Adversarial Event Detection for Reliable Text Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text-based event detection methods require manual labeling of training data, leading to inefficiencies and reduced accuracy due to the need for extensive human intervention.
Innovation Solution
An adversarial network is trained using a first data set of standard event instances and a second data set of non-standard event instances, comprising a generator and discriminator, to automatically identify reliable event instances and minimize noise, thereby enhancing detection efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual labeling is used to prepare training data, then the quality and reliability of training data is improved, but the time consumption and labor cost increase significantly
Solution Approach 1:
The patent uses pre-labeled standard event instances as templates to automatically generate non-standard event instances through text substitution and transformation. This copying approach allows the model to learn from standardized patterns without requiring manual labeling of each individual instance, thus maintaining data quality while reducing preparation time
Solution Approach 2:
The patent performs preliminary labeling of standard event instances in advance, and then uses these pre-labeled instances to automatically generate the training dataset. This preliminary action eliminates the need for manual labeling during the main training process, significantly reducing time consumption while preserving data quality
2Measurement precision
If manually labeled training data is used, then the accuracy of event detection is improved, but the productivity and efficiency of model training deteriorates
Solution Approach 1:
The system automatically generates non-standard event instances from standard event instances using text transformation rules and patterns. This self-service mechanism allows the model to create its own training data without external manual intervention, improving productivity while maintaining the accuracy benefits of structured labeling
Solution Approach 2:
The patent transforms standard event instances into non-standard instances by changing parameters such as text structure, event types, and contextual information while preserving the underlying labeling patterns. This parameter transformation enables efficient data generation that maintains detection accuracy
3Stability of the object's composition
If only standard event instances are used for training, then the model learns consistent patterns, but the adaptability to real-world diverse events deteriorates
Solution Approach 1:
The patent segments event instances into standard templates and variable components. By separating the consistent structural patterns from the variable event-specific content, the model can learn stable patterns from standard instances while adapting to diverse real-world events through the variable components in non-standard instances
Solution Approach 2:
The patent adds a new dimension to the training data by creating non-standard event instances that vary in text structure, event types, and contextual information while maintaining the core labeling patterns. This dimensional expansion allows the model to maintain consistency learning while gaining adaptability to diverse events
Data Source
AI summary
A training method includes obtaining a first data set and a second data set, each of the first data set and the second data set including event instances, the event instances include text and events corresponding to the text. The training method also includes training an adversarial network using the first data set and the second data set, the adversarial network includes processing circuitry configured as a generator and a discriminator. The discriminator is configured to output first reliable probabilities of the event instances in the first data set, and second reliable probabilities of the event instances inputted by the generator. A loss function of the adversarial network is used to adjust a parameter of the adversarial network, to maximize the first reliable probabilities and minimize the second reliable probabilities. The method further includes obtaining, by the trained adversarial network, a reliable event instance in the second data set.