Probabilistic Fact-Checking for Relational Database Claims

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional approaches for verifying the correctness of statements in text documents are limited in accuracy and efficiency, relying on human fact checkers, natural language query interfaces, or textual web query results, which are inadequate for reliable fact-checking.

Innovation Solution

A processing platform that extracts claim keywords from text documents, identifies query fragments from relational databases, and uses probabilistic inferencing to evaluate candidate queries against the database, determining the consistency of claims with relational data sets, thereby providing accurate and efficient fact-checking.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional approaches (human fact checkers, natural language query interfaces, textual web query results) are used to verify statements in text documents, then the verification process can be performed, but the accuracy and efficiency are substantially limited

Engineering Contradiction:

Improvefact-checking accuracyVSAvoidfact-checking efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces conventional mechanical approaches (human fact checking, natural language query interfaces) with a probabilistic inferencing system that directly queries relational databases. This substitution enables automated evaluation of candidate queries against raw relational data, significantly improving both accuracy and efficiency of fact-checking operations.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces probabilistic inferencing as an intermediary layer between text claims and database verification. This intermediary generates and evaluates multiple candidate queries with associated probabilities, enabling systematic comparison of claims against relational data while handling uncertainty in the verification process.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If human fact checkers are used to verify statements, then verification can be performed with contextual understanding, but the process is time-consuming and less efficient

Engineering Contradiction:

Improveverification reliabilityVSAvoidfact-checking time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent enables the verification system to perform self-service by automatically generating candidate queries, evaluating them against the database, and determining claim consistency without human intervention. The probabilistic inferencing system autonomously handles the entire verification process, eliminating time-consuming manual fact-checking while maintaining reliability through systematic probabilistic evaluation.

Inventive Principle:

Principle #25Self-service

3Adaptability or versatility

If natural language query interfaces to knowledge bases are used, then fact verification can be performed, but the functionality is limited and accuracy is reduced

Engineering Contradiction:

Improvefact-checking functionalityVSAvoidfact-checking accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent creates a universal verification system that works across multiple data types and query scenarios by directly interfacing with relational databases rather than being constrained to specific knowledge bases or natural language interfaces. The probabilistic inferencing framework provides multi-functional capability to handle various claim types while maintaining high accuracy through direct database evaluation.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11113275B2Verifying text summaries of relational data sets

Publication Date: 2021.09.07 CORNELL UNIVERSITY

AI summary

A processing platform in illustrative embodiments comprises one or more processing devices each including at least one processor coupled to a memory. The processing platform is configured to obtain a text document containing at least one claim about data of a relational database. Claim keywords are extracted from the text document, and query fragments are identified based at least in part on indexing of one or more relational data sets of the relational database. The relevance of the claim keywords to the query fragments is determined, and candidate queries are identified based at least in part on probabilistic inferencing from the determined relevance of the claim keywords to the query fragments. The candidate queries are evaluated against the relational database, and consistency of the claim with the data of the relational database is determined based at least in part on results of the evaluation of the candidate queries.