[
https://issues.apache.org/jira/browse/SOLR-1599?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Graham Poulter updated SOLR-1599:
---------------------------------
Description:
In Solr 1.4, the IDF (Inverse Document Frequency) is calculated on all of the
documents in an index. This introduces relevance problems when using a single
schema to store multiple entity types, for example to support "search for
tracks" and "search for artists". The ranking for search on the _name_ field
of _track_ entities will be (much?) more accurate if the IDF for the name field
does not include counts from _artist_ entities. The effect on ranking would be
most pronounced for query terms that have a low document frequency for _track_
entities but a high frequency for _artist_ entities.
The current work-around to make the IDF be entity-specific is to use a separate
Solr core for each entity type sharing the schema - and repeating the process
of copying solrconfig.xml and schema.xml to all the cores. This would be more
complicated with replication, and even more complicated with index
distribution, because you must now maintain a core for _artists_ and a core for
_tracks_ on each node.
David Smiley, author of "Solr 1.4 Enterprise Search Server", has filed
SOLR-1158, where he suggests calculating _numDocs_ after the application of
filters. He recognises however that the document frequency (DF_t) for each
query term in a _track_ search would also needs to exclude _artist_ entities
from the DF_t total to get the correct IDF_t=log(N/DF_t). DF_t must be
calculated at index time, when Solr does not know what filters will be applied.
I suggest using a metadata field _entitytype_ specified on submitting a batch
of documents, where the schema specifies the list of allowed entity types. In
the example the document could specify either entitytype="track" or
entitytype="artist" (defaulting to _track_). During indexing each entity type
has its set of document frequencies, so the term "foo" will have a DF for
entitytype="artist" and a different DF for entitytype="track". This might be
implemented by instantiating a separate Lucene index for each configured entity
type. Filtering on entitytype="artist" would then be implemented by searching
only the _artist_ index, analogous to searching only on the _artist_ core in
the multi-core workaround.
With this solution (entity type metadata field implemented with separate Lucene
indeces) a single Solr core can support many different entity types that share
a common schema but use partially overlapping subsets of fields, instead of
having to configure maintain, replicate and distribute for every entity type.
was:
In Solr 1.4, the IDF (Inverse Document Frequency) is calculated on all of the
documents in an index. This introduces relevance problems when using a single
schema to store multiple entity types, for example to support "search for
tracks" and "search for artists". The ranking for search on the _name_ field
of _track_ entities will be (much?) more accurate if the IDF for the name field
does not include counts from _artist_ entities. The effect on ranking would be
most pronounced for query terms that have a low document frequency for _track_
entities but a high frequency for _artist_ entities.
The current work-around to make the IDF be entity-specific is to use a separate
Solr core for each entity type sharing the schema - and repeating the process
of copying solrconfig.xml and schema.xml to all the cores. This would be more
complicated with replication, and even more complicated with index
distribution, because you must now maintain a core for _artists_ and a core for
_tracks_ on each node.
David Smiley, author of "Solr 1.4 Enterprise Search Server", has filed
SOLR-1158, where he suggests calculating _numDocs_ after the application of
filters. However, _numDocs_ is just the total number of documents: the
document frequency (DF) for a query term of a _track_ search would also need to
exclude _artist_ entities from the DF_t total to get the IDF_t=log(N/DF_t).
However, DF_t needs to be calculated at index time, when Solr has no idea what
filters will be applied.
I suggest using a metadata field _entitytype_ to specified on submitting a
batch of documents, with a configured list of allowed values: in the example
the document could specify either entitytype="track" or entitytype="artist"
(defaulting to _track_). The document frequency would then calculated for each
entity type during indexing. so for term "foo" there will be two DF's stored:
the DF of "foo" for entitytype="artist" and the DF of "foo" for
entitytype="track". This might be implemented by instantiating a separate
Lucene index for each configured entity type. Filtering on entitytype="artist"
would then be implemented by searching only the _artist_ index.
With this solution (entity type metadata field implemented with separate Lucene
indeces) a single Solr core can support many different entity types that share
a common schema but use partially overlapping subsets of fields, instead of
having to configure maintain, replicate and distribute for every entity type.
> Improve IDF and relevance by separately indexing different entity types
> sharing a common schema
> -----------------------------------------------------------------------------------------------
>
> Key: SOLR-1599
> URL: https://issues.apache.org/jira/browse/SOLR-1599
> Project: Solr
> Issue Type: New Feature
> Components: Schema and Analysis
> Reporter: Graham Poulter
> Original Estimate: 504h
> Remaining Estimate: 504h
>
> In Solr 1.4, the IDF (Inverse Document Frequency) is calculated on all of the
> documents in an index. This introduces relevance problems when using a
> single schema to store multiple entity types, for example to support "search
> for tracks" and "search for artists". The ranking for search on the _name_
> field of _track_ entities will be (much?) more accurate if the IDF for the
> name field does not include counts from _artist_ entities. The effect on
> ranking would be most pronounced for query terms that have a low document
> frequency for _track_ entities but a high frequency for _artist_ entities.
> The current work-around to make the IDF be entity-specific is to use a
> separate Solr core for each entity type sharing the schema - and repeating
> the process of copying solrconfig.xml and schema.xml to all the cores. This
> would be more complicated with replication, and even more complicated with
> index distribution, because you must now maintain a core for _artists_ and a
> core for _tracks_ on each node.
> David Smiley, author of "Solr 1.4 Enterprise Search Server", has filed
> SOLR-1158, where he suggests calculating _numDocs_ after the application of
> filters. He recognises however that the document frequency (DF_t) for each
> query term in a _track_ search would also needs to exclude _artist_ entities
> from the DF_t total to get the correct IDF_t=log(N/DF_t). DF_t must be
> calculated at index time, when Solr does not know what filters will be
> applied.
> I suggest using a metadata field _entitytype_ specified on submitting a batch
> of documents, where the schema specifies the list of allowed entity types. In
> the example the document could specify either entitytype="track" or
> entitytype="artist" (defaulting to _track_). During indexing each entity
> type has its set of document frequencies, so the term "foo" will have a DF
> for entitytype="artist" and a different DF for entitytype="track". This
> might be implemented by instantiating a separate Lucene index for each
> configured entity type. Filtering on entitytype="artist" would then be
> implemented by searching only the _artist_ index, analogous to searching only
> on the _artist_ core in the multi-core workaround.
> With this solution (entity type metadata field implemented with separate
> Lucene indeces) a single Solr core can support many different entity types
> that share a common schema but use partially overlapping subsets of fields,
> instead of having to configure maintain, replicate and distribute for every
> entity type.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
