NLP Special Parameters

To add special NLP parameters, go to the NLU Configurations page and click the Intents tab. Click on the Thresholds and parameters section header. In the Parameters table add parameters as best suits your needs.

To add a parameter, click the Add icon (+) at the top-right corner of the table. An empty line is added to the Parameters table. From the drop-downs, select the Parameter and its Value.

If you want to add NLP experimental parameters, tap on Add experimental parameters and in the table that appears, add the desired parameters.

Test the NLP model. Write the phrase you want to test (the intent), then click on Test and choose the corresponding language. The Conversational AI will return the matching flow (as it would do during a conversation), together with the matching score for that intent. It is a feedback tool for improving the NLP model.

Note: The testing functionality only works for trained models!

Train the NLP model by clicking the Train button The training model includes ALL training phrases defined on ALL flows associated with the respective bot.

Hint: You can edit or delete a parameter by clicking the Edit / Remove icon displayed inline.

After adding or changing a NLP parameter, click the Save button at the bottom of the page to save the parameters, then click the Train button to train the ML model using the newly added / updated parameter.

This section describes the NLP special parameters broken down per their usage.

Other Classification Algorithms

Parameter	Description	Default value
NLU.NER.Classification.ModelType	This parameter allows you to choose between three model types for NER classification: Non Semantic - A standard model that does not consider the semantic meaning of words. Semantic - Uses a pre-trained Transformers model that understands word meaning in context. This model offers higher accuracy and is ideal for large datasets. For more information, see Natural Language Understanding. Note: All training data must be typos free and must be written with the proper diacritics; otherwise, flow matching might be done erroneously. Important! For DRUID on premise deployments, semantic-based classification is not available by default and it requires a special environment configuration including infrastructure upgrade to GPU processors. For more information, reach out to DRUID Support. Semantic Torch - An enhanced semantic model providing improved performance for NLP tasks. LLM (Technology preview). Enables intent classification using a large language model (LLM). For more information about this NLU model and how to use it, see NLU Intents Classification with LLM.	Non Semantic
NLU.NER.Classification.ModelEmbeddings	This parameter allows you to select an embedding model. Note: This parameter is available in DRUID 8.3 and higher. An embeddings model is a machine learning model that transforms data, such as text or images, into a vector of numbers (an embedding). This vector representation captures the semantic meaning or relationships within the data, allowing for more efficient comparisons, searches, and analysis. The following embedding models are available in DRUID: English: Creates embeddings optimized for English text, capturing semantic meaning in English language inputs. English – score correction: Generates English embeddings with additional score correction, improving the accuracy and relevance of similarity scores. HigherEducation.v1 (English): This specialized model is trained on a rich dataset of higher education content in English only, including university-specific information, leading to improved accuracy when dealing with queries related to this domain. MultiAspect: An enhanced version of our existing Multilingual model, MultiAspect offers improved performance across multiple languages and incorporates score correction for more accurate results. Multilingual: Produces embeddings for multiple languages, enabling cross-lingual semantic understanding. Multilingual – score correction: Offers multilingual embeddings with score correction, enhancing the precision of similarity scores across languages. Paraphrase: Specializes in identifying semantically similar phrases, even if they differ in wording, useful for paraphrase detection tasks. Note: The HigherEducation.v1 (technology preview) and MultiAspect embedding models are available in Druid version 8.13 and later. Hint: The Paraphrase embeddings model processes up to 125 tokens per paragraph and is ideal for short sentences, while other models support up to 512 tokens per paragraph.
NLU.NER.Classification.UseOVA	This parameter, if set to true, triggers another machine learning model for classification called One-Versus-All, which consists of separate binary classifiers - one binary classifier for each possible outcome. You can use it either on the default model, or on the semantic model if NLU.NER.Classification.UseSemantic is set to true.	false
NLU.NER.IndexBasedEntityExtractorV2	Acts as a fallback additional algorithm when the default extraction algorithm fails to extract. When set to true, it searches for more possible candidates from the entire user says inside the Lexicon. If the CRF algorithm extracts entities with a score under the EntityMinSingleThreshold, than the fallback algorithm searches within the Lexicon (entity index) the already matched words to ensure a higher matching score. The search will be done on the left and on the right of the matched word with a maximum offset of 3 words. If the algorithm finds better matches in the Lexicon, they will be added or replaced based on the matching score.	true
NLU.NER.Classification.UseKNN	This parameter acts as a fallback algorithm to help extract the correct answer from an index. Set it to true to ensure that no potential candidates are excluded from consideration.	False

Multi-sentence Support

Parameter	Description	Default value
NLU.NER.Classification.MultiSentenceSupport	Enables multiple intent detection. Setting this parameter to true will detect at predict time if a user has inserted in the chatbot snippet more than one sentence at once and will predict a match for each detected sentence. For more information, see Multiple Intents. Note: At training time, if this parameter is set to true and the author has inserted multiple sentences as training phrases, only the first sentence in the training phrase will be considered for training!	false

Spelling Correction

Spelling correction is done by default in DRUID chatbots at predict time. This means that even if a user writes in the chatbot snippet a phrase that contains an erroneous word, the spelling correction algorithm will search for the correct word within all training words.

If the author changes NLU.NER.SpellingCorrection.UseGlobalDataset to true, then another global dataset of words will be used for spelling, in the specific chatbot language. The author can disable spelling correction altogether by either setting NLU.NER.SpellingCorrection.SimilarityScore to 1, or by setting NLU.NER.SpellingCorrection.Enabled to false.

Parameter	Description	Default value
NLU.NER.SpellingCorrection.Enabled	Enables spelling correction when predicting an intent.	true
NLU.NER.SpellingCorrection.SentenceLevel	Enables spelling correction at a sentence level instead of word level, improving the spelling correction when predicting an intent. Note: The parameter is valid only if the parameter NLU.NER.SpellingCorrection.Enabled  is set to true. This parameter is available in DRUID 5.10 and higher.	false
NLU.NER.SpellingCorrection.SimilarityScore	The similarity score the Conversational AI used for misspelling correction. Note: The parameter is valid only if the parameter NLU.NER.SpellingCorrection.Enabled  is set to true. The default value is 0.8 and the maximum value is 1.0. The value 1.0 means no spelling correction is done.	0.8
NLU.NER.SpellingCorrection.UseGlobalDataset	When set to true, the NLP engine does the first spelling correction using the model data set (generated by intents) and the second spelling correction using a global data set / dictionary. Note: The parameter is valid only if the parameter NLU.NER.SpellingCorrection.Enabled  is set to true.	false

Punctuation removal

Parameter	Description	Default value
NLU.NER.Utterance.KeepPunctuation	Punctuation is removed both at training and predict time, only if NLU.NER.Classification.UseSemantic is set to false, meaning that the default algorithm is used for classification. If NLU.NER.Classification.UseSemantic is true, this parameter has the default value false. Note: This parameter is available in DRUID 5.12 and higher.	true

Parameter

Description

Default value

NLU.NER.Utterance.KeepPunctuation

Punctuation is removed both at training and predict time, only if NLU.NER.Classification.UseSemantic is set to false, meaning that the default algorithm is used for classification.

If NLU.NER.Classification.UseSemantic is true, this parameter has the default value false.

Note: This parameter is available in DRUID 5.12 and higher.

true

Diacritics removal

Parameter	Description	Default value
NLU.NER.DiacriticsRemoval.Enabled	The Conversational AI removes the diacritics. Diacritcs are removed both at training and predict time, only if NLU.NER.Classification.UseSemantic is set to false, meaning that the default algorithm is used for classification. If NLU.NER.Classification.UseSemantic is true, this parameter has the value False.	true

Parameter

Description

Default value

NLU.NER.DiacriticsRemoval.Enabled

The Conversational AI removes the diacritics.

Diacritcs are removed both at training and predict time, only if NLU.NER.Classification.UseSemantic is set to false, meaning that the default algorithm is used for classification.

If NLU.NER.Classification.UseSemantic is true, this parameter has the value False.

true

Stop Words removal

Parameter	Description	Default value
NLU.NER.StopWordsRemoval.Enabled	The Conversational AI removes both default stop words and user added stop words at training and predict time, only if NLU.NER.Classification.UseSemantic is set to false, meaning that the default algorithm is used for classification. If NLU.NER.Classification.UseSemantic is true, this parameter has the value False. Hint: You might want to check the predefined list of stop words per language available by default in DRUID bots.	true
NLU.NER.StopWordsRemoval.UseDefaultList	The Conversational AI uses the system predefined list of stop words. Hint: You might want to check the predefined list of stop words per language available by default in DRUID bots. If the author wants to add more stop words, he can do that by inserting them one below the other in Settings, Conversation AI area – StopWords. If the parameter NLU.NER.StopWordsRemoval.UseDefaultList is set to false, then only the stop words provided in the settings area will be taken into account. If the parameter NLU.NER.StopWordsRemoval.Enabled is set to false, then no stop words will be removed at training and predict time.	true
NLU.NER.StopWordsRemoval.UseLemma	The Conversational AI uses the canonical form of words. When this parameter is set to true, at both training and predict time, the canonical form of the custom stop words list is also removed.	true

Parameter

Description

Default value

NLU.NER.StopWordsRemoval.Enabled

The Conversational AI removes both default stop words and user added stop words at training and predict time, only if NLU.NER.Classification.UseSemantic is set to false, meaning that the default algorithm is used for classification.

If NLU.NER.Classification.UseSemantic is true, this parameter has the value False.

Hint: You might want to check the predefined list of stop words per language available by default in DRUID bots.

true

NLU.NER.StopWordsRemoval.UseDefaultList

The Conversational AI uses the system predefined list of stop words.

Hint: You might want to check the predefined list of stop words per language available by default in DRUID bots.

If the author wants to add more stop words, he can do that by inserting them one below the other in Settings, Conversation AI area – StopWords. If the parameter NLU.NER.StopWordsRemoval.UseDefaultList is set to false, then only the stop words provided in the settings area will be taken into account.

If the parameter NLU.NER.StopWordsRemoval.Enabled is set to false, then no stop words will be removed at training and predict time.

true

NLU.NER.StopWordsRemoval.UseLemma

The Conversational AI uses the canonical form of words. When this parameter is set to true, at both training and predict time, the canonical form of the custom stop words list is also removed.

true

Evaluation

Parameter	Description	Default value
NLU.NER.Validation.CrossValidate.Enabled	Enables model cross evaluation metrics. If this parameter is set to true at training time, regardless of the classification model used, the training will be done 5 times, by randomly splitting the data in an 80/20 fashion, where 80% is used for training and 20% for testing. After the 5 trainings are done, results are averaged and can be downloaded via the DownloadTrainLog button. Hint: By using this method, the training time will multiply by 5.	false

Parameter

Description

Default value

NLU.NER.Validation.CrossValidate.Enabled

Enables model cross evaluation metrics.

If this parameter is set to true at training time, regardless of the classification model used, the training will be done 5 times, by randomly splitting the data in an 80/20 fashion, where 80% is used for training and 20% for testing. After the 5 trainings are done, results are averaged and can be downloaded via the DownloadTrainLog button.

Hint: By using this method, the training time will multiply by 5.

false

Reranker for flow matching

Parameter	Description	Default value
NLU.NER.Classification.UseReranker	Enhances flow matching by evaluating all flows that meet or exceed the threshold specified by the NLU.NER.Classification.Reranker.MinThreshold experimental parameter. By default, the NLU.NER.Classification.Reranker.MinThreshold parameter is set to 0.1, but you can adjust this value in the Experimental Parameters section. The DRUID Reranker assesses user inputs against each training phrase within a flow individually, providing a more refined match. Note: Enabling the Reranker may significantly increase training time, especially for larger test sets. This feature is available in DRUID version 7.13 and later.	false
NLU.NER.Classification.Reranker.JoinUtterances	Combines all training phrases of a flow into a single unit. This allows the Reranker to compare user inputs against the overall meaning of the flow, rather than individual phrases, improving the match accuracy.	false

Parameter

Description

Default value

NLU.NER.Classification.UseReranker

Enhances flow matching by evaluating all flows that meet or exceed the threshold specified by the NLU.NER.Classification.Reranker.MinThreshold experimental parameter. By default, the NLU.NER.Classification.Reranker.MinThreshold parameter is set to 0.1, but you can adjust this value in the Experimental Parameters section.

The DRUID Reranker assesses user inputs against each training phrase within a flow individually, providing a more refined match.

Note: Enabling the Reranker may significantly increase training time, especially for larger test sets. This feature is available in DRUID version 7.13 and later.

false

NLU.NER.Classification.Reranker.JoinUtterances

Combines all training phrases of a flow into a single unit. This allows the Reranker to compare user inputs against the overall meaning of the flow, rather than individual phrases, improving the match accuracy.

false

Other

Parameter	Description
NLU.NER.Utterance.Case	This parameter controls how NLP handles the casing of words in user utterances. Values: 0 (Lower case): (default) Converts all words in the utterance to lowercase before performing NLP. 1 (Upper case): Converts all words in the utterance to uppercase before performing NLP. 2 (None): Makes NLP case-insensitive. This means NLP focuses on the meaning of the words rather than their capitalization. Use this parameter to adjust NLP behavior based on your user experience.

Parameter

Description

NLU.NER.Utterance.Case

This parameter controls how NLP handles the casing of words in user utterances.

Values:

0 (Lower case): (default) Converts all words in the utterance to lowercase before performing NLP.
1 (Upper case): Converts all words in the utterance to uppercase before performing NLP.
2 (None): Makes NLP case-insensitive. This means NLP focuses on the meaning of the words rather than their capitalization.

Use this parameter to adjust NLP behavior based on your user experience.

Experimental Parameters

Parameter	Description	Default value
NLU.NER.Classification.DeterministicTraining	This parameter enhances bot accuracy, allowing the model to provide more precise responses. By default, it is set to False to maintain backward compatibility. Using this parameter will significantly increase the training duration: roughly double for non-semantic models and up to 20 times longer for semantic models Note: This parameter is available in DRUID version 8.6 and higher.	False
NLU.NER.Classification.ModelType	This parameter appears in the NLU Experimental Parameters section in DRUID versions prior to 7.18 if NLU.NER.Classification.UseSemantic was set to "true". You can remove it or switch to the new NLU.NER.Classification.ModelType parameter, which provides you with the option to choose between three model types for NER classification.
NLU.NER.Classification.Reranker.MinTreshold	Defines the minimum flow matching score required for the DRUID Reranker to be used. When the flow matching score meets or exceeds this threshold, and if NLU.NER.Classification.UseReranker is enabled (set to true), the Reranker will be applied to improve flow matching accuracy.	0.1
NLU.NER.Classification.UseSemantic	This parameter appears in the Experimental Parameters section for backwards compatibility if it was set to "true" in DRUID versions prior to 7.18. You can either remove this parameter or replace it by setting NLU_NER_CLASSIFICATION_MODEL_TYPE to Semantic or Semantic Torch.

Entities

Parameter	Description	Default value
NLU.NER.CrfAlgorithmImplementation	The statistical algorithm used by the Conversational AI for structured prediction. Possible values: Crf++ Crf++V2 - An improved Crf++ algorithm. Note: In DRUID 5.19 and higher, it is the default value; therefore if you're using this parameter, we recommend you to set its value to Crf++V2 and train the bot. CrfSuite - an implementation of Crf++ for labeling sequential data. The CRF algorithm is used when training entities. If the author is working with large, indexed entities, we recommend to set this parameter to CrfSuite.	Crf++V2
NLU.NER.CrfFlowStepInputMappingAlgorithmImplementation	For input mapping training phrases defined on a step level DRUID creates a smaller NLP model. If you define a larger number of training phrases on steps, for better results, we recommend settings the parameter NLU.NER.CrfFlowStepInputMappingAlgoritmImplementation to "Crf++V2". Note: This parameter applies on a bot level, input mapping training phrases on all steps. You cannot set it per step.	Crf++V2

Parameter

Description

Default value

NLU.NER.CrfAlgorithmImplementation

The statistical algorithm used by the Conversational AI for structured prediction.

Possible values:

Crf++
Crf++V2 - An improved Crf++ algorithm.

Note: In DRUID 5.19 and higher, it is the default value; therefore if you're using this parameter, we recommend you to set its value to Crf++V2 and train the bot.

CrfSuite - an implementation of Crf++ for labeling sequential data.

The CRF algorithm is used when training entities. If the author is working with large, indexed entities, we recommend to set this parameter to CrfSuite.

Crf++V2

NLU.NER.CrfFlowStepInputMappingAlgorithmImplementation

For input mapping training phrases defined on a step level DRUID creates a smaller NLP model. If you define a larger number of training phrases on steps, for better results, we recommend settings the parameter NLU.NER.CrfFlowStepInputMappingAlgoritmImplementation to "Crf++V2".

Note: This parameter applies on a bot level, input mapping training phrases on all steps. You cannot set it per step.

Crf++V2

NeuralSpace

Parameter	Description
NeuralSpace.AuthorizationToken	The NeuralSpace Platform authentication token. To get the authentication token, log into the NeuralSpace Platform and click the Copy Auth Token button at the top-right corner of the Dashboard.
NeuralSpace.PlatformUrl	The URL of the NeuralSpace Platform if hosted on premises. The URL should be accessible from the Internet.

Parameter

Description

NeuralSpace.AuthorizationToken

The NeuralSpace Platform authentication token. To get the authentication token, log into the NeuralSpace Platform and click the Copy Auth Token button at the top-right corner of the Dashboard.

NeuralSpace.PlatformUrl

The URL of the NeuralSpace Platform if hosted on premises. The URL should be accessible from the Internet.

Deprecated Parameters

Parameter	Description
NLU.NER.Classification.UseAllUtterances	Important! This parameter has been deprecated due to Child Intents which better addresses the need of this parameter. We strongly recommend you not to use this parameter anymore. Useful when you have models with input mapping with entity expressions. This parameter on True means that all utterances are considered at training time. If set on False, only those phrases that do not contain entities or utterances with entities which have the parameter NLU.NER.EntityIndex.UseForClassification set to true (see NER parameter description) are considered at training.

Parameter

Description

NLU.NER.Classification.UseAllUtterances

Important! This parameter has been deprecated due to Child Intents which better addresses the need of this parameter. We strongly recommend you not to use this parameter anymore.

Useful when you have models with input mapping with entity expressions.

This parameter on True means that all utterances are considered at training time. If set on False, only those phrases that do not contain entities or utterances with entities which have the parameter NLU.NER.EntityIndex.UseForClassification set to true (see NER parameter description) are considered at training.