You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

85 lines
3.8 KiB

  1. # Setup Auto Labeling
  2. In this tutorial, you will learn how to set up and use the auto-labeling feature. Auto-labeling is a feature that automates labeling using the Web API. This is not required to use doccano, but if you set it up, you will be able to label data more efficiently.
  3. The tutorial is divided into several sections:
  4. - Select a Template will give you a starting point to follow the tutorial.
  5. - Set Request Parameters will teach you how to set parameters required to send a request.
  6. - Specify Response Mapping will teach you the way to extract the label information from the response.
  7. - Specify Label Mapping will give you the way to map the extracted labels to the internal labels of doccano.
  8. - Enable the Feature will show you how to enable the auto-labeling.
  9. In this tutorial, we will show you how to set up auto-labeling using Amazon Comprehend Sentiment Analysis as an example. Therefore, we assume that you have a text classification project in doccano, an AWS account and be able to generate access keys.
  10. ## Select a Template
  11. First, move to the "settings" page and open "Auto Labeling" tab. The new tab should display a "Create" button and an empty table. Click the button and select "Amazon Comprehend Sentiment Analysis" from the dropdown menu:
  12. ![](../images/auto-labeling/select_template.png)
  13. ## Set Request Parameters
  14. Next, you need to set parameters to send an API request. In the case of Amazon Comprehend Sentiment Analysis, the following parameters are required:
  15. - aws_access_key
  16. - aws_secret_access_key
  17. - region_name
  18. - language_code
  19. In the following example, we set `us-west-2` as a `region_name` and `en` as a `language_code`:
  20. ![](../images/auto-labeling/set_parameters.png)
  21. Then, we will test them using the sample text to make sure whether the parameters are set correctly or not. In this case, we set "I like you" as a sample text and be able to get the response from Amazon Comprehend Sentiment Analysis. If you look at the Sentiment field in the response, you will see that its value is POSITIVE:
  22. ![](../images/auto-labeling/test_parameters.png)
  23. ## Specify Response Mapping
  24. Now, you can successfully fetch the API response. Next, you need to convert it to doccano format(below) with the mapping template([Jinja2](https://jinja.palletsprojects.com/en/2.11.x/) format).
  25. ```plain
  26. Text Classification
  27. [{ "label": "Cat" }, ...]
  28. Sequence Labeling
  29. [{ "label": "Cat", "start_offset": 0, "end_offset": 5 }, ...]
  30. Sequence to sequence
  31. [{ "text": "Cat" }, ...]
  32. ```
  33. In the case of Amazon Comprehend Sentiment Analysis, we want to get `Sentiment` value from the response. As we can access the entire response by the `input` variable, the mapping template looks like the following:
  34. ```json
  35. [
  36. {
  37. "label": "{{ input.Sentiment }}"
  38. }
  39. ]
  40. ```
  41. After setting the template, we will test them using the sample response. This response is the same one we fetched in the `Set Request Parameters` section.
  42. ![](../images/auto-labeling/test_mapping_template.png)
  43. ## Specify Label Mapping
  44. Once you specify the mapping template, you need to convert the label in the response into the one you defined at the label page.
  45. Click the `Add` button and fill in the `From` and `To` fields. `From` means the response label string. In this case, we can specify `POSITIVE`. `To` means the label of this project. In this case, we specify `positive`:
  46. ![](../images/auto-labeling/add_label_mapping.png)
  47. After adding the label mapping, we will test them using the sample response:
  48. ![](../images/auto-labeling/test_label_mapping.png)
  49. ## Enable the Feature
  50. Finally, move to the "annotation" page and click "Auto Labeling" button. It should display a "Slide" button for switching enable/disable auto-labeling feature. Try to enable it:
  51. ![](../images/auto-labeling/enable.png)
  52. Each time you view a new document, it will be labeled automatically.