How to
Using the Analyse sentiment node

The Analyse sentiment node in Pendula enables businesses to evaluate customer sentiment from inbound messages. By analysing tone and intent, this node helps workflows respond dynamically based on customer emotions, improving engagement strategies.

How it works

The Analyse sentiment node processes incoming messages and categorises them based on sentiment. This allows businesses to:

• Identify positive, neutral, or negative sentiment in customer responses.
• Route messages dynamically within workflows based on sentiment analysis.
• Automate escalations or tailored messaging strategies based on customer emotions.

Configuring the Analyse sentiment node

Customising the prompt

The prompt is structured into three sections, two of which are pre-defined and cannot be edited by users. These provide crucial instructions for sentiment classification.

Prompt introduction

The following context is included in every request:

Context: A mobile phone and internet provider requires analysis of the sentiment in text messages they receive from their customers.

Task: Determine the sentiment of customer responses based on the supplied message.

Prompt body

The body of the prompt is configurable and comes with a default example. Best practices include:

• Always provide the context of the customer message.
• Use double asterisks * to indicate bold headings.
• Use single quotes to indicate specific examples or quotes.
• Clearly define sentiment categories with examples.

Prompt conclusion

The following instructions are included at the end of each request:

Instructions:

• Always classify sentiment as either POSITIVE, NEUTRAL, or NEGATIVE.
• Follow the sentiment definitions and examples provided to ensure consistency.

Configuring the node

The Customise prompt field includes a default example which can be adjusted for specific use cases. Use the merge field explorer to select the messageBody field from the relevant Conversation or Inbound Message Trigger node to specify the text that will be analysed.

The Prompt tester allows users to test their prompt with sample responses, returning a classification along with a Confidence Rating and Confidence Rating Reason, explaining the result.

Outputs

The classification output is stored in the sentiment merge field, ensuring that only valid sentiment classifications are used.

Outcomes

• Success: The workflow continues if the sentiment is successfully classified.
• Failed: If an unexpected failure occurs, the experience can either end (if the path is disabled) or continue to the next configured node (if the path is enabled).

Best practices

• Ensure clarity in inputs: Provide structured input data for better accuracy.
• Handle uncertain cases: Define fallback actions for ambiguous sentiment.
• Monitor performance: Regularly review classification results to refine accuracy.