​
Purpose

Extracting data from tables can often be time-consuming and error-prone. Affinda’s advanced table extraction technology makes this process fast and reliable, even for complex layouts. To validate tables extracted in Affinda, you need to understand how to use the table editor. This tutorial provides a step-by-step guide. This tutorial is aimed at any users who are reviewing the outputs of tables from the Affinda platform. This may be for:
  1. Model improvement: Confirming documents where the table data is correct will improve model performance over time by adding the documents to ‘model memory’
  2. Ongoing document processing: ‘Human in the loop’ review of documents processed as part of general processes

​
Step-by-Step

Visual Learner? Follow along with our Video Tutorial:
1

Create new table (if required)

In most cases, the Affinda platform will have predicted a table that exists in the document. However, in some cases, you may need to create a new table annotation (or, in the case where you delete the initial model prediction).To do this, simply use your mouse to draw a box around the entire table area (excluding the table header) as you would with any other field. Select the appropriate table field from the list. Once the table annotation has been created, Affinda will then predict the rows within the table and the fields within each row.
2

Edit table boundary

When a table annotation is created by the model, you may need to adjust table boundaries to ensure all relevant fields are captured. To do so, click on the table annotation and edit the boundary by dragging with your mouse.When you edit the table boundary, Affinda will re-predict the table. Any confirmed rows or fields within the table will not be re-predicted.
3

Review first row

To edit a row, first click the entire table, then click the first row.

​
Adjust rows

Properly setting table rows is essential for ensuring accurate data extraction. Each row should capture only the data relevant to that line. Accurate row configuration also improves the performance of future model predictions.There are four main ways to adjust rows in the Affinda platform:Split Rows
If a single detected row spans multiple actual rows in the document, use the Split icon (located to the right of the row) to divide it into two separate rows automatically.Delete Rows
To remove an unwanted row from the top or bottom of the table, click the Trash icon next to that row.Merge Rows
If a row in the middle of the table is incorrectly separated and additional rows are predicted, clicking the Trash icon will merge that row with the one directly above.Adjust Row Boundaries
Use the draggable handles in the middle of a row to shift its boundary up or down, refining the area that each row covers.

​
Adjust fields

Once the row is correctly set, update any fields within the row as required. This is the same as editing other fields in the user validation interface.
4

Auto-Fill Down

Once the data in the first row is correct, click ‘Auto-Fill Down’. This feature uses the corrected row, along with any rows above it, to guide the model’s predictions for the rows below.Instead of manually editing each row, ‘Auto-Fill Down’ allows the model to learn the structure and patterns within the table, enabling it to quickly and accurately update the remaining rows. This significantly reduces manual effort while improving overall data quality.
5

Repeat steps 3 - 4 (if required)

While editing the first row and using Auto-Fill Down often results in fully accurate table data, occasional manual corrections may still be necessary. If so, simply repeat Steps 3 and 4, reviewing and applying Auto-Fill, on any rows that need adjustment.

​
Tip: Work from top to bottom for best results

Affinda’s Auto-Fill assumes all data in the current row — and all rows above it — are correct. To ensure accurate predictions, always review and edit rows sequentially from the top of the table downward.