Named Entity Recognition / NER
NLP task that detects and classifies special names such as the person inside the text.
The asset name recognition (Named Entity Recognition, NER) is a basic natural language processing task that automatically detects and classify "special name" expressions such as names, name names, corporate/company names, dates, currencies, etc. For example, "Mustafa Kemal Atatürk announced the Republic of Ankara on October 29, 1923" in a NER system "Mustafa Kemal Atatürk" can tag "29 October 1923" as the date and "Ankara". This task is often modeled as a label-based classification problem in each word (or token) in the array; classic methods provide much higher accuracy with Transformer-based models such as random areas (CRF) using conditional random fields (CRF).
NER is a foundation building stone of larger systems, so as to be a end-user application per itself. It is used as a critical pre-processing step in many practical applications such as making automatic event and person interest from news texts, automatically extracting party or pharmaceutical names from legal and medical documents, enabling search engines to index documents smarter, automatic detection of product or brand names from customer requests and creating information chart (knowledge graph). With the rise of large language models, NER can now be built for general purpose models with a convenient prompt without having to train a separate model.
NER is a foundation building stone of larger systems, so as to be a end-user application per itself. It is used as a critical pre-processing step in many practical applications such as making automatic event and person interest from news texts, automatically extracting party or pharmaceutical names from legal and medical documents, enabling search engines to index documents smarter, automatic detection of product or brand names from customer requests and creating information chart (knowledge graph). With the rise of large language models, NER can now be built for general purpose models with a convenient prompt without having to train a separate model.