Abstract
Generative models that utilize external information sources may not include mechanisms to quantitatively measure the contribution of each source document to a generated output. Disclosed are systems and methods for quantitatively attributing source importance. A technique can involve calculating numerical importance scores for constituent parts of a model's output, for instance, by analyzing internal model attention scores or semantic similarity to a user prompt. A remapping process then can trace these scored parts back to their original source documents to assign an overall importance score to each source. This quantitative attribution data can provide a basis for feedback mechanisms for content creators, facilitate targeted data acquisition to address knowledge gaps, and enhance transparency for users by ranking cited sources based on their contribution.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
N/A, "Quantitative Importance Attribution for Source Documents in Generative Models", Technical Disclosure Commons, (February 11, 2026)
https://www.tdcommons.org/dpubs_series/9315