关键词: MapReduce, Q-sample, 大数据, 数据清洗, 相似连接

Abstract: Local similarity join can finds all local similar pairs from sets quickly,which is a basic operation in many areas,such as gene sequence alignment,near duplicate detection,data cleaning and so on.This paper focused on designing similarity join parallel algorithm with MapReduce,and proposed a Q-sample-based algorithm to solve the locating problem of local similarity join.The proposed algorithm employs a filter-verify based framework.In filter stage,a Q-sample partition scheme is adopted to generate high-quality signatures without losing any true pairs,and then more dissimilar string pairs are discarded.In verify stage,the LS-Join’s backward-forward verification method is improved with the technique of removing redundant match,combining consecutive match and combining non-consecutive match.In the experiments,the performances of the proposed algorithm with different size of datasets or different value of edit distances are scaled.Experimental results show that the proposed algorithm outperforms the current excellent algorithm LS-Join on big dataset.Theoretical analysis and experimental result demonstrate that the performance of local similarity join is improved by using the techniques of the proposed algorithm.

Key words: Big data, Data cleaning, MapReduce, Q-sample, Similarity join