Method of generating training data for chemical reactions, computing device
By mapping reagent roles to fine-grained functional roles using a large language model and combining chemical mechanism knowledge to generate chemical reaction training data, the problems of data imbalance and insufficient accuracy in existing technologies are solved, a high-quality training dataset is achieved, and the accuracy and generalization ability of the model in chemical reaction prediction and generation tasks are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI ARTIFICIAL INTELLIGENCE INNOVATION CENT
- Filing Date
- 2026-05-22
- Publication Date
- 2026-07-24
AI Technical Summary
The lack of high-quality positive and negative samples in existing chemical reaction training data leads to insufficient accuracy and generalization ability of the model in chemical reaction prediction and generation tasks. In particular, the inference ability is limited when faced with implicit conditions, and the negative samples generated by existing methods are of low quality or cannot be scaled up.
By mapping reagent roles to fine-grained functional roles through a large language model and combining chemical mechanism knowledge, high-quality positive and negative samples are generated. Perturbation strategies are used to remove or replace reagents under chemical reaction conditions to generate negative sample data that conforms to the chemical mechanism.
It significantly improves the quality and generalization ability of training data, enhances the accuracy and reliability of the model in chemical reaction prediction and generation tasks, and solves the problems of data imbalance and accuracy in existing technologies.
Smart Images
Figure CN122245480B_ABST