NER
Named Entity Recognition, an information extraction task in natural language processing that identifies and classifies named entities in text—such as person names, place names, and organization names—into predefined categories.
Timeline (last 90 days)
Extracting structured fields from unstructured voice text can be achieved through LLM function calling/structured output capabilities, or through traditional Named Entity Recognition (NER) models.
The LLM parsing path is more flexible and can handle relative time expressions, but relies on external API calls with latency and cost implications; the NER path is faster and can be deployed locally, but has weaker generalization for colloquial ambiguity.
The technical implementation of PII desensitization is mainly divided into two categories: rule-based regular expression matching and NLP/NER model-based entity recognition.
Transformer-based NER models can leverage contextual bidirectional attention mechanisms to classify different tokens within the same word string into predefined culinary attribute slots, making them more robust than traditional regex-based parsers on long-tail ingredient descriptions.
RecipeDB2 uses a Transformer-based named entity recognition model to parse ingredient phrases, splitting them into seven culinary attributes.
Semantica uses the classic NER/RE/EE three-step pipeline from the information extraction field to extract triples: NER identifies entities, RE determines the type of relationship between entities, and EE extracts events and their arguments.
信息抽取流水线包括命名实体识别（NER）、关系抽取（RE）和共指消解（Coreference Resolution）
All Facts (7)
Extracting structured fields from unstructured voice text can be achieved through LLM function calling/structured output capabilities, or through traditional Named Entity Recognition (NER) models.
50%UnverifiedThe LLM parsing path is more flexible and can handle relative time expressions, but relies on external API calls with latency and cost implications; the NER path is faster and can be deployed locally, but has weaker generalization for colloquial ambiguity.
50%UnverifiedThe technical implementation of PII desensitization is mainly divided into two categories: rule-based regular expression matching and NLP/NER model-based entity recognition.
50%UnverifiedTransformer-based NER models can leverage contextual bidirectional attention mechanisms to classify different tokens within the same word string into predefined culinary attribute slots, making them more robust than traditional regex-based parsers on long-tail ingredient descriptions.
50%UnverifiedRecipeDB2 uses a Transformer-based named entity recognition model to parse ingredient phrases, splitting them into seven culinary attributes.
50%UnverifiedSemantica uses the classic NER/RE/EE three-step pipeline from the information extraction field to extract triples: NER identifies entities, RE determines the type of relationship between entities, and EE extracts events and their arguments.
50%Unverified信息抽取流水线包括命名实体识别（NER）、关系抽取（RE）和共指消解（Coreference Resolution）
50%