Approximate String Matcher

This node calculates the string similarity score between values from one column in a reference table and one or more columns from the comparison table. Additionally, it allows to filter the rows based on if there is a similar match or not.
The node computes the similarity between strings using configurable algorithms such as Levenshtein Edit Distance, Longest Common Subsequence, or Positional Matching, and applies a user-defined threshold to determine which rows are considered a match.
Each field value in the comparison input is matched against each reference input value. The best match of any of these single comparisons is decisive for the filter decision. So we consider the columns in the comparison table as well as the rows from the reference table combined in a logical OR fashion.
The comparison algorithms are described in detail in the options section.
As output of this node the user can choose to select the rows that match according to the filter value, the rows that do not match the criteria or all rows.
In either case additional columns can be generated that contain the computed criteria, namely the numeric match value, the best reference match and a string showing the modifications of the comparison string to align with the reference string. These extra fields can support downstream processing and decision-making beyond simple filtering.
This node may only be used for private and non-commercial purposes. Commercial use requires a valid license from exorbyte GmbH. All rights reserved.
For more information contact consulting@exorbyte.com.

Options

Select Settings Group

Allowing the user to navigate through different sections of the configuration options

• Input
• Search
• Output

Select Column in Reference Input

Select a column of the Reference Input Table to be used as list of Reference Terms

Select Columns in Comparison Input

Select columns applicable to comparison to the Reference Terms

Add Column with Numeric Matching Value

Appends a column showing the calculated similarity between strings. This may be the algorithm-specific matching value or percentage similarity, depending on the selected numeric metric.
The column is named “<input column name> - Match Number” and is useful for scoring and applying thresholds.

Add Column with Character Match Sequence

Appends a column showing a symbolic alignment of characters in the comparison and reference strings.
'=' -> Match
'o' -> Substitution
'x' -> Deletion
'+' -> Insertion
'><' -> Transposition
The column name is “<input column name> - Match Sequence”.

Add Column with Best Reference Match

Appends the most similar reference string for each row in the comparison table. If there are multiple reference strings with the same Numeric Matching Value, the last Reference Column entry with best value is taken.
Ideal for deduplication, linking, and labeling workflows where a single best match is sufficient.
The column name is “<input column name> - Best Match”.

Matching Algorithm Selector

Select the algorithm used to calculate string similarity between reference and comparison inputs.
Options:

• Levenshtein - Calculation of the Edit Distance, which is the minimum number of edit operations needed to transform the comparison term into the reference term. Allowed edit operations are:
  • Insertion of a character
  • Deletion of a character
  • Substitution of a character
  • Transposition of two adjacent characters (also referred to as Damerau-Levenshtein extension)
  The Levenshtein Algorithm compares, by default, the whole strings from beginning to end. In special situations, one wants to ignore prefixes or suffixes and find the match at any position within the other term.
  For this, there are options that trigger certain parts of the comparison term or the reference term to be ignored at no “cost”, meaning not counting as errors. The length of the ignored portion is optimized as to produce the best possible match.
  You can choose 2 of the 4 options simultaneously. 3 or more are not meaningful, since one term could be completely ignored, not producing a relevant matching number.
  Options are the following (described in more detail in their own section):
  • Ignore Leading Characters in Reference Term
  • Ignore Leading Characters in Comparison Term
  • Ignore Trailing Characters in Reference Term
  • Ignore Trailing Characters in Comparison Term
• Positional Matching - Simply compares characters at fixed character positions (first with first, second with second…). This is needed for comparing e.g. IDs with a fixed format.
• Longest Common Subsequence - Finds the longest subsequence of characters that appear in each of both terms. The sequence itself is not necessarily unique. Only the length of the longest sequence is important.

Ignore Leading Characters in Reference Term

Skips leading characters in the reference string during matching until the comparison term starts matching.

Ignore Leading Characters in Comparison Term

Skips leading characters in the comparison string during matching until the reference term starts matching.

Ignore Trailing Characters in Reference Term

Skips trailing characters in the reference string during matching.

Ignore Trailing Characters in Comparison Term

Skips trailing characters in the comparison string during matching.

Case Sensitivity

Determines whether the matching algorithm is case-sensitive or case-insensitive.

Numeric Matching Value (Levenshtein)

The Levenshtein algorithm computes a specific matching value from which a similarity can be deduced.
You can choose whether to use the algorithm specific matching number or the normalized similarity score ranging from 0-100 for the filtering.

• The Levenshtein specific matching value is the Number of Errors, also referred to as edit distance or Levenshtein distance, which is the minimum number of edit operations necessary to transform the comparison term into the reference term. The number of errors is a “negative” number: similarity is highest for zero errors.
• The Similarity in Percent is a match score as a percentage. This option is algorithm independent. It is a normalized value always ranging from 0-100 for all algorithms, 100 being a full match (according to the algorithm).

Numeric Matching Value (Longest Common Subsequence)

The Longest Common Subsequence (LCS) algorithm computes a specific matching value from which a similarity can be deduced.
You can choose whether to use the algorithm specific matching number or the normalized similarity score ranging from 0-100 for the filtering.

• The LCS specific matching value is the Length of LCS, which is a sequence of characters that appears in both terms and that may be interrupted by an arbitrary number of other characters at any position. The length of LCS is a “positive” number, similarity is highest when the length is equal to the maximum of lengths of both terms.
• The Similarity in Percent is a match score as a percentage. This option is algorithm independent. It is a normalized value always ranging from 0-100 for all algorithms, 100 being a full match (according to the algorithm).

Numeric Matching Value (Positional Matching)

The Positional Matching algorithm computes a specific matching value from which a similarity can be deduced.
You can choose whether to use the algorithm specific matching number or the normalized similarity score ranging from 0-100 for the filtering.

• The Positional Matching specific matching value is the Number of matching character positions. This is a “positive” number as for LCS. Highest Similarity comes with a matching number of the maximum of the lengths of both terms.
• The Similarity in Percent is a match score as a percentage. This option is algorithm independent. It is a normalized value always ranging from 0-100 for all algorithms, 100 being a full match (according to the algorithm).

Row Filter Condition

Controls which rows are included in the output.
Options:

• Output matching rows - Only rows that meet similarity criteria are included in the output. Matching is defined as having a similarity equal or higher than the similarity defined by the Matching Value Threshold. Be aware that if using the raw numeric matching value a match might mean a lower(!) value than the threshold for a “negative” matching number like the number of errors.
  A row where at least one column matches is considered a match for the whole row resembling an “OR” behavior.
• Output non-matching rows - Only rows that do not meet the matching criteria are forwarded into the output. Any column that matches filters away such a row.
• No Filtering - All rows are forwarded into the output. This is used to just add the matching information from the output section and use that information later.

Matching Value Threshold - Maximal Number of Edit Operations

This setting allows you to set the filter criteria based on the selection of the Numeric Matching Value.
This setting only appears, if filtering is actually switched on by the previous setting.
If the algorithm specific matching value was chosen, it applies to this number. If similarity was chosen, the value here is also a similarity threshold.

Matching Value Threshold - Minimal Number of Matches

This setting allows you to set the filter criteria based on the selection of the Numeric Matching Value.
This setting only appears, if filtering is actually switched on by the previous setting.
If the algorithm specific matching value was chosen, it applies to this number. If similarity was chosen, the value here is also a similarity threshold.

Matching Value Threshold - Minimal Common Sequence Length

This setting allows you to set the filter criteria based on the selection of the Numeric Matching Value.
This setting only appears, if filtering is actually switched on by the previous setting.
If the algorithm specific matching value was chosen, it applies to this number. If similarity was chosen, the value here is also a similarity threshold.

Matching Value Threshold - Minimal Matching Percentage

This setting allows you to set the filter criteria based on the selection of the Numeric Matching Value.
This setting only appears, if filtering is actually switched on by the previous setting.
If the algorithm specific matching value was chosen, it applies to this number. If similarity was chosen, the value here is also a similarity threshold.

Views

This node has no views

Workflows

• Approximate String Matcher Overview ExamplesKNIME Hub
• Business Email Fuzzy Classification (Dictionary-Based)KNIME Hub
• Customer Name Search ToolKNIME Hub
• Customer Review AnalysisKNIME Hub
• Explainable Fuzzy Matching - Typo Error StatisticsKNIME Hub
• Fault Tolerant Customer Search Data AppKNIME Hub
• Frequency-Aware Anomaly DetectionKNIME Hub
• Fuzzy Category Cleaner - Preparing Categorical Data for Machine LearningKNIME Hub
• Fuzzy Keyword Detection in Customer ReviewsKNIME Hub
• Fuzzy Keyword DetectorKNIME Hub
• Fuzzy Row Filter (Single Column)KNIME Hub
• KYC Sanction Screening with Fuzzy MatchingKNIME Hub
• Logistics Container ID Fuzzy Matching with Approximate String Matcher NodeKNIME Hub
• Simple Fuzzy AggregationKNIME Hub
• Simple Fuzzy Match Example with Approximate String MatcherKNIME Hub
• Simple Typo AnalysisKNIME Hub
• Simple Typo-Tolerant Lookup Tool for LLMsKNIME Hub
• Vendor Name Similarity Scoring Workflow [forum-post-88014]KNIME Hub

Links

• No links available

Developers