面向文本语义推理的标签感知去偏因果推断方法和系统
By constructing a causal graph and training the model with a total loss function, the problem of insufficient modeling of biased information in text semantic reasoning is solved, and refined processing of biased information is achieved, thereby improving the accuracy and applicability of the model's unbiased reasoning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI UNIV OF TECH
- Filing Date
- 2023-07-19
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies in text semantic reasoning either lack sufficient granularity in modeling biased information or focus too much on removing biased information, resulting in poor model generalization performance and decreased accuracy in predicting semantic inference relationships.
By constructing a causal graph, biased information representations in labels are obtained. The average causal effect and spurious association paths are utilized, and the model is trained using the total loss function to achieve the debiased causal effect from input to output. This method is applicable to various text semantic representation and inference methods.
It achieves refined modeling of biased information, improves the accuracy and generalization performance of unbiased reasoning, and ensures the accuracy and applicability of semantic reasoning.
Smart Images
Figure CN116894439B_ABST