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

Technique for personalized question answering by orchestrating different system using corpus relevancy

IP.com Disclosure Number: IPCOM000239698D
Publication Date: 2014-Nov-26
Document File: 3 page(s) / 48K

Publishing Venue

The IP.com Prior Art Database

Abstract

Question Answering systems ingest different corpus by processing the documents in different ways and creating different types of artifacts. For example using structure of document, semantics of documents, tables and many more. The way a given question is analyzed to find an answer depends on the underlying artifacts associated with the documents in the corpus. In this publication, we consider multiple question answering systems working on different corpora and propose a way to orchestrate these systems to find an answer to a given question which is personalized based on the user profile attributes.

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

Page 01 of 3

Technique for personalized question answering by orchestrating different system using corpus relevancy

Question Answering systems ingest different corpus by processing the documents in different ways and creating different types of artifacts. For example using structure of document, semantics of documents, tables and many more. The way a given question is analyzed to find an answer depends on the underlying artifacts associated with the documents in the corpus. In this publication, we consider multiple question answering systems working on different corpora and propose a way to orchestrate these systems to find an answer to a given question which is personalized based on the user profile attributes.

Advantages of the proposed Technique:
- Quickly identify whether a given question can be answered by any of the corpora without actually executing the query. Thus a short circuit way of determining whether we should go for the full blown processing or not.

- Instead of giving bad answer, the system can potentially return an answer like "I don't know"

- Profile based personalization i.e. many times it happens that similar questions are asked by different people but the answer they expect is different. So a QA system could utilize this profile based features and return different answers for different people.

Problem:

Orchestrate multiple question answer systems(QAS) with profile based personalization by determining suitability of a corpus for a particular question
Input:

Sequence of questions along with user profiles and user feedback on the answers returned from the system
Output:

- For every question, a ranked list of answers with their source of information (corpus) or the particular QAS giving the answer and a consolidated confidence score
- An internal model, containing information about individual QAS and user profiles, updated based on the user feedback. This would be used for subsequent questions

High Level Diagram:

Solution:


For every corpus we will build the list of topics covered by it along with dimensions like user attributes, weights for each topic based on user attributes and so on. Building this would be done by utilizing the questions, answers along with evidences with user feedback. Each topic would be associated with the list

1


Page 02 of 3

of user profile att...