Understand concept-level sentiment
Journey Insights Hub evaluates sentiment around individual concepts rather than assigning one sentiment label to an entire record.
This matters because one record can contain several experiences.
For example:
The app is easy to navigate, but the login process keeps failing.
A record-level classification might obscure the difference. Concept-level analysis can recognize positive language around navigation and negative language around login. The model uses the surrounding language and the context of the text analysis project to classify relevant concept matches as positive, negative, or neutral.
Because Sentiment analyzes the language itself, it can provide useful information even when the text analysis project does not include a numeric rating field. If suitable scores are available, you can combine Sentiment with Drivers for a more complete analysis.
Find concepts associated with strong feeling
In Configure, select Concepts linked to strong feeling when you want to discover concepts with especially positive or negative sentiment.
Concepts linked to strong feeling are not automatically a priority list. A concept may show strong sentiment while appearing in relatively few records. Consider sentiment strength together with match volume, business importance, and supporting context. When monitoring established topics, use an active concept list to compare the same business themes consistently.
Compare sentiment across groups
Apply a meaningful filter to compare the same concept across products, regions, departments, customer segments, channels, ratings, or time periods.
For a reliable comparison:
Keep the concept definition consistent
Change one filter at a time
Check the number of matches in every group
Review examples from each sentiment category
Look for differences in context, not only percentage
The concepts linked to strong feeling may change when you apply a filter because the platform evaluates the sentiment evidence available within the selected records. These comparisons can reveal whether a topic is broadly viewed the same way or whether the experience varies across the organization or customer journey.
Extract representative evidence
Select a concept and open Records to review the records behind its sentiment distribution. Look for:
Clear examples of positive, negative, and neutral language
Negation or indirect wording
Industry-specific expressions
Records containing both positive and negative experiences
Different meanings grouped under the same concept
Exceptions to the overall pattern
Records are also useful for finding representative quotations for presentations or reports. Select quotations that reflect the wider pattern rather than choosing only the most dramatic language.
Pair Sentiment with Drivers and Volume
Sentiment, Drivers, and Volume provide complementary information:
Sentiment shows the qualitative tone surrounding a concept.
Drivers shows whether the concept is associated with a different numeric outcome.
Volume shows how widely the concept appears.
For example, Drivers may show that login is associated with lower ratings. Sentiment can then help distinguish between a mild inconvenience, repeated frustration, and a serious inability to access the service. Volume shows how many records are affected.
A concept can be highly negative but appear in very few records, or it can be widely discussed with only mildly negative sentiment. Consider all three perspectives before deciding how significant a finding is.
Validate the interpretation
Sentiment provides a general indication of emotional tone and works best alongside a review of the underlying feedback. Before sharing a result:
Confirm the number of matching records
Review more than one example
Check whether the concept has multiple meanings
Compare relevant groups
Consider whether the dataset represents the affected population
Confirm important conclusions using the original text
The strongest Sentiment findings combine emotional direction, volume, variation across groups and representative evidence.