英语翻译Coactive learning proceeds in the following online fash-
英语翻译
Coactive learning proceeds in the following online fash- ion.In each step,a ranking is presented to the user that (approximately) maximizes the current estimate of the sub- modular utility function.As feedback,the algorithm ob- serves the (possibly diverse) set of documents the user reads in the presented ranking.After receiving this feedback,the algorithm updates its model.Even though we allow user feedback to be imperfect,noisy,and only “weakly informa- tive”(in a specific sense),we are able to prove guarantees on the performance of the algorithm.Unlike the theorems in [14],our guarantees apply even though submodular models only allow for approximate inference.Finally,experiments demonstrate the empirical effectiveness of the proposed ap- proach in learning both relevance and diversity.