Abstract
A technique for improving answer quality in conversational support chatbots without model retraining is disclosed. The support provided by such chatbots often falls short of that provided by experienced human support agents. Much of a human agent’s effectiveness comes from experience, through which an understanding of what tends to help is built. Replicating this experience by training a model on past interactions would require constant and costly retraining, since new product features continually change the range of possible queries. In the disclosed approach, support logs are condensed on a recurring basis into a database of issues, recommendations, and user feedback, such as satisfaction scores and thumbs up or down ratings. For each user query, the database, or a relevant subset, is supplied to the model as context, and highly rated recommendations are prioritized. Lightweight, in-field reinforcement is thereby achieved without model post-training.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Dupree, Andrew and Santarelli, Halli, "Support History as Reinforcement Context for Conversational Support Chatbots", Technical Disclosure Commons, (October 28, 2026)
https://www.tdcommons.org/dpubs_series/12507