Sentiment Analysis

  • Release version: Xanadu
  • Updated August 1, 2024
  • 2 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Sentiment Analysis

    Sentiment Analysis in ServiceNow's Task Intelligence for Customer Service helps you understand customer emotions by analyzing case and email text. This enables more empathetic and targeted customer interactions by evaluating sentiment at case creation and during case updates. The feature currently supports English-language cases in the Xanadu release.

    Show full answer Show less

    Key Features

    • Sentiment Evaluation: Automatically assesses sentiment in email subject lines, bodies, case descriptions, and customer comments.
    • Sentiment Labels and Scores: Classifies sentiment as Positive (1.0), Neutral (0.5), or Negative (0.0) with a confidence level.
    • Tracking Sentiment Over Time: Updates current sentiment on case updates, compares it to the original sentiment, and tracks trends as Improving, Declining, or Neutral.
    • Data Storage: Sentiment results are stored in the Predictor Results for Task table, accessible by users with the mladmin role for monitoring and analysis.
    • Non-interactive Feedback: The system does not collect agent feedback on sentiment predictions; predictions default to being considered correct.

    Key Outcomes

    • For Agents: Prioritize work based on current sentiment and monitor sentiment trends to ensure cases are resolving positively.
    • For Managers: Route cases to agents with appropriate empathy skills, monitor and reassign cases to prevent escalations, and identify coaching needs by reviewing cases with negative sentiment outcomes.

    Sentiment Analysis can help you gauge customer emotions, enabling you to provide more empathetic and compassionate customer experiences.

    Use the sentiment analysis feature included with Task Intelligence for Customer Service to:
    • Evaluate email and case text.
    • Identify the current sentiment of new cases.
    • Identify the ongoing sentiment of updated cases.
    • Display this information to agents and managers.
    Figure 1. Case list with sentiment fields
    My Cases list view in CSM Configurable Workspace showing positive, negative, and neutral case sentiment fields.

    Agents can use current case sentiment to prioritize their work and ongoing sentiment as it trends over time to see if cases are moving in the right direction.

    Managers can use sentiment to route cases to agents with the right empathy skills, monitor cases and reassign as needed, and avoid escalations. Manager can also identify coaching opportunities by looking at cases that ended on a negative sentiment.
    Note:
    In the Xanadu release, the sentiment analysis feature can predict sentiment for cases created in English.

    Sentiment analysis machine learning models

    Sentiment analysis uses a pre-trained machine learning model to evaluate email and case text and predict sentiment. This analysis takes place when a case is created and when it is updated by the customer.
    Table 1. Sentiment analysis for cases
    Cases scenario Description
    When a case is created
    The sentiment analysis model evaluates the following text to make a prediction:
    • Text in the subject line and body of emails.
    • Text in the short description and description of cases.
    If the model can make a prediction, it returns the following information:
    • A sentiment label and corresponding sentiment level.
      • Positive (1.0)
      • Neutral (0.5)
      • Negative (0.0)
    • A confidence level for the prediction.

    If the model can make a prediction, the sentiment is added to the Original sentiment field.

    If the model can't make a prediction, the Original sentiment is not set.

    This system stores the sentiment prediction information in the Predictor Results for Task table.
    When a case is updated
    The sentiment analysis model evaluates the following text to make a prediction:
    • The text from the body of a reply email.
    • Comments that a customer adds to the case.
    If the model can make a prediction, it returns the following information:
    • An updated sentiment label and corresponding sentiment level.
    • A confidence level for the prediction.
    The system:
    • Updates the Current sentiment field with the current sentiment.
    • Compares the updated current sentiment to the original current sentiment, calculates the change in sentiment, and updates the Sentiment over time field.
      • If there is an increase in the score, the Sentiment over time field shows Improving.
      • If there is a decrease in the score, the Sentiment over time field shows Declining.
      • If there is no change in the score, the Sentiment over time field continues to display the previous value.
      Note:
      If the Original sentiment is Neutral and the Current sentiment is Neutral, then the Sentiment over time is Neutral.

    If the model can't make a prediction, no information gets recorded and the value in the Current sentiment field remains the same.

    For more information about the pre-trained machine learning model, see Create a model to predict case sentiment.

    Prediction feedback

    The system stores feedback on prediction results in the Predictor Result [ml_predictor_results] table. Users with the ml_admin role can access the table and view the results. For sentiment analysis:
    • The default value in the Predicted correctly field for each sentiment prediction is set to true.
    • The Final input value and Final output value fields remain empty because sentiment analysis predictions do not collect feedback from agents.

    The Predictor Result table also stores information about skipped and failed predictions. For more information about this table, see Components installed with Task Intelligence for Customer Service.