Optimal Labeling Approach for Date Extraction in Custom Model

Question

Optimal Labeling Approach for Date Extraction in Custom Model

Remy 0

I’m seeking guidance on the best way to label date-related data for a custom extraction model. Below is an example of the type of data I need to extract, which varies in format as shown in the table.

Key Questions:

Should I label Day, Month, Year, and AM/PM separately, or is it better to use a single bounding box covering all or a subset?
- If using a single box, what sub-type should I assign? (Date > dmy?)
- If using separate boxes, what sub-types should I assign for each field based on the variations in format?

I asked CoPilot the same question twice and received completely contradictory answers:

First response: Label separately as Date > Not Specified
Second response: Label as a single bounding box as Date > dmy

Field	Example Values
Day	1st, 2nd, 3rd, 3, 15, 25
Month	Dec, December, 12
Year	2023, 2024, 23, 24
AM / PM	Checkbox or other label?

Thanks for your support!

1 answer

Your answer

Answer 1

SriLakshmi C 3,415 Microsoft External Staff

Hello Remy,

Greetings and Welcome to Microsoft Q&A! Thanks for posting the question.

I understand that you are looking for Labeling date-related data for a custom extraction model,

If your dataset has consistent date formats, you can use a single bounding box to label the entire date. In this case, assign the sub-type based on the format, such as Date > dmy for day-month-year formats or Date > mdy for month-day-year formats. This approach is ideal when the dataset contains uniform and predictable date structures.

However, if your dataset is highly inconsistent or you need to extract specific date components, it is better to use separate bounding boxes for each part of the date. Label each component individually with specific sub-types:

Day: Assign Date > Day (e.g., "1st", "15", "3rd").
Month: Assign Date > Month (e.g., "Dec", "December", "12").
Year: Assign Date > Year (e.g., "2023", "23").

Choosing between these two approaches depends on the variability of your dataset and the granularity of the information you need to extract.

I Hope this helps. Do let me know if you have any further queries.

If the response helped, please do click Accept Answer and Yes for was this answer helpful.

Thank you!

SriLakshmi C 3,415 Reputation points Microsoft External Staff

2025-01-29T19:34:41.2033333+00:00

Hi Remy,

Did you get any chance to check the above response. Thank you!
Remy 0 Reputation points

2025-01-29T20:06:28.12+00:00

Hi SriLakshmi,

I do not have Date > Day/Month/Year sub-types.
I only have Date > not-specified/dmy/mdy/ymd sub-types, per the attached image.

Do I need to enable a feature somewhere?

Thank you
Remy 0 Reputation points

2025-01-29T20:13:32.4566667+00:00

Hi SriLakshmi,

I do not have Date > Day/Month/Year sub-types. My only options are Date > not-specified, dmy, mdy, ymd per the image below.
SriLakshmi C 3,415 Reputation points Microsoft External Staff

2025-01-29T22:30:29.9633333+00:00

Hi Remy,

Sorry for the inconvenience caused to you,

It's better to use a single bounding box to label the entire date is ideal for datasets with consistent and predictable formats, ensuring efficient extraction and reducing errors. This approach simplifies labeling, improves model accuracy, and eliminates the need for post-processing. It works best in structured documents like invoices and reports, where dates follow a standard format. However, if date structures vary significantly, a more granular approach may be needed.

Thank you!
SriLakshmi C 3,415 Reputation points Microsoft External Staff

2025-01-30T20:38:14.9033333+00:00

Hi Remy,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

Thank you!
Remy 0 Reputation points

2025-01-31T00:30:32.5066667+00:00

Hi SriLakshmi,

I have retrained using bounding boxes across the entire area and set the subtype to Date => DMY. However, it is extracting all of the text and date data. Should this be my expectation? When I used individual bounding boxes for the day, month, and year sections, I extracted just those values. This is preferable.
Remy 0 Reputation points

2025-01-31T00:33:11.89+00:00

I have retrained by placing a single bounding box over the entire area, but it is extracting "day of" and other words. I felt that I had better success when I labeled the day, month, and year, separately. Are my expectations incorrect, or am I doing something incorrect?
Saideep Anchuri 4,605 Reputation points Microsoft External Staff

2025-01-31T03:59:17.7466667+00:00

Hi Remy,

If you are having better results by annotating day, month, year individually, we would recommend you continue that way and amalgamate results of day, month, year. And also increase dataset size and quality of dataset.

kindly refer below link: custom-label

Thank You.
Remy 0 Reputation points

2025-01-31T12:37:11.35+00:00
Please see the attached image. Per the custom-label best practices in the provided link, I have used shift-select to capture three areas, but I am prompted with the Label assignment warning.

If my goal with this scenario is to extract this text as 16/12/20 in a Date > dmy field, what are best practices for:

labeling actual text e.g., "16" "December" "20" in this scenario

drawing bounding box(es) around these regions if text is blank
SriLakshmi C 3,415 Reputation points Microsoft External Staff

2025-01-31T17:55:49.9066667+00:00

Hi Remy,

The best practice for labeling actual text, such as "16" "December" "20", is to use a single bounding box to encompass the full date and assign it to the Date > dmy field. This approach ensures that the model extracts the date components as a unified entity, allowing consistent formatting as 16/12/20. By avoiding separate bounding boxes for day, month, and year, the model can better recognize and standardize date structures, improving extraction accuracy and reducing post-processing complexity.

Please refer this Labeling guidelines.
SriLakshmi C 3,415 Reputation points Microsoft External Staff

2025-02-03T10:18:46.71+00:00

Hi Remy,

Did you get any chance to check the above response. Thank you!

Share via

Optimal Labeling Approach for Date Extraction in Custom Model

1 answer

Your answer