Optimize matching rules
Define deduplication on a match entity
In addition to cross-entity match rules, you can also specify deduplications rules. Deduplication is another process when matching records. It identifies duplicate records and merges them into one record. Source records get linked to the merged record with alternate IDs.
Deduplicated records will be used in the cross-entity matching process. Deduplication happens on individual entities and can be configured every entity used in match pairs.
Specifying deduplication rules isn't mandatory. If no such rules are configured, the system-defined rules are applied. They combine all records into a single record before passing the entity data to cross-entity matching for enhanced performance.
When you are specifying how to combine the duplicate records, you can choose one of three options:
- Most filled: Identifies the record with most populated attribute fields as the winner record. It's the default merge option.
- Most recent: Identifies the winner record based on the most recency. Requires a date or a numeric field to define the recency.
- Least recent: Identifies the winner record based on the least recency. Requires a date or a numeric field to define the recency.
Adjusting and modifying rules
When you first run a matching rule, you will likely find that the results are not quite what you were expecting or you might not know what it did. You have several options for evaluating the quality of your match pairs and improving the results.
Your first task is to examine the insights that are provided on the Match page. At the top of the page, you can review the unique and matched records from the tiles. The two separate tiles that you can examine are:
The left tile, which shows the number of unique profiles that the system identified.
The right tile, which shows the total number of matches across all your match pairs.
The right tile will provide you with more context into the first number. The closer that number is to the first number, the better the application did at matching items. If that number is low in comparison, you might want to consider making some changes to your matching rule.
You can explore further into the results by examining a number of records that came from each match pair by expanding the item to view matching percentages for each rule that is defined. Typically, it's a good idea to review at least a part of the rule to validate that records were matched according to your expectations.
Try experimenting with different thresholds around your conditions to identify the optimal thresholds. Use the Edit option on the match pair rule that you want to experiment with. Identify the condition that you want to experiment with. Each criterion is represented by one row in the Match rule pane.
By spending some time experimenting with these thresholds, you will gain a better understanding of the effects for each of the threshold levels. As you adjust the levels, you can compare how many records will be matched under each of the threshold levels and view the records under each option.
Make changes to optimize your matches
After you have experimented with different options to better understand the match qualities, you can use your knowledge to potentially reconfigure some of your match parameters. Different options that you could consider are:
Change the order of your rules - If you defined multiple rules, it might be worth changing their order so you can yield a better match quality. You can reorder the match rules by selecting the Move Up and Move Down options that are provided in the match rules grid.
Duplicate your rules - If you defined a match rule and want to create a similar rule with modifications, you can make a copy without having to manually recreate all the conditions that are defined by selecting the Duplicate option.
Edit your rules - This option includes several important changes that you should try as you optimize the match quality. The following options are accessible through the rule's Edit button:
Changing attributes for a condition - This method can be done by reselecting new attributes within the specific condition row.
Changing threshold for a condition - This method can be quickly achieved through the threshold bar.
Changing normalization method for a condition - This method can be done by reselecting the normalization method.