How to Use Learn Models for Record Matching

This method makes it easy to use a Learn model to evaluate feature vectors consisting of query scores. This approach is suitable for advanced record matching applications, including record equivalence and deduplication applications.

Each feature score is computed by a TIBCO Patterns query. Each query that produces a feature score can be any of the query types described in Designing Queries for TIBCO Patterns, including complex queries. An RLINK score combiner is used to combine the scores of all these queries and obtain the evaluation score using the Learn model. This allows a Learn model to be applied directly within the query to generate the match score for a record.

The RLINK score combiner is invoked like any other score combiner, for example, AND or OR. To configure an RLINK score combiner, provide the name of the model, along with the query inputs for each feature. The query configuration must exactly match the configuration used to train the model. When a model is exported from the Learn UI, it also exports a Java class with a static method that returns a NetricsQuery object. The returned NetricsQuery object incorporates the RLINK score combiner and the sub-queries for all features. Using the exported Java class ensures the correct configuration of the query.