tMatchGroup properties for Apache Spark Batch

These properties are used to configure tMatchGroup running in the Spark Batch Job framework.

The Spark Batch tMatchGroup component belongs to the Data Quality family.

This component is available from the 8.0 R2024-08 Talend Studio monthly release.

Important: When using the Simple VSR algorithm, the results depend on the input data and the Spark API shuffling. This causes different results at each Job run. To optimize the result stability, it is highly recommended to use the tPartition component right before the tMatchGroup component.

To optimize the result stability, you can also:

Set the same Blocking key (tMatchGroup) and Partition key (tPartition).
Select unique values as the Sort key (tPartition). For example, the ID.

The component in this framework is available in all Talend Platform products with Big Data and in Talend Data Fabric.

Basic settings

Schema and Edit schema	A schema is a row description. It defines the number of fields (columns) to be processed and passed on to the next component. When you create a Spark Job, avoid the reserved word `line` when naming the fields. Click Sync columns to retrieve the schema from the previous component connected in the Job. The output schema of this component contains the following read-only fields: The output schema of this component contains the following read-only fields: `GID`: Provides a group identifier of the data type String. `GRP_SIZE`: Counts the number of records in the group, computed only on the master record. `MASTER`: Identifies, by true or false, if the record used in the matching comparisons is a master record. There is only one master record per group. Each input record will be compared to the master record, if they match, the input record will be in the group. `SCORE`: Measures the distance between the input record and the master record according to the matching algorithm used. In case the tMatchGroup component is used to have multiple output flows, the score in this column decides to what output group the record should go. `GRP_QUALITY` provides the quality of similarities in the group by taking the minimal matching value. Only the master record has a quality score.
	Built-In: You create and store the schema locally for this component only.
	Repository: You have already created the schema and stored it in the Repository. You can reuse it in various projects and Job designs.
Matching Algorithm	Simple VSR is set by default. Information noteRestriction: The T-Swoosh algorithm is not available for the Apache Spark Batch component. You can use it for the Standard component.
	Click the import icon to import a match rule from the Talend Studio repository. For further information about how to import rules, see Importing match rules from the repository
Key Definition - Input Key Attribute column	Select the columns from the input flow on which you want to apply a matching algorithm. Information noteNote: When you select a date column on which to apply an algorithm or a matching algorithm, you can decide what to compare in the date format. For example, if you want to only compare the year in the date, in the component schema set the type of the date column to Date and then enter "yyyy" in the Date Pattern field. The component then converts the date format to a string according to the pattern defined in the schema before starting a string comparison.
Key Definition - Matching Function column	Select a matching algorithm from the list. For more information on each algorithm, see Matching algorithms.
Key Definition - Custom Matcher Class column	When you select Custom as the matching type, enter the path pointing to the custom class (external matching algorithm) you need to use. This path is defined by yourself in the library file (.jar file) which you can import by using the tLibraryLoad component. For example, to use a MyDistance.class class stored in the directory org/talend/mydistance in a user-defined mydistance.jar library, the path to be entered is org.talend.mydistance.MyDistance.
Key Definition - Tokenized measure column	Tokenization is the concept of splitting a string into words. Select the method to use to compute a tokenized measure for the selected algorithm: No: No tokenization method is used on the string. With this option, "John Doe" and "Jon Doe" should match. Same place: Splits the two strings by words to two lists, list1 and list2. Associates each element from list1 with the element which has the same position in list2. Using this method, "She is red and he is pink" and "Catherine is red and he is pink" should match. Same order: Splits the two strings by words to two lists, list1 and list2 and assumes that list1 is shorter than list2. Tries to associate the elements from list1 with the elements in list2 taken in the same order. Using this method, "John Doe" and "John B. Doe" match. This method should be used only with strings which has a few words, otherwise the number of possible combinations can be large. Any order: Splits the two strings by words to two lists, list1 and list2 and assumes that list1 is shorter than list2. Tries to assign each word of list1 to a word of list2, in order to have the highest global similarity (with respect to the used similarity). Using this method, "John Doe" and "Doe John" match.
Key Definition - Confidence Weight column	Set a numerical weight for each attribute (column) of the key definition. You can enter a number or a context variable. The value must be an integer greater than 0.
Key Definition - Handle Null column	To handle null values, select from the list the null operator you want to use on the column: Null Match Null: A Null attribute only matches another Null attribute. Null Match NONE: A Null attribute never matches another attribute. Null Match ALL: A Null attribute matches any other value of an attribute. For example, two columns, name and firstname where the name is never null, but the first name can be null. If two records: "Doe", "John" "Doe", "" Depending on the operator you choose, these two records may or may not match: Null Match Null: They do not match. Null Match NONE: They do not match. Null Match ALL: They match. For the records: "Doe", "" "Doe", "" Null Match Null: They match. Null Match NONE: They do not match. Null Match ALL: They match.
Match Threshold	Enter the match probability. Two data records match when the probability is greater than or equal to the set value. You can enter a different match threshold for each match rule.
Blocking Selection	If required, select the columns from the input flow according to which you want to partition the processed data in blocks, this is usually referred to as "blocking". Blocking reduces the number of pairs of records that needs to be examined. In blocking, input data is partitioned into exhaustive blocks designed to increase the proportion of matches observed while decreasing the number of pairs to compare. Comparisons are restricted to record pairs within each block. Using blocking columns is very useful when you are processing very big data.

Advanced settings

Multiple output	Select the Separate output check box to have several output flows: Uniques: when the group size is equal to 1, the record is listed in this flow. When records are not unique, they can be: Matches: when the group quality is greater than or equal to the threshold you define in the Confident match threshold field, the record is listed in this flow. Suspects: when the group quality is less than the threshold you define in the Confident match threshold field, the record is listed in this flow. Confident match threshold: set a numerical value between the current Match threshold and 1. From this threshold, you can be confident in the quality of the group.
Multi-pass	Select this check box to enable a tMatchGroup component to receive data sets from another tMatchGroup that precedes it in the Job. This will refine the groups received by each of the tMatchGroup components through creating data partitions based on different blocking keys. With multi-pass matching, all master records are generated but intermediate master records are removed from the output flow. Only final master and original records are kept at the end. When single master records from the 1st tMatchGroup merge into one group after passing the 2nd tMatchGroup, their order in the group may change every time you run the Job. For an example Job, see Matching customer data through multiple passes.
Sort the output data by GID	Select this check box to group the output data by the group identifier. The output is sorted in descending alphanumeric order by group identifier.
Output distance details	Select this check box to add an output column MATCHING_DISTANCES in the schema of the component. This column provides the distance between the input and master records in each group. Information noteImportant: When you use two tMatchGroup components in a Job and select this option in one component, you must select this check box in both tMatchGroup components before linking them together. If you linked the components before selecting this check box, select this check box in the second component in the Job flow and then, in the first component. Otherwise, you may have an issue as there are two columns in the output schema with the same name. Selecting this check box in only one tMatchGroup component may cause schema mismatch issues.
Display detailed labels	This checkbox is available when Output distance details is selected. Select this check box to have in the output MATCHING_DISTANCES column not only the matching distance but also the names of the columns used as key attributes in the applied rule. For example, if you try to match on first name and last name fields, lname and fname, the output would be fname:1.0\|lname:0.97 when the check box is selected and 1.0\|0.97 when it is not selected.
Deactivate matching computation when opening the wizard	Select this check box to open the Configuration wizard without running the match rules defined in the wizard. This enables you to have a better experience with the component. Otherwise, the wizard may take some time to open.

Global Variables

Global Variables	ERROR_MESSAGE: the error message generated by the component when an error occurs. This is an After variable and it returns a string. This variable functions only if the Die on error check box is cleared, if the component has this check box. A Flow variable functions during the execution of a component while an After variable functions after the execution of the component. To fill up a field or expression with a variable, press Ctrl+Space to access the variable list and choose the variable to use from it. For more information about variables, see Using contexts and variables.

ERROR_MESSAGE: the error message generated by the component when an error occurs. This is an After variable and it returns a string. This variable functions only if the Die on error check box is cleared, if the component has this check box.

A Flow variable functions during the execution of a component while an After variable functions after the execution of the component.

To fill up a field or expression with a variable, press Ctrl+Space to access the variable list and choose the variable to use from it.

For more information about variables, see Using contexts and variables.

Usage

Usage rule	This component is an intermediary step. It requires an input flow as well as an output flow.

Matching algorithms

This list describes each matching algorithm from the Matching Function column:

Exact: Matches each processed entry to all possible reference entries with exactly the same value. It returns 1 when the two strings exactly match, otherwise it returns 0.
Exact - ignore case: Matches each processed entry to all possible reference entries with exactly the same value while ignoring the value case.
Soundex: Matches processed entries according to a standard English phonetic algorithm. It indexes strings by sound, as pronounced in English, for example "Hello": "H400".It does not support Chinese characters.
Levenshtein (edit distance): Calculates the minimum number of edits (insertion, deletion, or substitution) required to transform one string into another. Using this algorithm in the tMatchGroup component, you do not need to specify a maximum distance. The component automatically calculates a matching percentage based on the distance. This matching score will be used for the global matching calculation, based on the weight you assign in the Confidence Weight field.
Metaphone: Based on a phonetic algorithm for indexing entries by their pronunciation. It first loads the phonetics of all entries of the lookup reference and checks all entries of the main flow against the entries of the reference flow. It does not support Chinese characters.
Double Metaphone: A new version of the Metaphone phonetic algorithm, that produces more accurate results than the original algorithm. It can return both a primary and a secondary code for a string. This accounts for some ambiguous cases as well as for multiple variants of surnames with common ancestry. It does not support Chinese characters.
Soundex FR: Matches processed entries according to a standard French phonetic algorithm.It does not support Chinese characters.
Jaro: Matches processed entries according to spelling deviations. It counts the number of matched characters between two strings. The higher the distance is, the more similar the strings are.
Jaro-Winkler: A variant of Jaro, but it gives more importance to the beginning of the string.
Fingerprint key: Matches entries after doing the following sequential process:
1. Remove leading and trailing whitespace.
2. Change all characters to their lowercase representation. It does not support Chinese characters.
3. Remove all punctuation and control characters.
4. Split the string into whitespace-separated tokens.
5. Sort the tokens and remove duplicates.
6. Join the tokens back together. Because the string parts are sorted, the given order of tokens does not matter. So, Cruise, Tom and Tom Cruise both end up with a fingerprint cruise tom and therefore end up in the same cluster.
7. Normalize extended western characters to their ASCII representation, for example gödel to godel. This reproduces data entry mistakes performed when entering extended characters with an ASCII-only keyboard. However, this procedure can also lead to false positives, for example gödel and godél would both end up with godel as their fingerprint but they are likely to be different names. So this might work less effectively for datasets where extended characters play substantial differentiation role.
q-grams: Matches processed entries by dividing strings into letter blocks of length q in order to create a number of q length grams. The matching result is given as the number of q-gram matches over possible q-grams.
Hamming: Calculates the minimum number of substitutions required to transform one string into another string having the same length. For example, the Hamming distance between "masking" and "pairing" is 3.
custom...: Lets you load an external matching algorithm from a Java library using the custom Matcher column.

For further information about how to load an external Java library, see tLibraryLoad.

For further information about how to create a custom matching algorithm, see Creating a custom matching algorithm.

For a related scenario about how to use a custom matching algorithm, see Using a custom matching algorithm to match entries.

Did this page help you?

If you find any issues with this page or its content – a typo, a missing step, or a technical error – please let us know!

Leave your feedback here