LSTM Multi-Encoder for Relationship Identification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional methods for identifying relationships from text data are inefficient due to manual data labeling and semi-automated approaches that generate noisy alignments, requiring additional manual processing to produce reliable training data for relationship identification models.

Innovation Solution

A system and method that aligns sentences with previously stored facts, computes confidence scores for these alignments, and uses them, along with extracted features, to train a relationship identification model using a Long Short-Term Memory (LSTM)-based multi-encoder, allowing for the use of all alignments in training while weighting them by confidence scores to improve model reliability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual labeling is used to generate training data, then the quality and reliability of training data is improved, but the time consumption and efficiency deteriorate significantly

Engineering Contradiction:

Improvequality of training dataVSAvoidtime consumption

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system uses automatically generated alignments from the alignment unit as training data without requiring manual labeling. The model trains on these self-generated alignments, making the system self-sufficient and eliminating the time-consuming manual labeling process while maintaining acceptable training data quality.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The alignment unit acts as an intermediary that automatically generates alignments between sentences and facts, which then serve as training data. This intermediary component bridges the gap between raw text data and the relationship identification model, providing training data without manual intervention.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Productivity

If automated alignment is used to generate training data, then the efficiency is improved, but the quality and reliability of training data deteriorates due to noisy alignments

Engineering Contradiction:

Improveefficiency of training data generationVSAvoidquality of training data

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system generates all possible alignments between sentences and facts using the alignment unit, including noisy or incorrect alignments. Rather than filtering for only high-quality alignments, the system accepts excessive alignments and relies on the model training process and confidence scoring to handle the noise, thereby maintaining high efficiency.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system computes confidence scores for each alignment to provide feedback on the quality of the automated alignment. This confidence score mechanism allows the system to identify and weight reliable alignments higher during training, thereby maintaining data quality while preserving the efficiency benefits of automated alignment generation.

Inventive Principle:

Principle #23Feedback

3Productivity

If all alignments are used in training without filtering, then the productivity is improved, but the measurement precision of relationship identification deteriorates due to noisy data

Engineering Contradiction:

Improvetraining speedVSAvoidaccuracy of relationship identification

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system changes the parameter of training data weighting by introducing confidence scores. Instead of treating all alignments equally or filtering them, the system adjusts the weight of each alignment based on its confidence score, allowing noisy alignments to have lower influence while maintaining the benefit of using all alignments for training.

Inventive Principle:

Principle #35Parameter changes

4Reliability

If manual selection of alignments is performed to remove noisy data, then the quality of training data is improved, but the complexity of the process and time consumption increase

Engineering Contradiction:

Improvequality of training dataVSAvoidprocess complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system replaces the mechanical process of manual alignment selection with an automated confidence score computation mechanism. The confidence score system automatically evaluates and weights alignments, substituting the need for manual inspection and selection, thereby reducing process complexity while maintaining data quality.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS10902350B2System and method for relationship identification

Publication Date: 2021.01.26 YAHOO ASSETS LLC

AI summary

The present teaching relates to method, system, and medium for generating training data for generating a relationship identification model. Sentences are received as input. Each of the sentences is aligned with a fact previously stored to create an alignment. Confidence scores for the alignments are computed and then used, together with the alignments to train a relationship identification model.