An adaptive sliding window based continuous Top-K dominating queries

Abstract

Top-K dominating query selects k data objects and influences the highest number of objects in a dataset. This is a decision supportable query since it provides data analysts a best way for finding significant objects. This search is not only for the earlier examination of large upper bounds that leads to earlier identification of results, but also eliminates partial dominance relationship between the entries, which facilitates the computation of tight lower bounds for these candidates. This is sufficient for applications operating on static or almost static dataset, where updates are rare. As many modern applications adopt the streaming model of computation, they require continuous query processing algorithms to refresh the query result. The existing work compares the event based algorithm, brute force approach and advanced algorithm. the event based algorithm and advanced algorithm consistently outperform the baseline algorithm. Eventually advanced algorithm shows the best overall performance being orders of magnitude faster than the brute force approach. it also implements two approximate algorithms approximate hoeffding bound algorithm and approximate minimum score algorithm which sacrifice accuracy for faster computation. Approximate hoeffding bound algorithm based on sampling offers probabilistic guarantees regarding the approximation error. on the other hand, approximate minimum score algorithm based on event pruning leads to faster processing with less accuracy compared to approximate hoeffding bound algorithm. the advanced algorithm, approximate hoeffding bound algorithm and approximate minimum score algorithm can work in combination. the enhancement work proposes the improved event based algorithm for finding dominant item set over segmented sliding windows in a data stream instead of using advanced algorithm. Experimental results show that the proposed method is quite efficient, scalable and achieves high accuracy.