Configuring sentiment

Medallia Experience Cloud analyzes sentiment by applying the Action Intelligence sentiment model to unstructured text feedback. This includes text from:

  • Text fields in surveys
  • Chat transcripts
  • Medallia Speech audio files transcribed to text

If you are planning to implement sentiment, first evaluate whether sentiment processing is available for your language, as listed in Text Analytics supported languages.

Restriction: Sentiment restricts how you build topics.

When sentiment is not enabled, the system finds topic matches across an entire comment. However, when sentiment is enabled, the system only finds topic matches within the same sentence.

This change can have the biggest impact on rules you create that use the NOT logic component. When sentiment is configured, the system will only eliminate comments where the NOT component is in the same sentence as the rule that matches the topic. For more information, see Rule components, How rules are applied, and Word matching in standard topics sections in Standard topics.

Configuring sentiment

Text Analytics uses the Action Intelligence sentiment model for processing sentiment.

  1. Navigate to the Reporting > Text Processing > Topic & Theme Tagging > Settings screen.

  2. Change TextAnalyticsVersion to SENTIMENT_SUPPORT.

    Important: Once you change the TextAnalyticsVersion property to SENTIMENT_SUPPORT, you cannot undo this action and go back to TOPICS_ONLY.

  3. Click Save.

    After saving, you see boxes populated with the Q-fields from your tag pool (starting halfway down the page).

  4. For questions with an inherently positive or negative sentiment, apply that sentiment to the question fields. Text Analytics takes the default sentiment into account when calculating sentiment for text entered in these fields.

    • Move any Q-fields with a positive context (such as, Tell us how we exceeded your expectations) into the MemberPositiveSurveyFields box.
    • and move any Q-fields with a negative context (such as, Tell us what we could have done better) into the MemberNegativeSurveyFields box.

    Fields not moved in either box are assumed to have neutral sentiment by default.

  5. Navigate to the Reporting > Text Processing > Topic & Theme Tagging > Sentiment screen.
  6. To force neutral sentiment for specific channels or fields, set one or more of the following properties and then click Save.
    • Neutral sentiment for Agent side of conversational data — When on, sentiment for the agent side of all conversational data, such as data in the q_speech_comment and q_conversational_transcript_comment fields, is calculated as neutral.
    • Neutral sentiment for Customer side of conversational data — When on, sentiment for the customer side of all conversational data is calculated as neutral.
    • Comment Fields for which Sentiment is Disabled — Fields in the ride-side box are forced to have neutral sentiment. For example, you might want case notes fields to be neutral.
      Note: Fields selected in the Comment Fields for which Sentiment is Disabled property override settings for those fields in the MemberPositiveSurveyFields and MemberNegativeSurveyFields properties.

  7. Navigate to the Reporting > Text Processing > Global Text Analytics Settings > Settings screen.

  8. For each of the languages you are processing with Text Analytics, turn on the Use Medallia AI Sentiment Model property if it appears. See Text Analytics supported languages for the list of languages with sentiment processing available.

  9. Click Save.
    Note: Immediately after you save the page, the sentiment analysis process is applied to all incoming surveys.

  10. (Optional) Navigate to the Reporting > Text Processing > Sentiment Reporting screen to alter the default reporting options that apply to sentiment. See Properties of the Sentiment Reporting screen, below, for a description of each reporting option.

  11. Validate your configuration and initiate comment processing after the initial setup process is complete. Make sure to choose a processing option that applies sentiment. See Text Analytics processing and initialization for more information.

Properties of the Sentiment Reporting screen

These properties influence how the system displays sentiment results in Text Analytics reports. These properties only apply when sentiment is enabled.

Full filter enabled
When on, all sentiment categories are displayed in reports. When off, the Neutral and Mixed sentiment categories are included or excluded from reports.

When full filter is not selected, the mixed and no opinion sentiment categories do not appear.

Note: This setting does not apply to custom reports with fixed sentiments, such as in the following example:

Fixed sentiments are specified in the AA2 XML for the custom report

Sentiment filter disabled
When on, the sentiment filter is available in the Category Filter dropdown. By default, a sentiment filter appears as a second dropdown of the Category Filter in many reports, including:

  • Custom reports with the SURVEY_TAG or SURVEY_TAG2 optionbox filters specified in the AA2 XML

  • Dashboards with CATEGORY_FILTER or CATEGORY FILTER2 specified for the MemberHomepageOptionboxPlugins field

  • Responses Filter report

There are two filters in the Category Filter. The sentiment filter allows you to filter by sentiment category.