Bookshelf v7.5: Prerequisites for Smart Answer Setup

Prerequisites for Smart Answer Setup

To complete the prerequisites for setup, use the instructions in the RME Deployment Best Practices Guide, and the Banter Workbench User's Guide on the third-party documentation CD-ROM. Reading this guide helps you to learn the terminology used to describe Smart Answer.

The prerequisite tasks are performed outside of Siebel eBusiness Applications and should be finished before you set up Siebel Smart Answer.

NOTE: It is strongly recommended that you review the RME Deployment Best Practices Guide, and Banter Workbench User's Guide before performing any prerequisite tasks or Smart Answer setup.

Perform the following tasks in the sequence shown:

Gather and analyze the data you want Smart Answer Manager to categorize.

This data will comprise the source data for the corpus file that you will use to generate the KB file.

Determine the categories that you will use in your KB for classification and training.

Choosing categories is one of the most important steps in building an initial KB file. Categories are classifications assigned to each type of inquiry or request received from customers. You define your own categories based on your business requirements during the Smart Answer planning process. Categories may represent actual content or intent or may identify another attribute of the text, such as the sender's name. Examples of category names might be product complaint, service complaint, and order status.

Smart Answer Manager returns category names and their scores to your Siebel application. eMail Response can use this information to automatically respond to emails, suggest categories of response to agents, route emails, and perform other workflow tasks. Smart Answer Manager stores all data that it needs to correctly identify intents and concepts in the KB, including information about the individual categories. This information is the foundation for the concept models.

Smart Answer for customer or employee applications uses the content items associated with the category to display various content types like Solutions, Decision Issues, Literature Items, and Resolution items.

Install Banter's third-party software. You need Banter Workbench 5.5s to create and train the KB file. This tool replaces both the KB Editor and the RME Jumpstart used in previous releases. The installation files are on the Windows Client Third-Party CD-ROM in the \Thirdpty\Banter folder.

NOTE: It is recommended that you create a local folder on your local machine, name it Smart Answer Tools, and install Banter Workbench in that folder.

Install Banter Workbench 5.5s.

Banter Workbench 5.5s allows you to create or import a corpus of texts, train and test a KB file, and analyze the performance of the trained KB file. Setup.exe installs the software.

Run the Setup.exe file and install it in your local folder. This program also contains a Knowledge Base Editor which helps you create the structure of the KB file. It has two parts: dictionaries and methods that are used to analyze a message, and concept models that are used to detect a message's intent.

It is recommended that you begin with a flat KB, unless you are creating a multilingual KB, for simplicity.

If your KB file grows to contain a large number of categories, you might need to convert it to a hierarchical KB file to increase the accuracy of Smart Answer feedback.

You can use Workbench to turn a flat KB file into a hierarchical KB file by using the KB Editor functionality and manually converting it to hierarchical KB by adding branches for each group of categories. However, if you change the structure of a KB file, you will lose all the learning and feedback that has been accumulated and will have to train and test the KB file again. It is recommended that you keep an updated corpus of emails for retraining.

To use a multilingual KB, you must create a hierarchical KB. For additional information, see Banter Workbench User's Guide.

When creating a hierarchical KB, you must assign branches and categories unique names. For example, you cannot create a branch called Order Status and a category called Order Status. If you assign the same name to a branch and a category, the category name will overwrite the branch name and only the category will appear in the Siebel catalog file.

Category names can be only 100 characters in length.

Start a new project in the Banter Workbench. Follow the steps for starting new projects in Banter Workbench User's Guide.

NOTE: You should name the project with the KB filename for consistency as this will eventually be youvtkb file name to be imported into your catalog and which will reside in the Siebel file system.

Define the corpus.

By defining the corpus, you are indicating which email fields you want to train your KB on. Out of the box, eMail Response sends body and subjects to the categorization engine for categorization. You can add any number of fields to be trained on as long as those fields exist in your corpus.

When you set up the NLP usage options in Workbench, you should associate NLP options with all fields you plan to send to the categorization engine in Smart Answer. If you do this, the confidence levels generated by the "Create Using All Selected, Analyze Using All Selected" option will match the scores generated in eMail Response for a text of incoming email.

Create and edit your corpus file.

A corpus contains your texts for training assigned to categories. A corpus can be created in a number of formats. For a list, see Banter Workbench User's Guide.