Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Method and Apparatus for Context Based Machine Learning Model

IP.com Disclosure Number: IPCOM000247499D
Publication Date: 2016-Sep-10
Document File: 3 page(s) / 64K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a hybrid machine-learning model for a Question & Answer (Q&A) engine that not only learns based on context, but also accounts for various parameters in real time when responding to questions.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 3

Method and Apparatus for Context Based Machine Learning Model

Cognitive Entities (CE) models are cognitive models enabled via a Big Data platform. CEs are aimed to remember the past, interact with humans, continuously learn, and refine the responses for future. Cognitive systems require initial training to build a ground truth, and then self-learn later. Currently, such learning is generic, not context-driven learning. Thus, a static Machine Learning model enables the cognitive system to know the answer without any context.

Typical example query: What is the favorite food of Americans?

Possible answer with related context: porridge (on a rainy day), pancakes or bagels (on a sunny morning), burger (on a sunny afternoon)

Mapping the ground truth thus requires real context with the minimum data of time and location and outputting the answers in the context relevant. An approach is needed to provide cognitive models that interact with humans and are as natural as possible.

The novel contribution is a hybrid machine-learning model for a Question & Answer (Q&A) engine that not only learns based on context, but also accounts for various parameters in real time when responding to questions. This system provides users with more relevant answers by learning the ground truth in its given context, using the context features in a cognitive component that can connect to external sources, and then using the external information to determine correct answers to user's questions.

The hybrid machine learning model engine offers the following novel capabilities:

1) To create a context-driven ground truthfor a given interaction input 2) To have an atomic context or a combination of contexts to the ground truth 3) To integrate with external sources or external Internet of Things (IoT) in real time or near real time and respond to input interactions to align with the context (e.g., location, time, or other specific features)

4) To use external information from [3] and context-based ground truth from [2] to customize the response based on the current identified context

5) (Optional) To define priorities and override inter-context factors (e.g., time is of greater priority than the location)

Current Q&A systems rely on supervised machine learning techniques. These systems feed/teach machines the correct answers for a large number of possible questions (i.e., the ground truth), and then classify the answer to the current user question by calculating the similarity to the training set questions and selecting the response corresponding for most similar one.

The novel system enhances this approach by adding both a context component and a cognitive component to the training stage.

Examples of context components are (but are not limited to):


 during summer , in Romania , Germany, and [x, y, z other countries] the favorite

1


Page 02 of 3

  beverage is beer ;
 when it is hot , in Romania , the favorite beverage is beer ;
 during summer , in India , the fav...