Half Transductive Ranking
AISTATS(2010)
摘要
We study the standard retrieval task of rank- ing a xed set of items given a previously un- seen query and pose it as the half transduc- tive ranking problem. The task is transduc- tive as the set of items is xed. Transduc- tive representations (where the vector rep- resentation of each example is learned) al- low the generation of highly nonlinear embed- dings that capture object relationships with- out relying on a specic choice of features, and require only relatively simple optimiza- tion. Unfortunately, they have no direct out- of-sample extension. Inductive approaches on the other hand allow for the representa- tion of unknown queries. We describe algo- rithms for this setting which have the advan- tages of both transductive and inductive ap- proaches, and can be applied in unsupervised (either reconstruction-based or graph-based) and supervised ranking setups. We show em- pirically that our methods give strong perfor- mance on all three tasks.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络