How to Manage Data

You can check and revise the data you integrated to your NLU API Dashboard. If necessary, you can upload more data as well.

Check and Revise Inference Result

Integrated data goes through inference by the AI model. You can check the result and revise it from the dashboard. Here's how to do that for each API type.

Text Classification, Sentiment Analysis

You can check the inference result in the Dashboard menu and the Inference menu.

In the Dashboard menu, you'll see the inference result in the Inference column. In the Confirmed column next to it, you can input the actual result manually when the inference result is incorrect. You can train your AI model using this data.

In the Inference menu, the inference result appears in the Intent column. If you click the exclamation mark under the Diff column, you can see the detailed information including the inference result, manually corrected answer, etc. The exclamation mark appears as blue if the inference result is the same with the correct answer and red if not.

NER

You can check the inference result in the Dashboard menu and the Inference menu. In both menus, you can find the inference result in the Phrase column.

Unlike the Text Classification or Sentiment Analysis, you can revise the inference result in the Phrase column under the Dashboard menu. Drag to select the keyword to create a new named entity, or mouse over and click the X icon on an existing entity to delete it.

In the Dashboard menu, the Phrase column will show the revised result if you make any revision. It means that each phrase in the table can be either an original inference result or a revised one. If you want to check out the original result after a revision, click the status in the Status column.

In the Inference menu, the result shown is always the original inference result regardless of your revision. If you click the exclamation mark under the Diff column, you can see the detailed information including the inference result, manually corrected answer, etc. The exclamation mark appears as blue if the inference result is the same with the correct answer and red if not.

Review Conversion

You can check the inference result in the Dashboard menu and the Inference menu. Check out the Review column under the Dashboard menu and the Inference column under the Inference menu. Please note that there is no separate Status column in the Review Conversion dashboard.

In the Dashboard menu, you can check the inference result and revise it. Click the + button to add a new review or to change the score of the existing reviews. You can remove an existing review by clicking on the X icon on the review.

In the Inference tab, you can see both the revised result and the original inference result in the table. The revised results are under the Tagged column with red text for any revision made, and the original inference result under the Inference column.

Status

In the Status column under the Text classification, Sentiment Analysis, and NER dashboard, you can see four different statuses depending on the inference status and how it is confirmed by agents.

  1. Inferring: Inference is not done yet. Once inferring is done, the status changes to 2, 3, or 4.

  2. Not Confirmed: Inference is done and either it's not confirmed by an agent or the training is turned off for the phrase.

  3. Correct: Inference is done and the result is confirmed correct by an agent. There are two possible scenarios for this status. 1) When the inference is done and the training is turned on for the phrase. Therefore if you want to confirm that the inference result is correct, you can simply turn on the training toggle without revising the result. 2) When there's tag information included in the uploaded data and the inference result is the same with the tag(s). In this case, the training toggle and the test toggle is turned on automatically.

  4. Incorrect: 인퍼런스 결과와 고객이 확인한 바가 다른 상태입니다. 인퍼런스된 결과를 수정하시거나, 데이터 업로드시 기입하신 태그와 인퍼런스 결과가 다를 경우, 트레이닝 버튼을 켜시면 해당 상태가 표기됩니다.

Apply Filters

You can apply filters for the data you see on the dashboard. These are the different sets of filters you can apply for each project type.

Text classification, Sentiment Analysis

'Confirmed' column state, data type (training / test / all), date created.

NER

Data type (training / test / all), date created.

Review conversion

Review category, data type (training / test / all), date created.

Upload Data

You can directly upload more data for training or test.

  1. You can enter the phrase you want to add. The phrase appears in the table right away.

  2. You can upload supporting documents with multiple phrases in it. The document may or may not include the tag (manual analysis result) info as well. Check out the supported file types and the data format for each project type below.

Common

If you don't want to include tag data and only want to upload phrases, TXT, TSV, JSON files are supported for every project type. Each phrase should be started in a new row. Please see the examples below.

<TXT, TSV>

What's the age limit for the policy A?
How can I request for a claim?

<JSON>

[
{
"text": "What's the age limit for the policy A?"
},
{
"text": "How can I request for a claim?"
}
]

If you want to include tag data as well, please see below.

Text classification, Sentiment Analysis

TSV, JSON types are supported.

<TSV> Category and the phrase should be in the same row, divided by tab.

PolicyDetail What's the age limit for the policy A?
ClaimQuestion How can I request for a claim?
Positive Service quality was far beyond expectation!
Neutral Not suberb, but not terrible.
Negative Won't buy here again. Aweful product quality.

<JSON>

[
{
"text": "What's the age limit for the policy A?",
"intent": {
"name": "PolicyDetail"
}
}
]

Key

Description

Data Type

text

The phrase to upload

String

intent

An object to put the category info in

Object

name

Category of the text

String

NER

Supports JSON type only.

<JSON>

[
{
"text": "Can I combine the policy with my auto policy?",
"entities": [
{
"start": 6,
"end": 23,
"token": "combine the policy",
"tag": {
"name": "REQUEST"
}
},
{
"start": 33,
"end": 43,
"token": "auto policy",
"tag": {
"name": "POLICYTYPE"
}
}
]
}
]

Key

Description

Data Type

text

The phrase to upload

String

entities

An object to define the attributes of the named entities

Array(object)

start

Where the named entity starts in the phrase. Starts from 0 and increased by 1 per each character, including space and the special characters

Integer

end

Where the named entity ends in the phrase. Starts from 0 and increased by 1 per each character, including space and the special characters

Integer

token

The keyword detected as the named entity

String

tag

An object to put the named entity (category of the keyword) in

object

name

The named entity (category) of the detected keyword

String

Review Conversion

Supports TSV and JSON.

<TSV> The first row works like a header row of a table. After the first title is 'Review', the categories to analyze the reviews with follows, separated by tab key. From the second row, the review data and the ratings for each category come in the order of the first row, separated by the tab key.

REVIEW PRICE QUALITY COLOR DESIGN
A bit pricy but comfy, good finish and color. -1 1 1
Perfect design, much better quality than expected. 1 2

Key

Value

REVIEW

A bit pricy but comfy, good finish and color.

PRICE

-1

QUALITY

1

COLOR

1

DESIGN

N/A

<JSON>

[
{
"text": "Had to replace it but great customer service and follow-up!"
"review_sentiments": [
{
"category": "Quality",
"sentiment": "-1"
},
{
"category": "Service",
"sentiment": "2"
}
]
}
]

Key

Description

Data Type

text

The phrase to upload (Review)

String

review_sentiments

An object to put the review categories and the scores

Object

category

The categories to analyze the review

String

sentiment

The score for the category, a number between -2 and 2

String

Named Entity Dictionary

NER project has the Named Entity Dictionary menu. You can add, delete or download named entities and the keywords you want the AI to detect here.