An aspect-level sentiment triple extraction method based on a diffusion model
By employing a boundary denoising diffusion process based on a diffusion model and a contrastive denoising training strategy, the problem of word generation limitations in generative models is solved, achieving efficient and accurate prediction of multi-word aspect/opinion terms and improving the performance of aspect-level sentiment triple extraction.
Patent Information
- Application Number
- CN202411194940.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2026-06-26
- Estimated Expiration
- 2044-08-28
AI Technical Summary
Existing generative models focus only on generating individual words during autoregressive decoding, limiting their ability to utilize holistic semantics when dealing with multi-word aspects/opinion terms.
We adopt a non-autoregressive diffusion model framework, defining the aspect-level sentiment triple extraction task as a boundary denoising diffusion process. By introducing Gaussian noise to simulate boundary uncertainty and introducing a contrastive denoising training strategy, we directly model the boundary index and use comprehensive contextual information for prediction.
It significantly improves the accuracy of aspect and opinion term boundary prediction, enhances the robustness and predictive performance of the model, and reduces the generation of erroneous triples.