Entity NER-specific Parameters

The table below describes the entity parameters you can set for NER when creating a Lexicon (an entity index).

Parameter	Description	Default value
NLU.NER.EntityIndex.AlternativeCalculationForProbability	An alternative method for computing the probability of an entity, that favors exact matches within the index. Note: This parameter is available in DRUID 5.2 and higher and has the default value "false". Important! In DRUID 5.8 and higher, this parameter provides an alternative method for computing the probability of an entity that favors partial matches within the index. The default value is "false".	false
NLU.NER.EntityIndexType	The Lexicon (entity index) type: Inverted - An optimized algorithm that performs an extended and quick search inside the Lexicon increasing the number of candidates to a user says. LinkedSearch - use this value if the Lexicon contains millions of records and the ML engine has to perform searches in these large sets of data. SymSpell	Inverted

Parameter

Description

Default value

NLU.NER.EntityIndex.AlternativeCalculationForProbability

An alternative method for computing the probability of an entity, that favors exact matches within the index.

Note: This parameter is available in DRUID 5.2 and higher and has the default value "false".

Important! In DRUID 5.8 and higher, this parameter provides an alternative method for computing the probability of an entity that favors partial matches within the index. The default value is "false".

false

NLU.NER.EntityIndexType

The Lexicon (entity index) type:

Inverted - An optimized algorithm that performs an extended and quick search inside the Lexicon increasing the number of candidates to a user says.
LinkedSearch - use this value if the Lexicon contains millions of records and the ML engine has to perform searches in these large sets of data.
SymSpell

Inverted

Experimental Parameters

Parameter	Description	Default value
NLU.NER.EntityIndex.UseForClassification	Multiplies the training phrases on the flow (within the model) by the number of entity records (elements) in the Lexicon (indexed entity). Important! Because the parameter's default value is too high, we strongly recommend you to use this parameter only together with NLU.NER.Classification.MaxNoOfUterancesFromEntityValuesForNer.	50
NLU.NER.Classification.MaxNoOfUterancesFromEntityValuesForNer	To ensure that the training model is balanced, set this parameter to 3 up to 5. The value will apply per utterances per entity record. Note: This parameter is available in DRUID 5.17. For example, there is a Lexicon for colors, which has 7 elements and there are 3 training phrases on the flow. 21 training phrases will be added to the model; a training phrase per element in the Lexicon.
NLU.NER.EntityIndex.RescaleProbs	Set the parameter to true to increase the extraction score which ensures that NER identifies more entity candidates. The parameters increases the extraction score as follows: 0 remains 0 0.1-30 -> 30-60 30-60 -> 60-90 60-100 -> 90-100 Example: In entity index there is a list of smartphones. The user asks for the price of an iPhone. The list of smartphones contains a long list of iPhone models, some with long names. Because the extraction score for long iPhone model names decreases, we set the NLU.NER.EntityIndex.RescaleProbs parameter to true to ensure that long iPhone names are also included in the list of candidates identified by NER. Note: This parameter is available in DRUID 5.18.

Parameter

Description

Default value

NLU.NER.EntityIndex.UseForClassification

Multiplies the training phrases on the flow (within the model) by the number of entity records (elements) in the Lexicon (indexed entity).

Important! Because the parameter's default value is too high, we strongly recommend you to use this parameter only together with NLU.NER.Classification.MaxNoOfUterancesFromEntityValuesForNer.

NLU.NER.Classification.MaxNoOfUterancesFromEntityValuesForNer

To ensure that the training model is balanced, set this parameter to 3 up to 5. The value will apply per utterances per entity record.

Note: This parameter is available in DRUID 5.17.

For example, there is a Lexicon for colors, which has 7 elements and there are 3 training phrases on the flow. 21 training phrases will be added to the model; a training phrase per element in the Lexicon.

NLU.NER.EntityIndex.RescaleProbs

Set the parameter to true to increase the extraction score which ensures that NER identifies more entity candidates.

The parameters increases the extraction score as follows:

0 remains 0
0.1-30 -> 30-60
30-60 -> 60-90
60-100 -> 90-100

Example: In entity index there is a list of smartphones. The user asks for the price of an iPhone. The list of smartphones contains a long list of iPhone models, some with long names. Because the extraction score for long iPhone model names decreases, we set the NLU.NER.EntityIndex.RescaleProbs parameter to true to ensure that long iPhone names are also included in the list of candidates identified by NER.

Note: This parameter is available in DRUID 5.18.