Multi-instance Multi-answer Training for Hybrid Table Text QA

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing question answering systems are inadequate for handling hybrid contexts that combine tables and unstructured textual information, particularly in scenarios with multiple instances and multiple answers, leading to inefficiencies and inaccuracies in answering natural language questions.

Innovation Solution

The system employs multi-instance, multi-answer training techniques to enhance text and table question answering. This involves a processor-executable component that includes an answer extractor trained on denoised multiple-instance answer matching data, and a query response component that determines the correct answer based on answer scores and candidate answer data items.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing question answering systems are used for hybrid contexts combining tables and unstructured text, then the system structure is simple, but the accuracy and reliability in handling multiple instances and multiple answers deteriorates

Engineering Contradiction:

Improveaccuracy in handling multiple instances and multiple answersVSAvoidsystem structure complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system segments the question answering task into distinct components: a table question answering module that handles tabular data with multiple instances and answers, and a text question answering module for unstructured text. Each module is specialized to handle its specific data type, improving overall reliability without requiring complete system redesign.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system creates a unified question answering framework that can process both table-based questions and text-based questions through a common architecture. The multi-instance multi-answer mechanism serves multiple functions: it handles exact matches, fuzzy matches, and contextual disambiguation across different data types.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Productivity

If existing question answering systems are used, then the ease of operation is maintained, but the productivity and efficiency in answering complex queries deteriorates

Engineering Contradiction:

Improveefficiency in answering complex queriesVSAvoiduser interaction complexity

Core Design Contradiction:

ProductivityVSEase of operation

Solution Approach 1:

The system performs preliminary processing of both table data and text data into standardized representations before question answering. Training data is pre-processed to include multiple instances and multiple answers, and the model is pre-trained to recognize patterns in hybrid contexts, enabling faster inference without increasing user interaction complexity.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If existing question answering systems are used, then the device complexity is low, but the measurement precision in determining correct answers from multiple candidates deteriorates

Engineering Contradiction:

Improveaccuracy in determining correct answerVSAvoidanswer extraction mechanism complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system implements a feedback mechanism where the answer extractor generates candidate answers with confidence scores, which are then evaluated against multiple instances and multiple answers from training data. The model receives feedback during training on which candidates are correct, improving its ability to distinguish correct answers from incorrect ones in hybrid contexts.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system replaces traditional rule-based answer selection mechanisms with a machine learning-based answer extractor model. This model learns to determine correct answers from multiple candidates by analyzing patterns in training data that includes multiple instances and multiple answers, achieving higher measurement precision without manual rule configuration.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS12210538B2Multi-instance, multi-answer training for table and text question answering

Publication Date: 2025.01.28 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Techniques for enhanced table and text question answering based on multi-instance, multi-answer training are presented. An answer extractor component can determine answer scores associated with candidate answer data items based on analysis of a set of data, comprising row data items of a table and passage data items associated with the table, and a context of a query of the set of data. The answer extractor component can be trained based on application of denoised single-instance and multiple-instance answer matching data associated with contexts to an answer extractor model to generate a trained answer extractor model of the answer extractor component. A query response component can determine a correct answer data item responsive to the query from the candidate answer data items based on the answer scores associated with the candidate answer data items, wherein the candidate answer data items can be reranked based on reweighted answer scores.