Custom Named Entity Recognition, Resolution
Better query understanding for your business



<1 ms
10x faster than most NER, slot filling*2

Real time language processing

Fast analysis of natural language with performance of under 1millisecond*1 for < 10 word queries. With this sub millisecond latency NLU performance, ThatNeedle is order of magnitude faster*2 than some of the big names providing such services. This is ideal for real time NLP applications involving text processing. ThatNeedle performs not just recognition but also performs entity processing/slot filling as required.

Goodbye GPUs and Accelerators

Our libraries work on CPUs and this makes for cost efficient operations.
This is unlike many other frameworks where GPU or hardware accelerator is mandatory for getting the system ready for natural language processing tasks.



0 gpu
Works great on CPUs

0 cloud
No network latency, better privacy

Offline, On Prem operation

You can work with NLP offline without the overhead of network latency. It will work offline without internet and can be hosted on your own servers residing on your premises. This not only removes network latency for NLP but also helps with security and various data privacy compliances such as GDPR.

Will Work for your domain

An NLP framework that is suited for general purpose applications like Siri and Alexa assistants will not work well for domain specific applications. You need something that will be aware of your domain. ThatNeedle NER has the ability to be configured for your domain quickly.

Vertical

0 iob
Work Without IOB datasets

Goodbye to IOB & similar annotated datasets for training

For domain specific applications, geting IOB format annotated datsets (and other similarly annotated datasets) can be a problem. Even if you manage to get such annotations done, it is slow and expensive. With ThatNeedle NER, you are not dependent on IOB annotations. We are designed to work with minimum processing from raw data.

Natural language numbers

ThatNeedle will automatically handle natural language numbers in words and convert them to digits.

One=1
Words to digits conversion
Easy
Focus on business logic

For Regular programmers
No NLP Phd required

Your developers can setup with just a few lines of code in your favourite programming language such as Python, C++, etc. With ThatNeedle libraries, you can extract custom named entities from unstructured text query with just 1 line of code. The library performs some basic entity resolution and basic fuzzy matching as well, which can be further enhanced as per business needs.


Entity Extraction example


Input could be :
blahblah with one hundred sixty six milli litres of paint thinner
The default / "out of the box" output could be a JSON formatted string similar to the one illustrated below (representative only, actual might differ):
{"MEASURE":"166ml", "PRODUCT":"paint thinner",}
i.e the blahblah would be ignored; however it is possible that blahblah could be an important entity for your special niche. In that case you could custom define blahblah as a special entity named blah_entity. After that, the new special entity would also be recognized and labelled accordingly.
See the screencast below.


Entity Resolution Demo

This demo shows how effectively certain ambiguties can be resolved based on the context for effective resolution of entities

Tell us about your business for custom optimized real time libraries:

contact us

Bespoke Entity Extraction (Custom NER)

Let us know about your custom entity recognition needs. Some topic extraction solutions restrict the entities to nouns, proper nouns etc. But depending on the business needs, you might want to have some particular types identified and extracted as entities. You should be able to define what to extract as custom entity and what not to label as an entity. If done naively, this is a tricky exercise and people often end up burning their hands. We will create the best solution for your text analysis and named entity recognition needs. We can custom create and test custom models for your niche and give you the pre-trained software solution that is ready to use for your niche and specific needs.
While the software allows the user to define custom entities and annotation, any other customization cost would be over and above the default price mentioned.
The default language is English, but the technology is capable of effective handling other languages, includes Asian languages like Chinese, Japanese, Arabic etc. These are traditionally a challenge, but our algorithms are designed to solve these natural language understanding issues. Please take a look at our other NLP libraries also.


About Named Entity Recognition

NER or Named Entity Recognition / Entity extraction identifies, extracts and labels the information in text into pre-defined categories, or classes such as location, names of people, brand, product etc. It is a loosely used term to also include entity-extraction of information such as dates, numbers, phone, url etc. Entities could be any useful data or information for example, date time, names, location, dimensions etc that could be stored or used for text processing. Some extractors, identify proper nouns or nouns as entities but thats too rigid and is not a good rigid rule. A good entity extractor should be able to take a string of unstructured text and identify and produce annotated output or a structured output that helps in intelligent and better analysis of the text. Such intelligent understanding (NLU) of the intent of the user query will help in producing better responses from the system. If it is a search query, it would mean better understanding of the query and more relevant search results because of better intent inference. Entity resolution and fuzzy matching are also an important part of this process of disambiguation. There is no universal entity extractor and the needs of the business must be taken into account before selecting a software tool to perform such tasks. Many such general purpose tools might be able to parse general entities like date, time, location etc but give poor accuracy for the context of the business in question, and are therefore not fit to be used in specific niches. A good tool will recognize the context of the niche and give annotations and analysis accordingly.

Importance of fast entity extraction from natural language

Most language processing software cannot parse the query and analyse the query fast enough to be used effectively in user interfacing applications. As a result response from the backend query processing system appears to be slow and tests the patience of the end user. A fast response to the query is essential not only to delight the customer, but to keep him engaged. A slow application will give the user a good reason to direct his valuable attention elsewhere. ThatNeedle has always recognised the need for speed in NLP and is making the core engine faster everyday. We are also proud to say that we are 10x faster than some leading natural language entity extraction from text service providers This would make ThatNeedle an ideal candidate for real time extraction tasks from plain text. ThatNeedle NER can serve as an ideal text processing tool for big data scientists, data architects, semantic search solution providers, realtime natural language processing, large scale NLP etc

Out of the box, ThatNeedle could be used as an effective and faster NLP microservice alternative to Microsoft Luis, alternative to IBM watson, alternative to Wit.ai, alternative to api.ai, alternative to Natty etc

Even if you are using traditional specialized parsers like Natty for Java or any similar library for date extraction etc, you should compare the performance with ThatNeedle and decide which is superior for yourself! Thatneedle can provide great speed and accuracy because of its high precision high recall engineering.

How to select named entity extraction tools / software / framework

There are many entity extraction tools / named entity extraction software for NLP floating around in the market. Some are just repackaging open source software, some are repackaging white labelleled software. There are many open source NER tools, one prominent tool is Stanford NER (in Java). NLTK (Natural Language Tool Kit) is a very popular python library for natural language processing in python. Some tools would require a learning curve and getting familiar with parts of speech tagging and some language processing know how. While the learning curve is a crucial factor in a business decision, another important thing is to test and compare the tools. We encourage you to try out some open source and commercial software so that you can truly appreaciate the value that ThatNeedle brings to the table. See how it performs out of the box for the advertised test cases. You should then vary the test cases. Make a note of the speed and the accuracy of the of software performance. A good NLP topic extraction tool should also have some features to customize for your niche. There is a fair chance you do not have a good dataset to train a fresh ML system for your niche. Relevant data sets are hard to obtain and one that is sufficiently large in size is rare. While many open source data sets are available, many wont be relevant to you. Integration is also an issue that should be looked into before selecting anything. ThatNeedle was designed to make integration with any technology easy. You could be using python, java, c++, ruby, lisp, erlang, golang, scala, node.js etc and you would still be able to use ThatNeedle entity extractor. What's more the technology is cross platform and would work on your favourite OS. You are requested to please confirm once before purchasing. See also our new custom voice commands library offering and Biomedical research insights .

*1. Based on tests conducted by ThatNeedle in 2020. Performance tests are conducted using specific computer systems and reflect the approximate performance of ThatNeedle library. This test has been performed on a 2012 MacBookPro with an Intel Core i5 processor and 4GB RAM.

*2. Based on tests conducted by ThatNeedle in 2020. Performance tests are conducted using specific computer systems and reflect the approximate performance of ThatNeedle library and other systems. Their performance might have improved since.