Set up your first matching model

This article walks you through creating a matching model, from importing your records to measuring its accuracy. If you are not yet sure whether matching is the right model type, read What is a matching model? first.

Prerequisites

Before you begin, make sure you have:

• Access to the Models section of the Tekst platform.
• Your master data as a CSV or JSON file, with one column that uniquely identifies each record.
• A few example messages in mind, each with the record you would expect it to match.

Step 1: Create the model

1. Go to the Models section and select Create model.
2. Choose Matching model as the type.
3. Select Let's get started.
4. Enter a Model name and an optional Model description.
5. Optionally upload your master data file now by dragging a CSV or JSON file into the upload area. You can also do this later.
6. Select Create Model.

Tekst creates the model and opens its detail page.

Step 2: Import your records

If you did not upload a file when creating the model, do it now from the records area of the model detail page. For a full explanation of importing, updating, and re-importing records, see Managing master data records.

Step 3: Set the unique identifier column

Open the model's settings and choose the column that uniquely identifies each record under Unique identifier. The list is only available once records have been uploaded, because the options come from the columns in your file.

This is required before you can add test items, because the unique identifier is how Tekst knows which exact record a match refers to.

Step 4: Add domain knowledge

Capture any synonyms, rules, and exceptions that should guide the model toward the right record. Domain knowledge is a written context document that you build by describing what you want in the assistant chat next to it, rather than by filling in fields.

Good domain knowledge is often the difference between a model that handles your real-world data well and one that struggles with abbreviations and edge cases. For guidance, see Using domain knowledge to guide matching.

Step 5: Build a test set

A test set is a collection of messages for which you have confirmed the correct record. It is how Tekst measures accuracy.

1. Add test items, each with an example input.
2. For each one, pick the correct record (the "ground truth") using the record picker.
3. Include a spread of examples, including tricky ones with unusual wording.

Step 6: Review accuracy

Once you have imported records and added test items, Tekst evaluates the model automatically and reports its accuracy. Open the accuracy view to see how it is performing and where it is going wrong.

For a full explanation of how the score is calculated, see How matching accuracy is measured.

What happens next

As your team confirms or corrects matches, those corrections improve the model over time. You can also refine results by updating your domain knowledge or re-importing your records when your master data changes.