Entity Linking via Dual-Model Feature Representation Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing entity linking technologies face challenges in generating accurate semantic feature representations, leading to low accuracy in entity recognition and disambiguation in text and multi-modal content.
Innovation Solution
A data processing method and apparatus that utilize a two-tower model, where a context encoding model generates context feature representations and an entity encoding model generates semantic feature representations, with a feature representation loss determined by similarity between the two, allowing for model parameter adjustment to improve representation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional entity linking methods are used to generate semantic feature representations, then the process is simple, but the accuracy of the generated semantic feature representation is low
Solution Approach 1:
The patent divides the encoding task into two separate models: a context encoding model for processing context information and an entity encoding model for processing semantic feature data. This segmentation allows each model to specialize in its specific task, improving the accuracy of semantic feature representations while maintaining manageable complexity through modular architecture.
Solution Approach 2:
The patent introduces a feature representation loss function as an intermediary mechanism that bridges the context encoding model and entity encoding model. This loss function calculates similarity between context feature representations and semantic feature representations, enabling the entity encoding model to learn from contextual information without directly coupling the two models, thus improving accuracy while controlling complexity.
2Reliability
If semantic feature representation accuracy is improved through complex encoding models, then entity linking accuracy improves, but the model training complexity and computational cost increase
Solution Approach 1:
The patent employs preliminary action by pre-training the context encoding model on context information before using it to guide the entity encoding model. This staged approach allows the system to build up representations incrementally, improving entity linking accuracy while managing training complexity through a structured, multi-phase training process rather than attempting to optimize everything simultaneously.
Solution Approach 2:
The patent implements feedback mechanisms through the feature representation loss, which provides continuous guidance during training by measuring the similarity between context-based representations and entity-based representations. This feedback loop allows the entity encoding model to iteratively improve its semantic feature representations based on contextual information, enhancing reliability while keeping training complexity manageable through directed optimization.
Data Source
AI summary
In a data processing method, a first training sample is received. The first training content data is encoded by using a first context encoding model to obtain a first context feature representation corresponding to the first training entity. The first semantic feature data is encoded by using a to-be-trained first entity encoding model to obtain a first semantic feature representation corresponding to the first training entity. A first feature representation loss is determined based on first similarity information between the first context feature representation and the first semantic feature representation. First model parameters of the to-be-trained first entity encoding model are adjusted based on the first feature representation loss to obtain a trained first entity encoding model. Apparatus and non-transitory computer-readable storage medium counterpart embodiments are also contemplated.