Similarity search has become a principal operation not only in databases but also in diverse application domains. Very large datasets, however, pose a big challenge on its enormous volume-processing capability. In order to deal with the challenge, we propose a two-level clustering approach aiming at supporting fast similarity searches in massive datasets. In addition, we embed some pruning and filtering strategies into our methods so that redundancy-free data, data accuracy, inessential data accesses, unnecessary distance computations, and other following consequences are taken into account. Furthermore, we validate our methods by a series of empirical experiments in real big datasets. The results show that our approach performs better than the two inverted index-based approaches, especially when given big query batches.
|Publikationsstatus||Veröffentlicht - 2016|