Entity-relationship Annotation

Create annotation task

Log into the management console, enter “ Baidu Machine Learning (BML) > Data Annotation”, and click the “Create annotation task” button above the list:

Enter the necessary information in the pop-up window for creating an annotation task:

Contents to be input include:

Name of the annotation task (required): Name of the annotation task is composed of English characters, numbers and underlines. It cannot start or end with underlines, and the length is 2-30 characters.
Annotation scenario (required): Select “entity-relationship” annotation.
Data storage path (required): Storage path of the file to be annotated.
Annotation task description (optional): It is used to introduce specific annotation rules to annotation personnel, and support doc, docx and pdf formats.
Entity type (optional): Entity type used for annotation, such as “person name” and “company name”. If it is blank, you can enter “Tag management” during annotation to add one.
Relationship type (optional): Relationship type used for annotation, such as “father-child” and “husband-wife”. If it is blank, you can enter “Tag management” during annotation to add one.
Review required or not (required): You can select yes or no. If the data do not have to be reviewed after annotation, you can select “No”; if the data have to be reviewed, you can select “Yes” for second confirmation of the annotated information.
After clicking OK, you can see that the annotation task is created successfully.

Upload annotation data

Click “Upload” on the annotation list page or “Upload data” on the annotation task details page to add data for the annotation task.

Support “Local upload” and “Select from BOS” for data upload.
Support “Single file” and “Compressed package” for uploading. When upload a single file, support the file types of txt, doc, docx and pdf. A maximum of 4 files can be uploaded at one time, and each file shall not exceed 2M.

You can also directly select a single file from BOS to upload:

Click OK to enter the annotation task details page. After the data are uploaded and processed, you can see the file.

Conduct data annotation

After entering the annotation task details page, you can click “Annotation” on the data overview page or click the “Manual annotation” tab to enter the annotation page:

The entity-relationship triples are annotated with sentence as the unit. The BML annotation system will automatically segment the sentences in the files uploaded by users, and the default separators are Chinese and English periods, question marks and exclamation marks.？！）
Before annotation, you have to click tag management to add annotation tags.

For example, in this case, we add a “person name” tag in the entity tag and add tags of “uncle-nephew”, “father-child” and “husband-wife” in the “relationship tag”.

During annotation, select the content to be annotated first, and then select the corresponding tag in the pop-up window, for example:

In the annotation process, you have to respectively select entity 1, relationship and entity 2. After this, click “Submit” on the right so that the annotation can be effective.

After annotation, click the “Save and go to next file” button to finish annotation of this file.

Conduct data review

Click the “Result review” tab to review the annotation information which has been completed. If all annotations are correct, click the “Pass” button. If the annotations are incorrect, click the “Fail” button. For failed files, you can annotate them again on the manual annotation page.

View information about annotation tasks

Click “Annotation Task > Task Management” to check the information about annotation tasks, including task progress, annotation task description, upload history of data to be annotated and result export history.

Entity-attribute annotation

Create annotation task

Log into the management console, enter “Baidu Machine learning (BML) > Data Annotation”, and click the “Create annotation task” button above the list

Enter the necessary information in the pop-up window for creating an annotation task:

Contents to be input include:

Name of the annotation task (required): Name of the annotation task is composed of English characters, numbers and underlines. It cannot start or end with underlines, and the length is 2-30 characters.
Annotation scenario (required): Select the “Entity-attribute” annotation.

Data storage path (required): Storage path of the file to be annotated.

Annotation task description (optional): It is used to introduce specific annotation rules to annotation personnel, and support doc, docx and pdf formats.

Entity type (optional): Entity type used for annotation, such as “person name” and “company name”. If it is blank, you can enter “Tag management” during annotation to add one.

Attribute type (optional): Attribute type used for annotation, such as “career” and “nationality”. If it is blank, you can enter “Tag management” during annotation to add one.

Review required or not (required): You can select yes or no. If the data do not have to be reviewed after annotation, you can select “No”; if the data have to be reviewed, you can select “Yes” for second confirmation of the annotated information.

After clicking OK, you can see that the annotation task is created successfully.

Upload annotation data

Click “Upload” on the annotation list page or “Upload data” on the annotation task details page to add data for the annotation task.

Support “Local upload” and “Select from BOS” for data upload. Support “Single file” and “Compressed package” for uploading. When upload a single file, support the file types of txt, doc, docx and pdf. A maximum of 4 files can be uploaded at one time, and each file shall not exceed 2M.

You can also directly select a single file from BOS to upload:

Click OK to enter the annotation task details page. After the data are uploaded and processed, you can see the file.

Conduct data annotation

After entering the annotation task details page, you can click “Annotation” on the data overview page or click the “Manual annotation” tab to enter the annotation page:

The entity-attribute is annotated with sentence as the unit. The BML annotation system will automatically segment the sentences in the files uploaded by users, and the default separators are Chinese and English periods, question marks and exclamation marks. ？ ！）
Before annotation, you have to click tag management to add annotation tags.

For example, in this case, we add tags of “person name” and “apple” in the entity tag and add tags of “nationality”, “career” and “color” in the “attribute tag”.

During annotation, select the content to be annotated first, and then select the corresponding tag in the pop-up window, for example:

In the annotation process, you have to respectively select entity and attribute. After this, click “Submit” on the right so that the annotation can be effective.

After annotation, click the “Save and go to next file” button to finish annotation of this file.

Conduct data review

Click the “Result review” tab to review the annotation information which has been completed. If all annotations are correct, click the “Pass” button. If the annotations are incorrect, click the “Fail” button. For failed files, you can annotate them again on the manual annotation page.

View information about annotation tasks

Click “Annotation task>Task management” to check the information about annotation tasks, including task progress, annotation task description, upload history of data to be annotated and result export history.