Skip to main content

Find Frequently Asked Questions

You can use the "FAQ Finder" tool to analyze historic messages in your database and find the most common topics that users have chatted about.

You can use these FAQs to:

  • understand how users are interacting with your chatbot
  • define verified answers for the most common requests
info

The framework and this tool refer to these topics as "frequently asked questions", but it can handle any user input even if they're not in the form of a question.

How It Works​

The FAQ Finder runs a clustering algorithm in your chatbot's vector embedding space. Specifically, it runs DBSCAN on the query embeddings your server generates for each incoming user message.

The tool combines messages into semantically similar groups and then sorts the groups by size (i.e. number of messages in the group). You can then sample or generate a representative question for the entire group.

Use the FAQ Finder​

1. Configure Your Environment​

The FAQ Finder is a script in the scripts package of the chatbot framework repo. You need to configure it before you can run it.

Define a .env file for the scripts package that includes the following values. Make sure to replace the placeholders with your app's values.

/packages/scripts/.env
# The "from" cluster/database denotes where your chatbot conversation data is stored
FROM_CONNECTION_URI=<connection_uri>
FROM_DATABASE_NAME=<database_name>
# The "to" cluster/database/collection denotes where the FAQ data will be stored
TO_CONNECTION_URI=<connection_uri>
TO_DATABASE_NAME=<database_name>
TO_FAQ_COLLECTION_NAME=<collection_name>

2. Run the FAQ Finder​

Once the tool is configured, you can run it from the command line.

From the scripts directory:

npm run findFaq

Or from the repository root:

npm run scripts:findFaq

After you run the tool, the "to" collection defined in your configuration contains the FAQ groups.