Event Extraction via Segmented Detection and Reading Comprehension
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current event extraction methods in the financial sector face challenges in accurately extracting event types and elements from document-level text due to issues like escape symbols, element confusion, and the presence of uncanonical expressions, leading to low accuracy and performance.
Innovation Solution
The proposed method involves constructing an event detection model and a reading comprehension model to determine event types and elements by splicing text with predefined event types, using semantic extraction models and binary classification processing to identify event bodies and elements, thereby improving the accuracy and efficiency of event extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual analysis is used to process financial documents, then accuracy and detail can be maintained, but processing efficiency and scalability are severely limited
Solution Approach 1:
The event extraction process is segmented into multiple stages: event type detection using a classification model, event body extraction using a reading comprehension model, and event element extraction using another reading comprehension model. This segmentation allows each stage to focus on specific tasks, improving overall efficiency while maintaining accuracy through specialized processing at each step.
Solution Approach 2:
A query corpus is introduced as an intermediary component that bridges the text to be processed and the extraction models. The query corpus contains predefined event types and related information that guide the models through the extraction process, enabling accurate identification of event types and elements while improving processing efficiency.
2Measurement precision
If traditional event extraction methods are used, then processing speed might be maintained, but accuracy deteriorates due to escape symbols, element confusion, and uncanonical expressions
Solution Approach 1:
The query corpus is constructed in advance with predefined event types and their corresponding characteristics. This preliminary preparation allows the extraction models to reference established patterns during processing, improving accuracy in identifying event types and elements while reducing the complexity of real-time decision-making.
Solution Approach 2:
The system incorporates feedback mechanisms where the extraction models continuously reference the query corpus and adjust their predictions based on the context. This feedback loop enables the models to resolve ambiguities caused by escape symbols, element confusion, and uncanonical expressions, improving accuracy without requiring overly complex processing architecture.
3Loss of information
If comprehensive event extraction is performed, then information completeness is improved, but processing time increases
Solution Approach 1:
The extraction process is divided into sequential stages: first determining event types, then extracting event bodies, and finally identifying event elements. This segmentation enables the system to process information in manageable chunks, maintaining completeness by systematically addressing each aspect of event extraction while controlling processing time through staged computation.
Solution Approach 2:
The query corpus is prepared in advance with comprehensive event type definitions and extraction patterns. This preliminary preparation allows the models to efficiently extract complete event information during processing, as the extraction criteria are pre-established, reducing the time needed for comprehensive information extraction.
Data Source
AI summary
The application discloses an event extraction method, apparatus, device and storage medium, the method including acquiring a text to be processed, inputting the text to be processed into an event detection model to determine an event type involved in the text to be processed, constructing a first query corpus based on the event type, inputting the text to be processed and the first query corpus into a reading comprehension model to determine an event body corresponding to the event type and event elements corresponding to the event body, obtaining an event extraction result of the text to be processed based on the event body and the event elements. This method accurately determines the event type and the event elements in the text to be processed, and is adapted to event extraction in a document level text, improving the overall performance of event extraction and promoting the efficiency of event extraction.