What is data matching?
find duplicates, potential duplicates and non-duplicates in a data source;
analyze data and return weighted probabilities of matching;
merge identical or similar entries into a single entry; and
reduce disparity across different data sources.
Record linkage consists of identifying records that refer to the same entity in a data set.
deterministic record linkage, which is based on identifiers that match; and
probabilistic record linkage, which is based on the probability that identifiers match.
What to do before matching?
Data profiling helps assess the quality level of the data according to defined set goals.
Data quality issues can stem from many different sources including, legacy systems, data migrations, database modifications, human communication inconsistencies and countless other potential anomalies. Regardless of the source, data quality issues can impact the ability of business to use its data to make insightful decisions.
If data is of a poor quality, or managed in structures that cannot be integrated to meet the needs of the enterprise, business processes and decision-making suffer.
Compared to manual analysis techniques, data profiling technology improves the enterprise ability to meet the challenge of managing data quality and to address the data quality challenges faced during data migrations and data integrations.
You can standardize data against indices. Synonyms are standardized or converted to the "master" words. For further information on available data synonym dictionaries, see documentation on Talend Help Center.
You can use address validation components to standardize address data against Experian QAS, Loqate and MelissaData validation tools. The addresses returned by these tools are consistent and variations in address representations are eliminated. As addresses are standardized, matching gets easier.
You can use the tStandardizePhoneNumber component to standardize a phone number, based on the formatting convention of the country of origin.
For further information on phone number standardization, see tStandardizePhoneNumber.
You can use other more generic components to transform your data and get more standardized records, such as tReplace, tReplaceList, tVerifyEmail, tExtractRegexFields or tMap.
How do you match?
The classical matching approach
Blocking by partitions
Record linkage is a demanding task because each record must be compared to the other ones from the data set. To improve the efficiency of this task, the blocking technique is a required step most of the time.
Blocking consists of sorting data into similar sized partitions which have the same attribute. The objective is to restrict comparisons to the records grouped within the same partition.
To create efficient partitions, you need to find attributes which are unlikely to change, such as a person's first name or last name. By doing this, you improve the reliability of the blocking step and the computation speed of the task.
It is recommended to use the tGenKey component to generate blocking keys and to view the distribution of the blocks. For further information on generating blocking keys, see tGenKey.
Choosing metrics and defining matching rules
After blocking data into similar sized group, you can create match rules and test them before using them in the tMatchGroup component. For further information about match analyses, see "Create a match analysis" on Talend Help Center.
Phonetic algorithms such as Soundex or Metaphone. These are used to match names.
The Levensthein distance, which calculate the minimum number of edits required to transform one string to another.
The Jaro distance, which matches processed entries according to spelling deviations.
The Jaro-Winkler distance, which is a variant of Jaro giving more importance to the beginning of the string.
For further information about the tMatchGroup component, see tMatchGroup.
In general, different partitioning schemes are necessary. This requires using sequentially tMatchGroup components to match data against different blocking keys.
For an example of how to match data through multiple passes, see Matching customer data through multiple passes.
tRecordMatching joins compared columns from the main flow with reference columns from the lookup flow. According to the matching strategy you define, tRecordMatching outputs the match data, the possible match data and the rejected data. When arranging your matching strategy, the user-defined matching scores are critical to determine the match level of the data of interest. For further information about the tRecordMatching component, see tRecordMatching.
Two cheat sheets exist in Talend Studio to guide you through the creation of your first matching or deduplication Job. For further information on cheat sheets, see Talend Studio User Guide.
The machine learning approach
The machine learning approach is useful when you want to match very high volume of data.
The data matching process can be automated by making a model learn and predict matches.
The advantages of the machine learning approach over the classical approach are the following:
The different blocking mechanism permits faster and more scalable computation. In the machine learning approach, blocking is not partitioning: a record can belong to different blocks and the size of the block is clearly delimited, which may not be the case with the tGenKey component.
The rules learnt and stored by the machine learning model can be much more complex and less arbitrary than human-designed matching rules.
- Configuring components is more simple. The machine learning model learns automatically matching distances and similarity threshold, among other things.
The first step consists of pre-analyzing a data set using the tMatchPairing component. Unique records, exact match records, suspect match pairs and a sample of the suspect match pairs are outputted by tMatchPairing.
For examples of how to compute suspect pairs, see Computing suspect pairs and writing a sample in Talend Data Stewardship and Computing suspect pairs and suspect sample from source data.
The second step consists of labeling the suspect match pairs from the sample as "match" or "no-match" manually. You can leverage Talend Data Stewardship to make the labeling task easier.
For further information about how to add a Grouping campaign to identify duplicates in a data sample in Talend Data Stewardship, see Adding a Grouping campaign to identify duplicate pairs.
In Talend Data Stewardship, grouping tasks allow authorized data stewards to validate a relationship between pairs or groups of records. The outcome of a grouping task is the list of records associated to each other.
You can use more than two classes, for example “match”, “potential match” and “different”.
For further information on grouping tasks in Talend Data Stewardship, see Handling grouping tasks to decide on relationship among pairs of records.
The third step consists of submitting the suspect match pairs you labeled to the tMatchModel component for learning and outputting a classifier model.
For examples of how to generate a matching model, see Generating a matching model from a Grouping campaign and Generating a matching model.What is a good sample?
The sample should be well-balanced: the number of records in each class - "match" and "no match" - should be approximately the same. An imbalanced data sample yields an unsatisfactory model.
The sample should be diverse: the more diverse the examples in the sample are, the more effective the rules learnt by the model will be.
The sample should be the right size: if you have a large data set with millions of records, then a few hundreds or thousands of examples may be enough. If your data set contains less than 10 000 records, then the sample size should be between 1 and 10% of the full data set.
How does tMatchModel generate a model?
The machine learning algorithm computes different measures, which are called features, to get as much information as possible on the defined columns.
To generate the model, tMatchModel analyzes the data using the Random Forest algorithm. A random forest is a collection of decision trees used to solve a classification problem. In a decision tree, each node corresponds to a question about the features associated to the input data. A random forest grows many decision trees to improve the accuracy of the classification and to generate a model.
The fourth step consists of labeling suspect pairs for large data sets automatically using the model computed by tMatchModel with the tMatchPredict component.
For an example of predict labels on suspect pairs, see Labeling suspect pairs with assigned labels.
For further information on the machine learning approach, see Matching on Spark.
Surviving master records
Merging records using tRuleSurvivorship
Once you estimated duplicates and possible duplicates that are grouped together, you can use the tRuleSurvivorship component to create a single representation for each group of duplicates using the best-of-breed data. This representation is called a survivor.
For an example of how to clean and deduplicate a data set, see Creating a clean data set from the suspect pairs labeled by tMatchPredict and the unique rows computed by tMatchPairing and .
Using Talend Data Stewardship for clerical review and merging records
You can add merging campaigns in Talend Data Stewardship to review and modify survivorship rules, create master records and merge data.
For further information on merging campaigns in Talend Data Stewardship, see Adding a Merging campaign to deduplicate records.
- In Talend Data Stewardship, data
stewards are business users in charge of resolving data stewardship tasks:
Classifying data by assigning a label chosen among a predefined list of arbitration choices.
Merging several potential duplicate records into one single record.
Merging tasks allow authorized data stewards to merge several potential duplicate source records into one single record (golden record). The outcome of a merging task is the golden record produced by data stewards.
For further information on merging tasks in Talend Data Stewardship, see Handling merging tasks to deduplicate records.
Source records can come from the same source (database deduplication) or different sources (databases reconciliation).
How do you rematch using machine learning components?
Doing continuous matching
If you want to match new records against a clean data set, you do not need to restart the matching process from scratch.You can reuse and index the clean set and to do continuous matching.
To be able to perform continuous matching tasks, Elasticsearch version 5.1.2+ must be running.
The continuous matching process is made up of the following steps:
The first step consists of computing suffixes to separate clean and deduplicated records from a data set and indexing them in Elasticsearch using tMatchIndex.
For an example of how to index a data in Elasticsearch using tMatchIndex, see Indexing a reference data set in Elasticsearch.
The second step consists of comparing the indexed records with new records having the same schema and outputting matching and non-matching records using tMatchIndexPredict. This component uses the pairing and matching models generated by tMatchPairing and tMatchModel
For an example of how to matching new records against records from a reference data set, see Doing continuous matching using tMatchIndexPredict.
You can then clean and deduplicate the non-matching records using tRuleSurvivorship and populate the clean data set indexed in Elasticsearch using tMatchIndex.