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

Self-diagnosing natural language QA system that improves itself based on facial feature tracking

IP.com Disclosure Number: IPCOM000241901D
Publication Date: 2015-Jun-05
Document File: 2 page(s) / 141K

Publishing Venue

The IP.com Prior Art Database

Abstract

A self-diagnosing natural language question answering (QA) system that improves itself based on facial feature tracking is disclosed.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 2

Self-diagnosing natural language QA system that improves itself based on facial feature tracking

Disclosed is a self-diagnosing natural language question answering (QA) system that improves itself based on facial feature tracking.

Question answering is a hard problem in natural language processing (NLP) and machine learning (ML) systems. Often, it may be difficult to incorporate user feedback into the ML framework once the framework is in place. The disclosed system utilizes audio, visual/image analysis as well as active learning and reinforcement as feedback into an NLP ML system. The disclosed system avoids the need for manual intervention and data wrangling and adds a cognitive richness to the experience. The system analyzes facial expressions as part of a feedback mechanism.

An embodiment of the system has the following components:

A back-end natural language question answering system.

1.


3.

4.

have been present in the candidate answer variant set, or it could be some additional data that could be transformed into new features for a new training. Back-end re-training system which adds the information obtained from the user

       5.
feedback to the training set, and retrains the machine learning models. This component is based on active learning and online machine learning methodologies. Repeat steps until some sort of convergence. This cycle repeats for a number of


6.

times configurable in the system. If however, at some point during one of the answers, the system is not able to identify either positivity or negativity from the user's facial features, a pop-up appears that asks the user what they think about the answer.

Over time, repeatedly incorporating feedbac...