Back to Main Site

AKIN Natural Language Processing DEMO

Medical Model

(Extremely tolerant of variation)

This Demo detects two diagnoses below and their symptoms. Edit the text in the input box below and click 'Analyze Text' to see how well AKIN can detect these concepts:

  • Diabetes
    • Symptoms: Fatigue, Hunger, Frequent Urination, Blurred Vision, Thirst
  • Myocardial Infarction (Heart Attack)
    • Symptoms: Chest Pain, Labored Respiration, Arm Pain, Jaw Pain, Heartburn, Belching, Malaise, Head Ache, Vomiting

Only the symptoms listed above have been modeled and will be directly detected. This demo showcases AKIN's ability to detect these symptoms when they are written in a wide variety of ways, including typos, misspellings, grammatical errors, or unexpected composition. When diagnoses are detected, they are Inferred Concepts from AKIN Inference Rules, and have been configured based on basic medical knowledge widely available on the internet. This model is NOT meant to exactly reflect all symptoms of these diseases or provide exact diagnosis. It only illustrates AKIN's ability to detect some symptoms in free unstructured text, and make inferences based on those where appropriate.

Please input text below and test AKIN. A starting example of input text has been provided for you in the input box below. You can edit or replace it. The results are sorted by high relevance items first and then by word position.

First please confirm that you are not a robot to start.
The first call to the service may take a moment to bring the service up. After that, all subsequent queries to the service should be fast, unless there is a long pause in between queries.

The demo above provides an example of the AKIN NLP API’s ability to sense and understand the conceptual meaning of free unstructured text , and output that understanding into structured standardized representations. It's high sensitivity leads to greater adaptability to variation than other technologies. Additionally, AKIN provides traceability information for downstream consuming systems and users, delivering unparalleled transparency. The traceability info shown in this demo is only a small sample of the full set of information that AKIN provides to consuming systems.

In this scenario a person may have sent an electronic text message to a medical clinic, had their conversation transcribed, or might be a doctors note. AKIN analyzes the text to detect descriptions of symptoms related to Diabetes or Myocardial Infarction, and then performs assessment of inference rules to detect possible diagnoses which can be used by the host medical computer system to alert employees or practitioners, provide improved prioritization, or extract critical information and features for research purposes.

Below we've provided some examples of the types of variation-filled text a person might write, communicating their ideas and thoughts in multiple ways, including spelling and grammatical errors in bold. You can copy and paste these (or edit them) into the input box above and see what type of results you get:

Example 1

I was running the other day and suddenly felt a sudden sharpness in my strnum and found it hard to ctch mybreath. Also, Ihad a dull constant throbbing in my forehead the rest of the day.

Example 2

I can barley get up in the murning and feel like I have to constantly drink water

Example 3

Yesterday I go running and feel odd thing in mychest. My jaw start to tingle and I have hardtime brathing.

Example 4

I have a strange feeling in my chest, and my jaw aches. I've been really dizzy, throwing up, and have this wicked thrabbing in my head.

Example 5

I wanted to let you know that I've been feeling more tired lately. Also, it seems like I've always got the munchies… constantly needing to eat. I'm going to the batroom to pee alot. Also I've noticed something strange with my vision. Normally I can see just fine but now it’s hard to makeout objects or signs at night when I drive.

Example 6

The patient complains of low energi, blury vision, and constnt hanger.



High Level Features:
  • Built-In Artificial Intelligence
    • We've removed the need to do the expensive and time consuming work of feature engineering and extraction, testing and selecting machine learning algorithms and approaches, tagging tens or hundreds of thousands of records, and training ML Algorithms. Our AI Model allows you to simply configure/add the concepts you want to detect and AKIN can use that information in combination with its built-in intelligence to detect those concepts and make inferences and determinations with a high degree of accuracy.
  • Direct Injection of Knowledge
    • Sometimes you know exactly how you want the system to interpret some information or context, and you don't want it making probabilistic assessments. AKIN also supports explicit knowledge and rules.
  • Easy to configure and manage Domain Knowledge Models
    • Directly via API and Model Manager User Interface
  • Extremely tolerant of textual variations
    • Varied expressions, mispellings, and grammatical errors
  • Highly Transparent Model
    • Easy to find out why something was or was not detected, and make necessary adjustments or improvements
  • Built-In Native Concepts Types & Detection
    • Intents, Sentiment, Descriptors, Actions
    • Numbers, ordinals, dates/times, timespans, units of measure & account, assignment & equality operators, and grammatcial indicators
  • Customer defined custom entity types
    • Complex Entities with properties having multi-level hierarchies and multiple relationships
    • Graph + Hierarchical Relationships
  • Advanced Inferred Concept Detection
    • Some concepts are not explicitly stated but inferred based on the presence of other concepts and ideas
  • Command/Intent detection and user interaction
    • Detects concepts in large paragraphs or documents of text, as well as shorter user commands or intents equally well
  • Advanced Noise Reduction
  • Highly Optimized Performance
    • In-memory distributed processing
    • Sub-second response times
  • Deploy Anywhere
    • API dll can be hosted on any .Net compatible platform in the cloud or on premises
  • Multi-lingual support
    • Although currently only certified for English, AKIN has been designed to work extremely well across cultural domains, including Asian languages.
Configuration:

To configure AKIN, you feed it knowledge in the form of standardized values of concepts and ideas you want detected and fed to your downstream consuming systems. This is done either through easy to use API functions/methods, or via the Model Manager UI. Additionally, you give it synonymous and related terms for the concepts you define. AKIN uses this knowledge to make smart determinations about the text you want it to analyze. You don't have to spend hours tagging up thousands of records to "train" it in a haphazard, disorganized way. You directly feed it knowledge in an orderly fashion, and you can always see what it knows and what it doesn't know. Then, AKIN uses its sophisticated probabilistic AI to detect concepts and ideas within the text even when there are a lot of variations in the way something is written, such as spelling and grammatical errors.

Why do I Need Natural Language Processing?

For many types of businesses a large amount of information is still collected, bound up, and stored in unstructured or semi-structured text. Every year, businesses spend enormous amounts of money attempting to successfully manage and extract value from this data, that frequently require manual efforts enhanced by discovery technology that often feels rudimentary and inadequate.

Additionally, businesses and application developers are looking to create more natural interactions with their information systems for consumers, customers, and their employees. They want the ability for users to be able to write or speak inquiries and directives naturally, and have their information systems understand and be able to process this information. This is no easy task.

The problem is that everyone communicates differently. Individuals express their thoughts, ideas, and concepts in so many different, non-standard, unique, and individualistic ways. Exacerbating the problem, people also make mistakes, or speak different native languages leading to unexpected grammatical structures when they write or speak (speech to text). For example, in a medical domain people may have symptoms or issues they express in different unique ways. Several people might describe the issue of having difficulty breathing in a number of ways:

  • “I have a hard time breathing”
  • “Sometimes it’s painful when I inhale”
  • “I find myself gasping for air”
  • “There are times when I can’t catch my breath”
  • “It hurts to breathe”
  • Etc.

Downstream consuming software systems need to be able to understand those concepts and do things like alert key stakeholders or gather significant data for analysis and research. However, those downstream systems need a standardized representation of these concepts, in this case the symptom we describe above, something like “Difficulty Breathing” or “Labored Respiration”. Having a single representation of this symptom makes it extremely easy for other systems to use this information in a structured way.

AKIN Natural Language Processing provides unparalleled accuracy and performance allowing your business to effectively extract the value from your unstructured text and natural language queries. There is no need to rely solely on centralized Cloud-Only-Based solutions like Google, Alexa, or Cortana that tie you down to an ecosystem and take your data out of your hands. With AKIN you can host anywhere on-premises, and use any speech to text technology with it, like Dragon Naturally Speaking, and those already freely available on mobile devices.

AKIN has been designed to be super intelligent and high performance, and yet still be very lightweight and efficient. It can even be hosted directly on mobile devices or very lightweight client Virtual Machines.

For more information and licensing inquiries, please contact info@grappledata.com using a valid company/organization email address.