Text scoring in scripts

edit

Text scoring in scripts

edit

Text features, such as term or document frequency for a specific term can be accessed in scripts (see scripting documentation ) with the _index variable. This can be useful if, for example, you want to implement your own scoring model using for example a script inside a function score query. Statistics over the document collection are computed per shard, not per index.

Nomenclature:

edit

df

document frequency. The number of documents a term appears in. Computed per field.

tf

term frequency. The number times a term appears in a field in one specific document.

ttf

total term frequency. The number of times this term appears in all documents, that is, the sum of tf over all documents. Computed per field.

df and ttf are computed per shard and therefore these numbers can vary depending on the shard the current document resides in.

Shard statistics:

edit
_index.numDocs()
Number of documents in shard.
_index.maxDoc()
Maximal document number in shard.
_index.numDeletedDocs()
Number of deleted documents in shard.

Field statistics:

edit

Field statistics can be accessed with a subscript operator like this: _index['FIELD'].

_index['FIELD'].docCount()
Number of documents containing the field FIELD. Does not take deleted documents into account.
_index['FIELD'].sumttf()
Sum of ttf over all terms that appear in field FIELD in all documents.
_index['FIELD'].sumdf()
The sum of df s over all terms that appear in field FIELD in all documents.

Field statistics are computed per shard and therfore these numbers can vary depending on the shard the current document resides in. The number of terms in a field cannot be accessed using the _index variable. See word count mapping type on how to do that.

Term statistics:

edit

Term statistics for a field can be accessed with a subscript operator like this: _index['FIELD']['TERM']. This will never return null, even if term or field does not exist. If you do not need the term frequency, call _index['FIELD'].get('TERM', 0) to avoid uneccesary initialization of the frequencies. The flag will have only affect is your set the index_options to docs (see mapping documentation).

_index['FIELD']['TERM'].df()
df of term TERM in field FIELD. Will be returned, even if the term is not present in the current document.
_index['FIELD']['TERM'].ttf()
The sum of term frequencys of term TERM in field FIELD over all documents. Will be returned, even if the term is not present in the current document.
_index['FIELD']['TERM'].tf()
tf of term TERM in field FIELD. Will be 0 if the term is not present in the current document.

Term positions, offsets and payloads:

edit

If you need information on the positions of terms in a field, call _index['FIELD'].get('TERM', flag) where flag can be

_POSITIONS

if you need the positions of the term

_OFFSETS

if you need the offests of the term

_PAYLOADS

if you need the payloads of the term

_CACHE

if you need to iterate over all positions several times

The iterator uses the underlying lucene classes to iterate over positions. For efficiency reasons, you can only iterate over positions once. If you need to iterate over the positions several times, set the _CACHE flag.

You can combine the operators with a | if you need more than one info. For example, the following will return an object holding the positions and payloads, as well as all statistics:

`_index['FIELD'].get('TERM', _POSITIONS | _PAYLOADS)`

Positions can be accessed with an iterator that returns an object (POS_OBJECT) holding position, offsets and payload for each term position.

POS_OBJECT.position
The position of the term.
POS_OBJECT.startOffset
The start offset of the term.
POS_OBJECT.endOffset
The end offset of the term.
POS_OBJECT.payload
The payload of the term.
POS_OBJECT.payloadAsInt(missingValue)
The payload of the term converted to integer. If the current position has no payload, the missingValue will be returned. Call this only if you know that your payloads are integers.
POS_OBJECT.payloadAsFloat(missingValue)
The payload of the term converted to float. If the current position has no payload, the missingValue will be returned. Call this only if you know that your payloads are floats.
POS_OBJECT.payloadAsString()
The payload of the term converted to string. If the current position has no payload, null will be returned. Call this only if you know that your payloads are strings.

Example: sums up all payloads for the term foo.

termInfo = _index['my_field'].get('foo',_PAYLOADS);
score = 0;
for (pos : termInfo) {
    score = score + pos.payloadAsInt(0);
}
return score;

Term vectors:

edit

The _index variable can only be used to gather statistics for single terms. If you want to use information on all terms in a field, you must store the term vectors (set term_vector in the mapping as described in the mapping documentation). To access them, call _index.getTermVectors() to get a Fields instance. This object can then be used as described in lucene doc to iterate over fields and then for each field iterate over each term in the field. The method will return null if the term vectors were not stored.