0",
"indent":"1",
"hl.simple.pre":"{{{",
"hl.fl":"body_eng",
"hl.maxAnalyzedChars":"-1",
"wt":"json",
"hl":"true",
"rows":"1",
"fl":"ia,body_length,page_count",
"q":"ia:talesofpunjabtol00stee AND PUNJAB",
"q.op":"AND",
"f.body_eng.hl.snippets":"428",
"hl.simple.post":"}}}",
"hl.usePhraseHighlighter":"true"}},
However, the results show only 20 highlighted occurrences of word PUNJAB.
I tried f.body_eng.hl.snippets":"428" but this even isnt working.
body_eng is a big text field. The highlighting works only till some length. I have tried other words as well. In all the examples, highlighting works till around 54K letter counts.
What could be the reason?
First of all: 1.4 is a very old version of Solr. I'm not sure if per field values were supported at that time (Highlighting itself was introduced with Solr 1.3). The default highlighter was changed in 3.1.
You should however be able to highlight all occurences in a field by supplying a large value for hl.maxAnalyzedChars (not sure if -1 will do what you want). Another option to try should be to have a large hl.maxAnalyzedChars value and a large hl.fragsize value (use the same value for both fields and not 0).
If you're still unable to get it to work, test it on a more recent version of Solr to see if it's an issue that has already been fixed.
So, after lot of asking around, Its working now.
The query params is correct. The schema was causing problems. Changes done were -
<filter class="solr.SnowballPorterFilterFactory" language="English" />
was replaced with
with <filter class="solr.StopFilterFactory" ignoreCase="true" words="stopwords.txt" enablePositionIncrements="true" />
I'm trying to set up a fullname search in Solr. Until now I thought my work was fine until I've found something strange, and I can't figure out how to correct it.
So I want to be able to do searches on fullnames. My index is a database where I get first name and last name and put them in one multivalued field with keyword tokenizer.
Here's my fieldtype :
<fieldType name="text_auto" class="solr.TextField">
<analyzer>
<tokenizer class="solr.KeywordTokenizerFactory"/>
<filter class="solr.ASCIIFoldingFilterFactory"/>
<filter class="solr.LowerCaseFilterFactory"/>
</analyzer>
</fieldType>
Everything works fine, I can search only a first name OR lastname and it gives me the full names that exists, and it also works for full names in any order if there's no mispelling.
I just noticed something wrong ! For example, if I ask for Dupont dupont, it'll give me every Dupont that exists, even the ones for which the first name doesn't match with dupont. I guess it's because dup is found a second time in the fullname... The problem is that if they're looking for "dupont d", they'll find every Dupont that exist because "d" is contained in Dupont ! That's not what I want, I want to find every Dupont with a d in their first name (the other string).
So I need to find a way to make it work, I tried many different tokenizers and filters but I'm affraid it's not possible...
Thank you for any help you could provide me !
Sounds like you are searching with something like:
q=dupont d
Which will have no problem with finding the terms in any order, or even as the same term in the index, in the case of dupont dupont (I'm assuming, by the way, that you are setting the default operator to AND, since this sort of behavior is surprising). If you want to find the phrase "dupont d" in that order, you should search with a quoted phrase query:
q="dupont d"
or for dupont dupont
q="dupont dupont"