We present the AS-Index, a new index structure for exact string search in disk resident databases. AS-Index relies on a classical inverted file structure, whose main innovation is a probabilistic search based on the p...We present the AS-Index, a new index structure for exact string search in disk resident databases. AS-Index relies on a classical inverted file structure, whose main innovation is a probabilistic search based on the properties of algebraic signatures used for both n-grams hashing and pattern search. Specifically, the properties of our signatures allow to carry out a search by inspecting only two of the posting lists. The algorithm thus enjoys the unique feature of requiring a constant number of disk accesses, independently from both the pattern size and the database size. We conduct extensive experiments on large datasets to evaluate our index behavior. They confirm that it steadily provides a search performance proportional to the two disk accesses necessary to obtain the posting lists. This makes our structure a choice of interest for the class of applications that require very fast lookups in large textual databases. We describe the index structure, our use of algebraic signatures, and the search algorithm. We discuss the operational trade-offs based on the parameters that affect the behavior of our structure, and present the theoretical and experimental performance analysis. We next compare the AS-Index with the state-of-the-art alternatives and show that 1） its construction time matches that of its competitors, due to the similarity of structures, 2） as for search time, it constantly outperforms the standard approach, thanks to the economical access to data complemented by signature calculations, which is at the core of our search method.展开更多
文摘We present the AS-Index, a new index structure for exact string search in disk resident databases. AS-Index relies on a classical inverted file structure, whose main innovation is a probabilistic search based on the properties of algebraic signatures used for both n-grams hashing and pattern search. Specifically, the properties of our signatures allow to carry out a search by inspecting only two of the posting lists. The algorithm thus enjoys the unique feature of requiring a constant number of disk accesses, independently from both the pattern size and the database size. We conduct extensive experiments on large datasets to evaluate our index behavior. They confirm that it steadily provides a search performance proportional to the two disk accesses necessary to obtain the posting lists. This makes our structure a choice of interest for the class of applications that require very fast lookups in large textual databases. We describe the index structure, our use of algebraic signatures, and the search algorithm. We discuss the operational trade-offs based on the parameters that affect the behavior of our structure, and present the theoretical and experimental performance analysis. We next compare the AS-Index with the state-of-the-art alternatives and show that 1） its construction time matches that of its competitors, due to the similarity of structures, 2） as for search time, it constantly outperforms the standard approach, thanks to the economical access to data complemented by signature calculations, which is at the core of our search method.
