G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, (72):65--70, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and Perleberg [7] and BaezaYates and Navarro [8] The preceding methods did not take into account the location of the substrings in the pattern when defining the boundaries for the surviving regions. An obvious improvement on this has been considered independently by Baeza Yates and Navarro [9] and Hyyr [10] both using a requirement that corresponds to applying Lemma 3 without the N O functions. Of these mentioned schemes [8] and [10] concerned the case of indexed searching, while the others were focused on online searching. Both indexed methods used a simple q gram index, which ....

....algorithm of Myers, which we used in checking the surviving regions, is the fastest choice among the current approximate string matching algorithms, and so there is no likely improvement on this aspect either. One approach we have not tested is the use of hierarchical verification ( 16, [9]) with smaller substrings. But we do not expect this method to be too helpful, because it would seem to basically require either looking for 3 grams with no errors or 6 grams with one error, and these two choices would have much larger sets if initial hits than the current method. The tests done ....

Baeza-Yates R and Navarro G. Very fast and simple approximate string matching. Information Processing Letters, 72:65-70 (1999).

....preprocesar el texto, sino tambi en porque es parte fundamental de las t ecnicas de indexaci on. Destacamos a continuaci on nuestros mayores logros en este aspecto y los cap itulos donde se presentan. Este trabajo se ha publicado en [BYN96b, BYN96a, BYN97b, Nav97a, Nav97b, BYN98d, BYN98c, NBY98b, NBY98d, NR98b] y hay otros enviados para publicaci on (esto incluye el material del Cap itulo 2) ffl Desarrollamos en el Cap itulo 5 un nuevo algoritmo para b usqueda aproximada. Este combina el paralelismo de bits con t ecnicas de filtrado (conceptos que se explican en la tesis) Dado que el uso de ....

....not only because in many cases there is no time or space to preprocess the text, but also because it lies at the heart of most indexing techniques. We point out our major achievements in this part. This work has been published in [BYN96b, BYN96a, BYN97b, Nav97a, Nav97b, BYN98d, BYN98c, NBY98b, NBY98d, NR98b] and there are more submitted (this includes the results of Chapter 4) ffl We develop in Chapter 5 a new algorithm for approximate string matching. This combines bitparallelism with filtering techniques (to be explained later) As the use of bit parallelism limits the length of the ....

[Article contains additional citation context not shown here]

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Technical Report TR/DCC-98-6, Dept. of Computer Science, Univ. of Chile, 1998. Submitted. ftp://ftp.dcc.uchile.cl/pub/users/gnavarro/hpexact.ps.gz.

No context found.

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

....that O(s) space is available for the DFAs. The third approach filters the text to quickly discard large text areas, using a necessary condition for an approximate occurrence that is easier to check than the full condition. The areas that cannot be discarded are verified with a classical algorithm [18, 17, 5, 12, 14]. These algorithms achieve sublinear expected time in many cases for low difference ratios, that is, not all text characters are inspected. However, the filtration is not effective for higher ratios. The typical average complexity is O(kn log oe m=m) for ff = O(1= log oe m) The optimal average ....

....superimposition and hierarchical verification, together with a diagonally bit parallelized NFA [3, 13] The implementation was from the original authors. EXP: Partitioning the pattern into k 1 pieces and using hierachical verification with a diagonally bit parallelized NFA in the checking phase [12]. The implementation was from the original authors. Fig. 11 shows the test results for oe = 4, 13 and 52 and m = 30 and 55. This is only a small part of our complete tests, which included oe = 4; 13; 20; 26 and 52, and m = 10; 15; 20; 55. We chose oe = 4 because it behaves like DNA, oe = ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

....at most one piece. So the technique consists of performing a multipattern searching for the pieces without errors, and checking the text surrounding the occurrences of each piece for a complete approximate occurrence of the whole pattern. This leads to the fastest algorithms for low error levels [20, 19]. The property is not true if we add the transposition, because this operation can alter two contiguous pieces at the same time. Much better than splitting the pattern in 2k 1 pieces is to split it in k 1 pieces and leave one unused character between each pair of pieces [19] Under this ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

....that O(s) space is available for the DFAs. The third approach filters the text to quickly discard large text areas, using a necessary condition for an approximate occurrence that is easier to check than the full condition. The areas that cannot be discarded are verified with a classical algorithm [18, 17, 5, 12, 14]. These algorithms achieve sublinear expected time in many cases for low difference ratios, that is, not all text characters are inspected. However, the filtration is not effective for higher ratios. The typical average complexity is O(kn log oe m=m) for ff = O(1= log oe m) The optimal average ....

....superimposition and hierarchical verification, together with a diagonally bit parallelized NFA [3, 13] The implementation was from the original authors. EXP: Partitioning the pattern into k 1 pieces and using hierachical verification with a diagonally bit parallelized NFA in the checking phase [12]. The implementation was from the original authors. Fig. 11 shows the test results for oe = 4, 13 and 52 and m = 30 and 55. This is only a small part of our complete tests, which included oe = 4; 13; 20; 26 and 52, and m = 10; 15; 20; 55. We chose oe = 4 because it behaves like DNA, oe = ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

....that O(s) space is available for the DFAs. The third approach lters the text to quickly discard large text areas, using a necessary condition for an approximate occurrence that is easier to check than the full condition. The areas that cannot be discarded are veri ed with a classical algorithm [18, 17, 5, 12, 14]. These algorithms achieve sublinear expected time in many cases for low di erence ratios, that is, not all text characters are inspected. However, the ltration is not e ective for higher ratios. The typical average complexity is O(kn log m=m) for = O(1= log m) The optimal average ....

....superimposition and hierarchical veri cation, together with a diagonally bit parallelized NFA [3, 13] The implementation was from the original authors. EXP: Partitioning the pattern into k 1 pieces and using hierachical veri cation with a diagonally bit parallelized NFA in the checking phase [12]. The implementation was from the original authors. Fig. 11 shows the test results for = 4, 13 and 52 and m = 30 and 55. This is only a small part of our complete tests, which included = 4; 13; 20; 26 and 52, and m = 10; 15; 20; 55. We chose = 4 because it behaves like DNA, 13 ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65-70, 1999.

....approaches exist on classical approximate pattern matching [Nav00] three of which are of interest for this paper: 1) Dynamic Programming, 2) Bit Parallelism, and (3) Filtration. Refer to the survey [Nav00] In the present work, we adapt a simple but powerful filtration technique due to [WM92, NBY99] to work on compressed text. The idea is that if a pattern is split in k 1 nonoverlapping pieces, then at least one of the pieces must appear unaltered inside every occurrence with at most k errors. The reason is that each error can alter at most one piece. We split the pattern in k 1 ....

....only. We test the following algorithms: the previous work based on dynamic programming over compressed text [KNU00] DP) the previous work based on bit parallelism over compressed text [MKT 00] BP) the naive approach of decompressing plus searching using partitioning into k 1 pieces [NBY99] U PP) and using bit parallelism [Mye98] U BP) and our new algorithms using different multipattern search techniques: the Aho Corasick method [KTS 98] PP AC) our adaptation of the Boyer Moore method [NT00] PP BM) and our adaptation of the bit parallel method [NR99, KTS 99] PP BP) ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

....essentially in where (pattern or text) are the errors assumed to occur. That is, the branch explored in [34, 28] consider P = A and T 0 = B (where T 0 is an occurrence of P in T ) while the branch explored in [20, 18, 36] consider T 0 = A and P = B. A very simple application of the lemma [30] is to split the pattern in k 1 pieces and check the text area surrounding each exact occurrence of a piece in the text. However, Lemma 3.1 can be relaxed to permit the presence of some errors in the pieces: A Hybrid Indexing Method for Approximate String Matching 11 A B A1 A1 A2 A2 A3 ....

....node processing algorithm better, it is because the pattern pieces are so long that the exponential nature of the search cost on suffix trees will make the whole approach useless. On the other hand, there exist (filtering) algorithms that for low error levels are faster than our choice, e.g. [30], but those algorithms need to skip over the text, which is not possible in this setup. The use of the NFA node processing algorithm is only possible because of the simplicity of the DFS traversal. For instance, the idea does not work on the more complex setup of [37, 11] since these need some ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

.... and has a running time of O(mn) 26] In recent years several algorithms have improved the classical one [22] Some improve the worst or average case by using the properties of the dynamic programming matrix [30, 11, 16, 31, 9] Others filter the text to quickly eliminate uninteresting parts [29, 28, 10, 14, 24], some of them being sublinear on average for moderate ff (i.e. they do not inspect all the text characters) Yet other approaches use bit parallelism [3] in a computer word of w bits to reduce the number of operations [33, 35, 34, 6, 19] The problem of approximately searching a set of r ....

....Then we present the three new techniques. In Section 3 we present automaton superimposition , which extends a bit parallel simulation of a nondeterministic finite automaton (NFA) 6] In Section 4 we present exact partitioning , that extends a filter based on exact searching of pattern pieces [7, 6, 24]. In Section 5 we present counting , based on counting pattern letters in a text window [14] In Section 6 we analyze our algorithms and in Section 7 we compare them experimentally. Finally, in Section 8 we give our conclusions. Some detailed analyses are left for Appendices A and B. Although ....

[Article contains additional citation context not shown here]

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 1999. To appear. ftp://ftp.dcc.uchile.cl/pub/users/gnavarro/- hpexact.ps.gz.

....best available algorithm. For this alternative (which we call Best in the experiments) we still use our compression format, because it decompresses faster than Gnu gzip and Unix compress. Our decompression times are 2.09 seconds for WSJ and 1.80 for DNA. The search algorithms used are those of [15, 4, 13], which were the fastest for different m and k values in our texts. On the other hand, the OM type algorithms are unpractical for typical compression ratios (i.e. u=n at most 10) because of their need to keep count of the m k first and last characters of each block. The MP approach does not ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

....at most one piece. So the technique consists of performing a multipattern searching of the pieces without errors, and checking the text surrounding the occurrences of each piece for a complete approximate occurrence of the whole pattern. This leads to the fastest algorithms for low error levels [19, 18]. The property is not true if we add the transposition, because this operation can alter two pieces at the same time. Much better than splitting the pattern in 2k 1 pieces is to split it in k 1 pieces and leave one unused character between each pair of pieces [18] Under this partition a ....

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65--70, 1999.

No context found.

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, (72):65--70, 1999.

No context found.

G. Navarro and R. Baeza-Yates. Very fast and simple approximate string matching. Information Processing Letters, 72:65-70, 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC