Algorithms to determine matching values - 23.1

spectrum-inline

Product type
Software
Portfolio
Enrich
Integrate
Locate
Verify
Product family
Spectrum
Product
Spectrum > Quality > Context Graph
Spectrum > Discovery
Spectrum > Web Services
Spectrum > Spatial > Spectrum Spatial
Spectrum > Quality > Addressing
Spectrum > Dataflow Designer
Spectrum > Geocoding > Enterprise Tax
Spectrum > Data Integration
Spectrum > Quality > Spectrum Quality
Spectrum > Spectrum Platform
Spectrum > GeoEnrichment
Version
23.1
Language
English
Product name
Precisely Spectrum
Title
spectrum-inline
First publish date
2007

Acronym
Determines whether a business name matches its acronym by looking for acronym data; otherwise it creates an acronym using the first character of every word. Example: Internal Revenue Service and its acronym IRS would be considered a match and return a match score of 100.
Character Frequency
Determines the frequency of occurrence of each character in a string and compares the overall frequencies between two strings.
Daitch-Mokotoff Soundex
Phoenetic algorithm that allows greater accuracy in matching of Slavic and Yiddish surnames with similar pronunciation but differences in spelling. Coded names are six digits long, and multiple possible encodings can be returned for a single name. This option was developed to respond to limitations of Soundex in the processing of Germanic or Slavic surnames.
Date
Compare date fields regardless of the date format in the input records. Click Edit in the Options column to specify the following:
  • Require Month: prevents a date that consists only of a year from matching
  • Require Day: prevents a date that consists only of a month and year from matching
  • Match Transposed MM/DD: where month and day are provided in numeric format, compares suspect month to candidate day and suspect day to candidate month as well as the standard comparison of suspect month to candidate month and suspect day to candidate day
  • Prefer DD/MM/YYYY format over MM/DD/YYYY: contributes to date parsing in cases where both month and day are provided in numeric format and their identification can not be determined by context. For example, given the numbers 5 and 13, the parser will automatically assign 5 to the month and 13 to the day because there are only 12 months in a year. However, given the numbers 5 and 12 (or any two numbers 12 and under), the parser will assume whichever number is first to be the month. Checking this option will ensure that the parser reads the first number as the day rather than the month.
  • Range Options—Overall: allows you to set the maximum number of days between matching dates. For example, if you enter an overall range of 35 days and your candidate date is December 31st, 2000, a suspect date of February 5, 2001 would be a match, but a suspect date of February 6 would not. If you enter an overall range of 1 day and your candidate date is January 2000, a suspect date of 1999 would be a match (comparing December 31, 1999) but a suspect date of January 2001 would not.
  • Range Options—Year: allows you to set the number of years between matching dates, independent of month and day. For example, if you enter a year range of 3 and your candidate date is January 31, 2000, a suspect date of January 31, 2003, would be a match but a suspect date of February 2003 would not. Similarly, if your candidate date is 2000, a suspect date of March 2003 would be a match because months are not in conflict and it's within the three-year range.
  • Range Options—Month: allows you to set the number of months between matching dates, independent of year and day. For example, if you enter a month range of 4 and your candidate date is January 1, 2000, a suspect date of May 2000 is a match because there is no day conflict and it's within the four-month range, but a suspect date of May 2, 2000, is not, because the days conflict.
  • Range Options—Day: allows you to set the number of days between matching dates, independent of year and month. For example, if you enter a day range of 5 and your candidate date is January 1, 2000, a suspect date of January 2000 is a match because there is no day conflict but a suspect date of December 27, 1999, is not, because the months conflict.
Double Metaphone
Determines the similarity between two strings based on a phonetic representation of their characters. Double Metaphone is an improved version of the Metaphone algorithm, and attempts to account for the many irregularities found in different languages.
Edit Distance
Determines the similarity between two strings based on the number of deletions, insertions, or substitutions required to transform one string into another.
Euclidean Distance
Provides a similarity measure between two strings using the vector space of combined terms as the dimensions. It also determines the greatest common divisor of two integers. It takes a pair of positive integers and forms a new pair that consists of the smaller number and the difference between the larger and smaller numbers. The process repeats until the numbers are equal. That number then is the greatest common divisor of the original pair. For example, 21 is the greatest common divisor of 252 and 105: (252 = 12 × 21; 105 = 5 × 21); since 252 − 105 = (12 − 5) × 21 = 147, the GCD of 147 and 105 is also 21.
Exact Match
Determines if two strings are the same.
Initials
Used to match initials for parsed personal names.
Jaro-Winkler Distance
Determines the similarity between two strings based on the number of character replacements it takes to transform one string into another. This option was developed for short strings, such as personal names.
Keyboard Distance
Determines the similarity between two strings based on the number of deletions, insertions, or substitutions required to transform one string to the other, weighted by the position of the keys on the keyboard. Click Edit in the Options column to specify the type of keyboard you are using: QWERTY (U.S.), QWERTZ (Austria and Germany), or AZERTY (France).
Koeln
Indexes names by sound as they are pronounced in German. Allows names with the same pronunciation to be encoded to the same representation so that they can be matched, despite minor differences in spelling. The result is always a sequence of numbers; special characters and white spaces are ignored. This option was developed to respond to limitations of Soundex.
Kullback-Liebler Distance
Determines the similarity between two strings based on the differences between the distribution of words in the two strings.
Metaphone
Determines the similarity between two English-language strings based on a phonetic representation of their characters. This option was developed to respond to limitations of Soundex.
Metaphone (Spanish)
Determines the similarity between two strings based on a phonetic representation of their characters. This option was developed to respond to limitations of Soundex.
Metaphone 3
Improves upon the Metaphone and Double Metaphone algorithms with more exact consonant and internal vowel settings that allow you to produce words or names more or less closely matched to search terms on a phonetic basis. Metaphone 3 increases the accuracy of phonetic encoding to 98%. This option was developed to respond to limitations of Soundex.
Name Variant
Determines whether two names are variants of each other. The algorithm returns a match score of 100 if two names are variations of each other, and a match score of 0 if two names are not variations of each other. For example, JOHN is a variation of JAKE and returns a match score of 100. JOHN is not a variant of HENRY and returns a match score of 0. Click Edit in the Options column to select Name Variant options. For more information, see Name Variant Finder.
NGram Distance

Calculates in text or speech the probability of the next term based on the previous n terms, which can include phonemes, syllables, letters, words, or base pairs and can consist of any combination of letters. This algorithm includes an option to enter the size of the NGram; the default is 2.

NGram Similarity

Determines similarity between two strings based on the length of the longest common subsequence of phonemes, syllables, letters, words or base pairs.

The algorithm includes the following options:

  • Ngram size: Enter the size of the NGram. The default value is 2.
  • Drop Noise Characters: Select the check-box to replace punctuation with space.
  • Drop Spaces: Select the check-box to merge words.
Numeric String
Compares address lines by separating the numerical attributes of an address line from the characters. For example, in the string address 1234 Main Street Apt 567, the numerical attributes of the string (1234567) are parsed and handled differently from the remaining string value (Main Street Apt). The algorithm first matches numeric data in the string with the numeric algorithm. If the numeric data match is 100, the alphabetic data is matched using Edit distance and Character Frequency. The final match score is calculated as follows:

(numericScore + (EditDistanceScore + CharacterFrequencyScore) / 2) / 2

For example, the match score of these two addresses is 95.5, calculated as follows:

123 Main St Apt 567
123 Maon St Apt 567

Numeric Score = 100
Edit Distance = 91
Character Frequency = 91

91 + 91 = 182
182/2 = 91
100 + 91 = 191
191/2 = 95.5

Nysiis
Phonetic code algorithm that matches an approximate pronunciation to an exact spelling and indexes words that are pronounced similarly. Part of the New York State Identification and Intelligence System. Say, for example, that you are looking for someone's information in a database of people. You believe that the person's name sounds like "John Smith", but it is in fact spelled "Jon Smath". If you conducted a search looking for an exact match for "John Smith" no results would be returned. However, if you index the database using the NYSIIS algorithm and search using the NYSIIS algorithm again, the correct match will be returned because both "John Smith" and "Jon Smath" are indexed as "JANSNATH" by the algorithm. This option was developed to respond to limitations of Soundex; it handles some multi-character n-grams and maintains relative vowel positioning, whereas Soundex does not.
Note: This algorithm does not process non-alpha characters; records containing them will fail during processing.
Phonix
Preprocesses name strings by applying more than 100 transformation rules to single characters or to sequences of several characters. 19 of those rules are applied only if the character(s) are at the beginning of the string, while 12 of the rules are applied only if they are at the middle of the string, and 28 of the rules are applied only if they are at the end of the string. The transformed name string is encoded into a code that is comprised by a starting letter followed by three digits (removing zeros and duplicate numbers). This option was developed to respond to limitations of Soundex; it is more complex and therefore slower than Soundex.
Sonnex
This algorithm determines the similarity between two French-language strings based on the phonetic representation of their characters.
It returns a Sonnex coded key of the selected fields.
Soundex
Determines the similarity between two strings based on a phonetic representation of their characters.
SubString
Determines whether one string occurs within another.
Syllable Alignment
Combines phonetic information with edit distance-based calculations. Converts the strings to be compared into their corresponding sequences of syllables and calculates the number of edits required to convert one sequence of syllables to the other.

The following table describes the logical relationship between the number of algorithms you can use based on the parent scoring method selected.

Table 1. Matching Algorithm-to-Scoring Method Matrix
Scoring Method Algorithms
Single Multiple
Weighted Average n/a Yes
Average n/a Yes
Maximum Yes Yes
Minimum n/a Yes
Vector Summation n/a Yes