Long-Range Event Relation Extraction via Synthetic Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document analysis systems face challenges with computational accuracy and operational flexibility in determining long-range event relations within digital documents, primarily due to limited training data consisting of short-range event relations, leading to inaccuracies and inflexibility in handling event pairs separated by greater distances.
Innovation Solution
The system generates a synthetically augmented long-range event relation dataset by inserting contextually coherent synthetic sentences between event pairs in digital documents, using a generative language model to expand the range of event relations beyond the initial short-range threshold, thereby training an event relation extraction model capable of identifying long-range relations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If existing document analysis systems are trained only on short-range event relation data, then the training data requirement is reduced, but the accuracy and flexibility in determining long-range event relations deteriorates
Solution Approach 1:
The patent uses a generative language model to create synthetic event relation datasets that replicate the structure and patterns of real long-range event relations. These synthetic datasets copy the essential characteristics of actual event relations while providing the necessary training data that would otherwise be unavailable or insufficient, thereby enabling accurate long-range event relation extraction without requiring extensive real-world data
Solution Approach 2:
The system performs preliminary data generation and augmentation before the actual event relation extraction task. By pre-generating synthetic long-range event relation datasets and pre-training the model on this augmented data, the system prepares the necessary learning material in advance, allowing the model to accurately handle long-range event relations when they actually need to be extracted from documents
2Device complexity
If existing document analysis systems are trained only on short-range event relations, then the system complexity is reduced, but the adaptability to various event pair distances deteriorates
Solution Approach 1:
The patent modifies the training data parameters by creating synthetic datasets with varying event relation distances, contexts, and structures. This parameter transformation allows the model to learn patterns across different event pair distances without fundamentally changing the system architecture, thereby achieving adaptability to various event distances while maintaining manageable system complexity through data-level transformation rather than system-level complexity
3Measurement precision
If synthetic sentences are inserted to create long-range event relation datasets, then the event relation extraction accuracy for long-range relations is improved, but the data generation process complexity increases
Solution Approach 1:
The generative language model serves as an intermediary between the existing short-range event relation data and the desired long-range event relation training data. This intermediary component automatically transforms and augments the data, generating synthetic long-range event relations by inserting contextual sentences between event pairs, thereby achieving high accuracy in long-range event relation extraction while managing data generation complexity through automated intelligent transformation rather than manual complex processing
Data Source
AI summary
The present disclosure relates to systems, methods, and non-transitory computer-readable media that generates a long-range event relation dataset by augmenting a digital document with a set of synthetic sentences. For example, the disclosed systems access a digital document from a short-range event relation dataset that includes an event pair. In some embodiments, the disclosed systems generate a set of synthetic sentences utilizing a generative language model for inserting within the digital document between the event pair to satisfy a long-range event relation threshold. In these or other embodiments, the disclosed systems generate a long-range event relation dataset by augmenting the digital document within the short-range event relation dataset to include the set of synthetic sentences. In certain cases, the disclosed systems generate an event relation extraction model to determine long-range event relations by learning model parameters for the event relation extraction model from the long-range event relation dataset.