IntentDetector#
- class IntentDetector(config)#
Bases:
BatchedStep
Do intent detection
- Parameters
step (str, "app") – app
type (str, "query_processing") – query_processing
name (str, "intent_detector") – intent_detector
analyzed_input_field (str, "nlp") – analyzed spacy Doc
target_index_name (str, "") –
index to be searched
Deprecated since version 3.10.7: Target index name is now automatically detected from the pipeline
match_entity_as_phrase (list, ["PERSON", "PRODUCT", "FAC", "NORP","ORG", "GPE", "LOC"]) – list of NER tags that should be appended as a phrase query
match_entity_phrase_slop (int, 5) – Phrase slop that is applied for those auto-detected entities
match_entity_combine_operator (str, "OR") – If multiple entities are detected, should all of them match? OR / AND
label_lookup_match_ngram_field (bool, False) – [Label Retrieval] Enable matching ngram field (high recall)
label_lookup_fuzzy (bool, False) – [Label Retrieval] Enable fuzzy search (high recall)
label_lookup_prefix_queries (bool, False) – [Label Retrieval] Enable sub-query prefix queries (high recall)
label_lookup_most_common_rescoring (bool, True) – [Label Retrieval] Rescore matching label values by their respective overall doc–count (how often documents are tagged with this label)
label_match_rewrite_mode (str, "filter_query") – How to apply the detected label match: filter, boost, or rescore
label_match_combine_operator (str, "OR") – How to combine multiple label matches: OR / AND
label_match_category_weights (dict, {}) – Specify weight applied on the facet-filter, by type. Default is 10
date_match_on_facet (str, "item_created_at") – What field to be used for date filtering
date_match_rewrite_mode (str, "filter_query") – How to apply the detected date match: filter, boost, or rescore
date_match_recency_per_category (dict, {}) – TODO: Mapping of recency boost profile for detected category (from label-match lookup)
path (str, ".") – path
Methods Summary
process_doc
(doc)Process a document
process_entities
(doc)- rtype
List
[Intent
]
Methods Documentation
- process_doc(doc)#
Process a document