Test NLP
There are 2 ways to test NLP recognition results after training: Test each sentence (Test NLP) and multiple test
Last updated
There are 2 ways to test NLP recognition results after training: Test each sentence (Test NLP) and multiple test
Last updated
NLP testing shows the model's recognition results after training.
Step 1: Click Test NLP (1) in the quick access toolbar.
Step 2: At the display interface, enter the content of the sample sentence to be checked (2) then click Analyze (3) for the system to conduct the check.
Step 3: The result returns the Confidence index of the recognition result. Users evaluate the results of identifying the Intent, Entity Type, and sentiment index of the sample sentence just tested. Then, make necessary corrections if the recognition results are not accurate.
Trainers upload sample sentence files to the system for mass testing according to the following instructions:
Step 1: At the quick access toolbar, the user clicks on the Multiple Test icon (1), then clicks Upload (2)
Step 2: Select (Click here to download sample file) (3) download the NLP test sample sentence file.
Step 3: Enter the downloaded file with sample sentence data, intent, and the corresponding entity that needs to check the system's recognition results.
Step 4: After entering data, select file (4) to upload the sample test sentence file with imported data.
Step 5: Select Processing (5) then click Runing (6) for the system to conduct a series of checks.
Multiple test results in chart form indicate the following indicators:
The “Confident” index indicates the reliability of intent recognition results. Users can check sentences with an average or low confidence index on the Overview home page, to make necessary corrections to improve results.
Average level: Ranges from 50-79 (on a scale of 100).
Urgency level: Low range from 0-49 (on a scale of 100).
Index
Calculation
Correct Intent rate
The number of sentences with correct intent recognition results out of the total number of sentences tested.
Correct Entity Rate
The number of sentences with correct entity recognition results out of the total number of sentences tested.
Pass rate
The number of sentences that meet both criteria out of the total number of sentences tested.