Contributors: Jothiponsundar, Subhojeet
Today Mapping Assist serves hundreds of customers helping them quickly map fields and accelerate the speed of integration developments. Using Artificial Intelligence algorithm Mapping Assist finds the best match for a particular field.
With the new Incremental Learning feature Mapping Assist algorithm will adjust to user selections over time. The App Connect software comes bundled with a base Mapping Assist algorithm that provides state-of-the-art suggestions. Over time, the Incremental Learning feature uses Deep Learning algorithms to retrain and adjust to customer's mapping selections over time.
Let's demonstrate this by trying out a simple use-case. We will start by creating a simple API flow to Sync customer data from Microsoft Dynamics to Sales agreement in Microsoft Dynamics. For the field "Commission agreement" let us say a user selects a field which is not a top suggestion from the Mapping Assist feature. In this example let us consider we select "Payment terms" from MS Dynamics Retrieve Customers which is the second top suggestion.
Fig. 1: Before Training
As the "Payment terms" field is suggested as the second top suggestion, the score is comparatively low (47%). Still the user preferred to map this field and proceeded to map. This User mapping will be stored once the user executes the flow successfully.
As part of this incremental learning feature, a new training will be triggered on regular intervals with the data generated by user preference. In this example, Fig 1 user preference is taken into account and trained on top of base version of AI mapping assist model.
After training, the confidence score has improved for the same mapping suggestions from 47% to 69% and the position of the user preference also has been shifted to top ( Fig.3). The model has incrementally learned and fine-tuned to provide suggestion which the user has preferred earlier.
Fig. 3: After Training
As the incremental learning is happening on top of base version of Mapping Assist, the new model will be saved only after validating its accuracy & regression with base model. So the users don't have to worry about bias and regression issues of the new trained model.
To enable mapping assist & incremental learning features the user has to turn both the `Enable Mapping Assist` and `Mapping Assist incremental learning` toggles ON as shown in Fig.4. If the user wishes to use the base model version, then he can turn OFF the incremental learning feature though the toggle options any time.