A Study on Data Mining Optimizing Visual Search Re-Ranking Via Pair Wise Learning Image Datas

Conventional approaches to visual search re-ranking empirically take the “classification performance” as the optimization objective, in which each visual document is determined whether relevant or not, followed by a process of increasing the order of relevant documents. In this project, we first reestablish the fact that: the classification performance fails to produce a globally optimal ranked list. Hence, we formulate re-ranking as an optimization problem, in which a ranked list is globally optimal only if any arbitrary two documents in the list are correctly ranked in terms of relevance. This is different from existing Approaches which simply classify a document as “relevant” or not. To find the optimal ranked list, we convert the individual documents to “document pairs”, Each pair is represented as an “ordinal relation.” Then, we find the optimal document pairs which can maximally preserve the initial rank order while simultaneously keeping the consistency with the auxiliary knowledge mined from query examples and web resources as much as possible. We develop two pair wise re-ranking methods, difference pair wise re-ranking (DP-re-ranking) and exclusion pair wise reranking (EP-re-ranking), to obtain the relevant relation of each document pair. Finally, a round robin criterion is explored to recover the final ranked list.