Named-entity Recognition (NER)

What is Named-entity Recognition? 

Named-entity Recognition (NER) is a method of recognizing and classifying essential pieces of information from within larger unstructured text-based data into predefined categories such as person names, organizations, locations and more.

How Does Named-entity Recognition Work?

Named-entity recognition (NER) is a natural language processing technique used to pinpoint and classify named entities found in unstructured text and convert them into predefined categories such as names and locations. 

Named-entity recognition works by taking an unannotated section of text, such as:

Ryan sold 400 shares of in 2021.

And annotating it so that it highlights the different entities and names in the section:

[Ryan]personsold 400 shares of []organization in [2021]time. 

Various machine learning algorithms are typically used for NER, such as support vector machines and conditional random fields, as well as newer techniques like Recurrent neural networks (RNNs) and transformers (e.g., BERT).

Use Cases

Named-entity recognition is one of the most fundamental problems in NLP and has a variety of applications.

One of the most common areas where we can find named-entity recognition being used is within web search engines in order to understand the entities within a query. 

In brief, search engines aim to understand what the user is searching for and give them the best answer to their query. The way in which a search engine does this is by categorizing the content on a web using ‘keywords’ that are “understood” using NER. 

Another commonly seen application of NER is within product reviews. Named-entity recognition is used in order to extract product names and other important entities from online reviews. By doing this, people can gain insights on which of their products and features are being discussed and in what way.

A third use of named entity recognition is within chatbots

Named-entity Recognition and Chatbots 

As mentioned, named-entity recognition helps identity and categorize critical elements within textual data. This enables a lot of businesses to find insights from large unstructured datasets.

One way in which an organization could capitalize from the use of named-entity recognition is through the use of chatbots. 

Many businesses set up chatbots to help automate tedious processes such as documentation collection, customer support services and more, in order to save manual work for other complex tasks.

There are, however, a few challenges involved in doing this, specifically in terms of customer support services. Firstly, the chatbot must understand who it is speaking to in order to provide a good user experience. Secondly, they have to be able to extract important data in order to respond in a meaningful way.

These challenges can only be overcome if the chatbot is thoroughly trained with named-entity recognition algorithms to recognize these entities from within a chat and process them correctly. 

Furthermore, previously extracted information from named-entity recognition models can be used to further train the algorithm and attain higher accuracy levels. 

Below is an example of NER classifying named entities within a conversation of Hyro’s COVID-19 virtual assistant. In this specific case, the named-entity recognition has been trained to recognize medical symptoms from within a chat, such as coughing, and use additional knowledge about COVID-19 specific terms, such as identifying Rome as a high risk city.

Hyro’s Named-Entity Recognition

Recent Developments in Named-entity Recognition (NER) ​

Here are some recent developments in Named Entity Recognition (NER):


  • Harnessing Pre-trained Language Models: Recent progress in NER involves utilizing pre-trained language models such as BERT and RoBERTa, greatly enhancing NER system performance. These models are trained on extensive text datasets, enabling them to encode words and phrases in a manner that captures both their semantic and syntactic attributes. NER systems can leverage this knowledge to better detect and classify named entities.
  • Self-Supervised Learning’s Impact: Self-supervised learning, a form of machine learning that doesn’t necessitate labeled data, has emerged as a valuable tool for NER. Instead of relying on labeled data, models master tasks by predicting missing or distorted segments of input data. This methodology proves effective for NER, allowing models to learn essential features for recognizing named entities without manual annotation of extensive datasets.
  • Attention Mechanisms in Focus: Incorporating attention mechanisms empowers NER systems to concentrate on the most pertinent segments of input text while making predictions. This capability proves advantageous for identifying named entities embedded within intricate or noisy text contexts.
  • Enabling Ensemble Learning: Ensemble learning, a technique that amalgamates predictions from multiple models, enhances overall accuracy. In NER, this strategy is proven effective in ameliorating errors that individual models might produce.

These represent just a glimpse into the latest NER advancements. As research in this domain progresses, we can anticipate further enhancements in NER system performance.


Here are additional recent trends in NER:


  • NER Systems for Low-Resource Languages: While NER systems have predominantly been tailored for high-resource languages like English and Chinese, there’s a burgeoning interest in creating NER solutions for low-resource languages such as various African and indigenous languages.
  • NER Systems for Specialized Domains: Conventionally designed for recognizing named entities in general text, NER systems are now being honed for specific domains like medical and legal text, catering to domain-specific naming conventions and contexts.
  • Enhancing Robustness against Noise and Ambiguity: NER systems often grapple with noisy and ambiguous text. Recent research is dedicated to formulating NER systems with enhanced robustness to surmount these challenges effectively.

Helpful Resources

Unlock your digital potential with the #1
adaptive communications platform.