Abstract
Manual creation of themed navigational content for user engagement campaigns can be a slow, resource-intensive, and unscalable process, while general-purpose generative models may lack domain-specific safety guidelines and struggle with thematic consistency. A disclosed system can address these challenges through a dual-model, generator-evaluator framework. A primary generative model can create multiple thematic text candidates for a given instruction. A separate, secondary evaluator model may then assess these candidates using a hybrid evaluation process that combines model-based semantic analysis for attributes, such as safety and logical contradiction, with function-based checks for criteria like length and semantic similarity. The system can iteratively refine the output by re-initiating this generation-evaluation cycle if a candidate does not meet predefined quality standards. This approach can provide a scalable method for producing thematically consistent content suitable for deployment in applications.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Katz, Roy, "System for Generating Themed Content Using a Generator-Evaluator Framework", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/9203