一种基于预测分布几何约束的情感分类优化方法及系统
By introducing a joint loss function with geometric constraints on the prediction distribution and a warm-up strategy into the sentiment classification model, the prediction distribution structure is optimized, which solves the problem of prediction instability in multi-class sentiment tasks and improves the stability and generalization ability of the model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUNAN INST OF INFORMATION TECH
- Filing Date
- 2026-05-06
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies lack structural constraints on the predicted probability distribution of model outputs in multi-class, fine-grained emotion tasks, resulting in prediction instability and insufficient generalization ability. In particular, lightweight models are prone to getting stuck in local optima or training oscillations.
By introducing a joint loss function based on the geometric constraints of the prediction distribution, using Bregman divergence to measure the geometric distance of the prediction probability distribution, and combining a warm-up strategy to gradually introduce geometric constraints during training, the prediction distribution structure of the sentiment classification model is optimized.
It significantly improves the stability and consistency of sentiment classification results and enhances the model's generalization ability. In particular, by reasonably controlling the timing of introducing geometric constraints, it avoids the performance degradation of the lightweight model in the early stages of training and achieves more stable classification performance.
Smart Images
Figure CN122132565B_ABST