Language Intelligence (LI) Module

The Language Intelligence (LI) module takes any general piece of input and determines which module it should jump (or proceed) to based on the trained sample data.

This module accepts text as input which can be supplied via the following methods:

1. User responses are stored in variables.

2. Email-related system variables such as subject, body, and attachments.

3. Using APIs such as “OnIncomingEmail”.

   

Application

The LI Module can be used in any type of engagement that requires user input to be classified into one of the defined categories. For example, if you want the user to go through a specific journey based on their responses or classify inbound emails into one of the defined categories for further processing, the LI module can help you achieve this.  

Using this module transforms your structured workflow into a dynamic one. Based on the result of this classification, a Ushur workflow can take users through various paths.

In the example below, the user’s response to the “Question” module will be stored in the “uVar_Response” variable that is used in the LI Module for classification.

Single Topic Classification

Step 1: Uploading Sample Data

The classification engine relies on data samples uploaded as phrases or bulk sections of the next module (all engagement modules are equipped with this option).

The “Jump on Phrase” field in the LI module’s settings is what allows it to proceed to the correct next module based on the Phrases of that module. A Topic describes a specific topic that should trigger the module (i.e. the topic of an email is to send a positive review). A Phrase is a keyword or phrase that the LI module uses to train its model.

The following engagement modules are equipped with Phrases/Bulk upload feature:

• Invisible App

• Multiple Choice

• Open Response

• Message

• Voice

• Email

• Thank You

Before configuring the LI module, first configure the Phrases setting for all your topics. There are two ways to upload sample training data:

• Phrases: Manually add, edit, or delete phrases by directly typing a phrase. When manually entering phrases, the module will only have one corresponding Topic.

• Bulk Phrases: Upload phrases in bulk through a CSV file. The CSV must have “topic,phrase ” as the header (case sensitive) and contain a topic (same across all rows) and corresponding examples in each row.

Step 2: LI Module Settings

After setting up Phrases, it is time to insert and configure the following settings for the LI Module:

• Label: This name uniquely identifies this module within a Ushur workflow.

• Content: The text content (as variable/s) you want to classify can be listed here.

• LI Jumps Section: The next module you want your workflow to jump to if there is a match (data sets are configured in the subsequent module). The LI module will pick the default route (image below) when no match has been found. Select each module that you want the LI module to proceed to. For the given input, the model will determine the topic of the input and proceed to that module.

• Verified: Check this to mark that the LI model has already been trained and accuracy has been verified for that topic.

• Confidence: Set the threshold for the LI module to select the next module to compensate for inaccuracies if your sample data is not balanced (meaning there is a large difference in several sample data from one category to another).

  

Note

Ushur provides an easy-to-use AI platform by hiding AI/ML complexity from the user. Ushur’s data science team can help you with the training, testing, deployment, and optimization process of your ML project. Please contact your customer success manager for more details.

The LI module can be further customized for your use case. Next to the LI Jumps title, there is an indicator for LI model training. Green means the training was successful, orange means the training is in progress and red indicates an error in training. Clicking the indicator will show more information about its status.

To change model and corpus settings, click on the gear icon next to the status indicator. 

• LI Origin: Use the toggle to select the external LI or an Ushur LI model if your organization has the data model that you would like to use. 

• Corpus Type: if using an Ushur LI model, choose the corpus for the LI model to use.  

• LI Model (if using an Ushur LI model): Select the LI model to use.

  

Multi-Topic Classification

Multi-topic can be used in cases such as SmartMail where only classification tags are desired. In this case, a large set of topics and phrases is uploaded in a single CSV file. Classification results will be stored in a variable that can be used or passed to other work management systems so that it can be treated accordingly.

Step 1: Uploading Sample Data

In the following example, the CSV file contains sample emails for 4 categories which have been imported into an engagement module:

Step 2: LI Module Settings

After setting up Phrases, it is time to insert and configure the following settings for the LI module:

• Label: This name uniquely identifies this module within a Ushur workflow.

• Content: The text content (as variable/s) you want to classify can be listed here.

• LI Jumps Section: Next module you want your workflow to jump to if there is a match (data sets are configured in the subsequent module). The LI module will pick the default route (image below) when no match has been found. Select each module that you want the LI Module to proceed to. For the given input, the model will determine the topic of the input and proceed to that module.

• Verified: Check this to mark that the LI model has already been trained and accuracy has been verified for that topic.

• Confidence: Set the threshold for the LI module to select the next module to compensate for inaccuracies if your sample data is not balanced (meaning there is a large difference in the number of sample data from one category to another). 

Note

Click on the ellipsis on the right to assign a Ushur variable to store the topic name.

Testing and Evaluation

You can use an AI Evaluator to test the classification accuracy of your Ushur. The AI evaluator can be activated by clicking on “AI Mode” from the Ushur toolbar.

1. Click on AI in the top right corner of the module to open the AI evaluator window.

2. Enter a sample text.

3. Check for a match in the results table.

You can add the sentence to a specific topic if there is no match to re-train the model with this new sample phrase.