Evaluate Words Used in Multiple Intents

The DRUID Portal enables you to see for each bot an image composed of words used in the training model.

The WordCloud page gives you an overview of the training model on a word level. The size of each word indicates its frequency throughout the entire NLP model / test phrase set.

This tool helps you identify the words that appear frequently and might affect the ML model:

1. On the bot NLU menu, click Evaluation.
2. Click on the WordCloud tab.
3. Select the Data Set (Training Set or Test set).
4. From the Model field, select the model you want to evaluate (either the bot model or a parent / child model).
5. Click the Refresh button and click the WordCloud section header.

The bigger a word is in the image; the higher are the chances that it is a candidate for stop words.

Once you identify such words, you have two options:

• Mark the words as stop words.
• Change flow utterance.

Mark the words as stop words

To mark words as stop words, follow these steps:

1. On the NLU menu, click Configurations.
2. Click on the Intents tab, then click on the Stop words section header.
3. Enter the words identified in WordCloud.
4. Click the Save button to save the changes.
5. Click the Train button to train the bot.

If you go to Evaluation > WordCloud and do not select the ShowStopWords check box, the image will no longer show the words you’ve marked as stop words.

Change flow utterance

Scroll-down to the Report section and check the word: on how many flows and utterances it appears.

To view which flows contain the word, click on the word.

To view the utterances that contain the word on a specific flow, click the arrow in front of the flow.

If you want to edit a specific utterance, click on it. The Flow Details page appears by default on the Training Phrases.

After you edit the flow utterance, click the Train Set tab and click the Train button to train the bot with the new training phrase.