MoreLikeThis
MoreLikeThis enables queries for documents similar to a document in their result list.
It does this by using terms from the original document to find similar documents in the index.
There are several ways to use MoreLikeThis. The first, and most common, is to use it as a request handler. In this case, you would send text to the MoreLikeThis request handler as needed (as in when a user clicked on a "similar documents" link).
The second is to use it as a search component. This is less desirable since it performs the MoreLikeThis analysis on every document that matches a user query. This may slow search results.
Another approach is to use it as a request handler but with externally supplied text. This case, also referred to as the MoreLikeThisHandler, will supply information about similar documents in the index based on the text of the input document.
Finally, the MLT query parser can be used. This operates in much the same way as the request handler but since it is a query parser it can be used in filter queries, boost queries, etc., and results can be paginated or highlighted as needed.
How MoreLikeThis Works
MoreLikeThis
constructs a Lucene query based on terms in a document.
It does this by pulling terms from the list of fields provided with the request.
For best results, the fields should have stored term vectors (termVectors=true
), which can be configured in the schema.
If term vectors are not stored, MoreLikeThis can generate terms from stored fields.
The field used for the uniqueKey
must also be stored in order for MoreLikeThis to work properly.
Terms from the original document are filtered using thresholds defined with the MoreLikeThis parameters. Once the terms have been selected, a query is run with any other query parameters as appropriate and a new document set is returned.
MoreLikeThis Handler and Component
The MoreLikeThis request handler and search component share several parameters, but also have some key differences in response and operation, as described below.
Common Handler and Component Parameters
The list below summarizes the MoreLikeThis
parameters supported by Solr.
These parameters can be used with the MoreLikeThis search component or request handler.
mlt.fl
-
Required
Default: none
Specifies the fields to use for similarity. A list of fields can be provided separated by commas. If possible, the fields should have stored
termVectors
. mlt.mintf
-
Optional
Default:
2
Specifies the minimum frequency below which terms will be ignored in the source document.
mlt.mindf
-
Optional
Default:
5
Specifies the minimum frequency below which terms will be ignored which do not occur in at least this many documents.
mlt.maxdf
-
Optional
Default: none
Specifies the maximum frequency above which terms will be ignored which occur in more than this many documents.
mlt.maxdfpct
-
Optional
Default: none
Specifies the maximum document frequency using a ratio relative to the number of documents in the index. The value provided must be an integer between
0
and100
. For example,mlt.maxdfpct=75
means the word will be ignored if it occurs in more than 75 percent of the documents in the index. mlt.minwl
-
Optional
Default: none
Sets the minimum word length below which words will be ignored.
mlt.maxwl
-
Optional
Default: none
Sets the maximum word length above which words will be ignored.
mlt.maxqt
-
Optional
Default:
25
Sets the maximum number of query terms that will be included in any generated query.
mlt.maxntp
-
Optional
Default:
5000
Sets the maximum number of tokens to parse in each example document field that is not stored with TermVector support.
mlt.boost
-
Optional
Default:
false
Specifies if the query will be boosted by the interesting term relevance. Possible values are
true
orfalse
. mlt.qf
-
Optional
Default: none
Query fields and their boosts using the same format used by the DisMax Query Parser. These fields must also be specified in
mlt.fl
. mlt.interestingTerms
-
Optional
Default:
none
Adds a section in the response that shows the top terms (based on TF/IDF) used for the MoreLikeThis query. It supports three possible values:
-
list
lists the terms. -
none
lists no terms (the default). -
details
lists the terms along with the boost value used for each term. Unlessmlt.boost=true
, all terms will haveboost=1.0
.
To use this parameter with the search component, the query cannot be distributed. In order to get interesting terms, the query must be sent to a single shard and limited to that shard only (with the
shards
parameter). Multi-shard support is, however, available with the MoreLikeThis request handler. -
MoreLikeThis Request Handler
Request Handler Configuration
The MoreLikeThis request handler is not configured by default and needs to be set up before using it.
You can do this by manually editing solrconfig.xml
or with the Config API:
Manual Configuration
<requestHandler name="/mlt" class="solr.MoreLikeThisHandler">
<str name="mlt.fl">body</str>
</requestHandler>
Config API
curl -X POST -H 'Content-type:application/json' -d {
"add-requesthandler": {
"name": "/mlt",
"class": "solr.MoreLikeThisHandler",
"defaults": {"mlt.fl": "body"}
}
} http://localhost:8983/solr/<collection>/config
Both of the above examples set the mlt.fl
parameter to "body" for the request handler.
This means that all requests to the handler will use that value for the parameter unless specifically overridden in an individual request.
For more about request handler configuration in general, see the section Default Components.
Request Handler Parameters
The MoreLikeThis request handler supports the following parameters in addition to the common parameters above. It supports faceting, paging, and filtering using common query parameters, but does not work well with alternate query parsers.
mlt.match.include
-
Optional
Default:
true
Specifies if the response should include the matched document. If set to
false
, the response will look like a normal select response. mlt.match.offset
-
Optional
Default: none
Specifies an offset into the main query search results to locate the document on which the MoreLikeThis query should operate. By default, the query operates on the first result for the
q
parameter.
Request Handler Query and Response
Queries to the MoreLikeThis request handler use the name defined when it was configured (/mlt
in the above example).
The following example query uses a document (q=id:0553573403
) found in Solr’s example document set (./example/exampledocs
), and asks that the author field be used to find similar documents (mlt.fl=author
).
http://localhost:8983/solr/gettingstarted/mlt?mlt.fl=author&mlt.interestingTerms=details&mlt.match.include=true&mlt.mindf=0&mlt.mintf=0&q=id%3A0553573403
This query also requests interesting terms with their boosts (mlt.interestingTerms=details
) and that the original document also be returned (mlt.match.include=true
).
The minimum term frequency and minimum word document frequency are set to 0
.
The response will include a section match
, which includes the original document.
The response
section includes the similar documents.
Finally, the interestingTerms
section shows the terms from the author field that were used to find the similar documents.
Because we did not also specify mlt.boost
, the boost values shown for the interesting terms all display 1.0
.
{
"match":{"numFound":1,"start":0,"numFoundExact":true,
"docs":[
{
"id":"0553573403",
"cat":["book"],
"name":["A Game of Thrones"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":1,
"genre_s":"fantasy",
"_version_":1693062911089442816}]
},
"response":{"numFound":2,"start":0,"numFoundExact":true,
"docs":[
{
"id":"0553579908",
"cat":["book"],
"name":["A Clash of Kings"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":2,
"genre_s":"fantasy",
"_version_":1693062911094685696},
{
"id":"055357342X",
"cat":["book"],
"name":["A Storm of Swords"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":3,
"genre_s":"fantasy",
"_version_":1693062911095734272}]
},
"interestingTerms":[
"author:r.r",1.0,
"author:george",1.0,
"author:martin",1.0]}
If we had not requested mlt.match.include=true
, the response would not have included the match
section.
Streaming External Content to MoreLikeThis
An external document (one not in the index) can be passed to the MoreLikeThis request handler to be used for recommended documents.
This is accomplished with the use of Content Streams.
The body of a document can be passed directly to the request handler with the stream.body
parameter.
Alternatively, if remote streams are enabled, a URL or file could be passed.
http://localhost:8983/solr/mlt?stream.body=electronics%20memory&mlt.fl=manu,cat&mlt.interestingTerms=list&mlt.mintf=0
This query would pass the terms "electronics memory" to the request handler instead of using a document already in the index.
The response in this case would look similar to the response above that used a document already in the index. Note: it doesn’t work for SolrCloud, check below for the possible solution.
MoreLikeThis Search Component
Using MoreLikeThis as a search component returns similar documents for each document in the response set for another query. It’s important to note this could incur a cost to search performance so should only be used when the use case warrants it.
Search Component Configuration
The MoreLikeThis search component is a default search component that works with all search handlers (see also Default Components).
Since it is configured already, it doesn’t need any additional configuration unless you’d like to set parameters for a particular collection that override the MoreLikeThis defaults. To do this, you could configure it like this:
<searchComponent name="mlt" class="solr.MoreLikeThisComponent">
<str name="mlt">true</str>
<str name="mlt.fl">body</str>
</searchComponent>
The above example would always enable MoreLikeThis for all queries and will always use the "body" field. This is probably not something you really want! But the example serves to show how you might define whichever parameters you would like to be default for MoreLikeThis.
If you gave the search component a name other than "mlt" as in the above example, you would need to explicitly add it to a request handler as described in the section Referencing Search Components. Because the above example uses the same name as the default, the parameters defined override Solr’s default.
Search Component Parameters
The MoreLikeThis search component supports the following parameters in addition to the common parameters above.
mlt
-
Optional
Default: none
If set to
true
, activates theMoreLikeThis
component and enables Solr to returnMoreLikeThis
results. mlt.count
-
Optional
Default:
5
Specifies the number of similar documents to be returned for each result.
Search Component Query and Response
The response when using MoreLikeThis as a search component is different than when using the request handler.
In this case, we are using the /select
request handler and performing a regular query (q=author:martin
).
We’ve asked for MoreLikeThis to be added to the response (mlt=true
), but otherwise the parameters are the same as the earlier example (we’ve asked for interesting terms and set minimum term and document frequencies to 0
).
http://localhost:8983/solr/gettingstarted/select?mlt.fl=name&mlt.mindf=0&mlt.mintf=0&mlt=true&q=author%3Amartin
The response includes the results of our query, in this case 3 documents which have the term "martin" in the author field.
We’ve changed the field, however, to find documents that are similar to these based on values in the name
field (mlt.fl=name
).
In the response, a moreLikeThis
section has been added.
For each document in the results that match our query, a list of document IDs is returned with score values.
Each of these documents are similar to the document in the result list to varying degrees.
{
"response":{"numFound":3,"start":0,"maxScore":0.43659902,"numFoundExact":true, "docs":[
{
"id":"0553573403",
"cat":["book"],
"name":["A Game of Thrones"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":1,
"genre_s":"fantasy",
"_version_":1693062911089442816},
{
"id":"0553579908",
"cat":["book"],
"name":["A Clash of Kings"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":2,
"genre_s":"fantasy",
"_version_":1693062911094685696},
{
"id":"055357342X",
"cat":["book"],
"name":["A Storm of Swords"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":3,
"genre_s":"fantasy",
"_version_":1693062911095734272}]
},
"moreLikeThis":[
"0553573403",{"numFound":6,"start":0,"maxScore":1.6554483,"numFoundExact":true,
"docs":[
{
"id":"055357342X",
"score":1.6554483},
{
"id":"0553579908",
"score":1.6554483},
{
"id":"0805080481",
"score":1.3422124},
{
"id":"0812550706",
"score":1.284826},
{
"id":"978-1423103349",
"score":0.7652973}]
},
"0553579908",{"numFound":5,"start":0,"maxScore":1.6554483,"numFoundExact":true,
"docs":[
{
"id":"055357342X",
"score":1.6554483},
{
"id":"0553573403",
"score":1.6554483},
{
"id":"0805080481",
"score":1.3422124},
{
"id":"978-1423103349",
"score":0.7652973},
{
"id":"VDBDB1A16",
"score":0.68205893}]
},
"055357342X",{"numFound":5,"start":0,"maxScore":1.6554483,"numFoundExact":true,
"docs":[
{
"id":"0553579908",
"score":1.6554483},
{
"id":"0553573403",
"score":1.6554483},
{
"id":"0805080481",
"score":1.3422124},
{
"id":"978-1423103349",
"score":0.7652973},
{
"id":"VDBDB1A16",
"score":0.68205893}]
}]}
MoreLikeThis Query Parser
The mlt
query parser provides a mechanism to retrieve documents similar to a specific document, like the request handler.
It uses Lucene’s existing MoreLikeThis
logic and also works in SolrCloud mode.
The document identifier used here is the document’s uniqueKey
value and not the Lucene internal document id.
The list of returned documents excludes the queried document.
One benefit of the query parser is that it can be used in various places, not only in a standard q
parameter.
This allows MoreLikeThis to be added to boost queries, filter queries, function queries, etc.
Query Parser Parameters
This query parser takes the following parameters:
qf
-
Required
Default: none
Defines the fields to use as the basis for similarity analysis.
mintf
-
Optional
Default:
2
Defines the minimum frequency below which terms will be ignored in the source document.
mindf
-
Optional
Default:
5
Defines the minimum frequency below which terms will be ignored which do not occur in at least this many documents.
maxdf
-
Optional
Default: none
Sets the maximum frequency above which terms will be ignored which occur in more than this many documents.
minwl
-
Optional
Default: none
Sets the minimum word length below which words will be ignored.
maxwl
-
Optional
Default: none
Sets the maximum word length above which words will be ignored.
maxqt
-
Optional
Default:
25
Sets the maximum number of query terms that will be included in any generated query.
maxntp
-
Optional
Default:
5000
Sets the maximum number of tokens to parse in each example document field that is not stored with TermVector support.
boost
-
Optional
Default:
false
Specifies if the query will be boosted by the interesting term relevance. It can be either
true
orfalse
.
Query Parser Query and Response
The structure of a MoreLikeThis query parser request is like a query using Local Params>, as in:
{!mlt qf=name}1
This would use the MoreLikeThis query parser to find documents similar to document "1", based on the "name" field.
Additional parameters would be added inside the brackets, for example if we wanted to specify limits for mintf
and mindf
:
{!mlt qf=name mintf=2 mindf=3}1
If given a query such as the following based on the example documents provided with Solr:
http://localhost:8983/solr/gettingstarted/select?q={!mlt qf=author mintf=1 mindf=1}0553573403
The query parser response includes only the similar documents sorted by score:
{
"response":{"numFound":2,"start":0,"maxScore":1.309797,"numFoundExact":true,
"docs":[
{
"id":"0553579908",
"cat":["book"],
"name":["A Clash of Kings"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":2,
"genre_s":"fantasy",
"_version_":1693062911094685696},
{
"id":"055357342X",
"cat":["book"],
"name":["A Storm of Swords"],
"price":[7.99],
"inStock":[true],
"author":["George R.R. Martin"],
"series_t":"A Song of Ice and Fire",
"sequence_i":3,
"genre_s":"fantasy",
"_version_":1693062911095734272}]
}}
Query Parser for External Content
Use {!mlt_content}lorem ipsum
or {!mlt_content q='lorem ipsum'}
to find docs similar to an external content absent in index in SolrCloud mode like /mlt
handler and content streams. Parameters and response are the same as above. It queries fields passed via qf
parameter with the given content. When qf
is omitted it queries all fields in the schema that usually fails on numerics and other specific formatted field types. If you need to query different fields with different content combine several {!mlt_content qf=fieldA}lorem ipsum
with {!bool}
query or other.