Corpus Augmentation Engine for Cognitive QA Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional cognitive question answering systems lack mechanisms to identify and ingest new content efficiently, leading to low-quality answers due to insufficient knowledge base content, and face challenges in automating the ingestion of new information from online forums without being perceived as bots or astroturfing operations.
Innovation Solution
A system and method that utilize a corpus augmentation engine to generate forum questions based on low-quality topics, posting them to forums using a persona generated from user interactions, to extract and ingest new content into the knowledge base, improving answer quality by evaluating responses for potential inclusion in the knowledge base.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a machine or bot is used to automatically post to a forum to ingest new information, then the automation of content ingestion is improved, but the system is perceived as a bot or astroturfing operation which reduces reliability of content acquisition
Solution Approach 1:
The system employs human users as intermediaries who voluntarily post questions to forums on behalf of the QA system. These users are recruited from the forum community itself and are compensated for their participation, thereby mediating between the automated system and the forum environment to avoid bot detection while maintaining automation benefits
Solution Approach 2:
The system leverages the forum's own user base to generate content by offering incentives (such as forum credits or rewards) to users who post questions. This self-service approach allows the community to serve the system's needs while maintaining natural forum participation patterns that avoid detection
2Reliability
If manual processes are used where domain experts review and select documents for ingestion, then the quality of ingested content is improved, but the productivity of content acquisition deteriorates due to laborious manual processes
Solution Approach 1:
The system performs preliminary filtering and selection of questions by analyzing forum data to identify questions that are likely to yield high-quality answers. This preliminary action reduces the burden on human reviewers by pre-screening content before it reaches domain experts for final selection
Solution Approach 2:
The content acquisition process is segmented into distinct stages: automated question identification, human user posting to forums, answer collection, and expert review. This segmentation allows different tasks to be performed by appropriate methods (automated where possible, human where needed) to optimize both quality and productivity
3Adaptability or versatility
If the knowledge base corpus contains insufficient content, then the coverage of answerable questions is improved, but the quality of answers deteriorates due to low confidence measures
Solution Approach 1:
The system implements feedback loops where answer quality and confidence measures are continuously monitored. Low-confidence answers identify gaps in the knowledge base, which then trigger targeted content acquisition efforts to add relevant information, thereby improving both coverage and confidence over time
Data Source
AI summary
An approach is provided for automatically ingesting additional corpus based on an interaction history that is mined to identify a question that meets specified answer deficiency criteria, and then generate a second question which is correlated to the first question by requesting additional answer information for answering the first question, where the second question is posted to a forum using a selected persona so that forum responses can be monitored and ingested as additional content in the knowledge base.