Fine-grained relation extraction data enhancement method based on keyword prompt
Through the fine-grained relational extraction data enhancement method based on keyword hints, the problem of insufficient data generation in the prior art is solved, the generation of sentences with rich context diversity is realized, and the generalization ability of the relational extraction model is improved.
Patent Information
- Application Number
- CN202411699641.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-26
- Publication Date
- 2025-03-28
AI Technical Summary
Existing relational extraction data enhancement methods are difficult to maintain the relationship dependence between the generated data and the original data, resulting in semantic and relational bias of the generated data, while lacking sufficient contextual diversity, limiting the generalization ability of the model.
A fine-grained relational extraction data enhancement method based on keyword cues is adopted, and keyword cues closely related to relationship instances are obtained through statistical algorithms and semantic similarity. A prompt template based on keyword cues is constructed. Combined with the generation ability of large language models, sentences with rich context diversity are generated. At the same time, syntactic post-processing of distance editing through syntax tree to ensure the syntax correctness and consistency of the generated data.
Effectively maintain the relationship dependence consistency of generated sentences in the context, improve the diversity and quality of generated data, and enhance the generalization ability of relationship extraction models.