Sentiment analyzes the tone surrounding individual concepts in your text analysis project. Instead of assigning one sentiment to an entire record, Journey Insights Hub evaluates how a specific concept is discussed within its context.
This means that one record can contain different sentiments toward different concepts. A person might describe one part of an experience positively while expressing frustration with another.
Use Sentiment to answer questions such as:
Which concepts are discussed most positively or negatively?
Which topics receive mixed reactions?
Does the tone around a concept differ across groups?
What language is contributing to the sentiment result?
Is a commonly discussed topic also strongly felt?
How concept-level sentiment works
Journey Insights Hub uses a deep learning model to evaluate the language surrounding each qualifying concept match. It classifies the context as:
Positive
Negative
Neutral
The model considers surrounding language rather than assuming that a word always carries the same sentiment. This is important because the meaning and tone of a concept can change depending on how it is used.
Sentiment percentages should be treated as an indication of the overall pattern. Sarcasm, negation, mixed statements, specialized language, or unusual phrasing can affect automated classification, so important findings should always be checked in the source records.
How sentiment distribution is calculated
Sentiment distribution represents the share of positive, negative, and neutral classifications associated with a concept.
Journey Insights Hub:
Identifies the records in which the concept appears.
Classifies the concept’s context within each qualifying record.
Counts the positive, negative, and neutral classifications.
Calculates the percentage represented by each category.
For example, suppose “work-life balance” is classified in four records:
Two positive
Two negative
Zero neutral
Its sentiment distribution would be 50% positive, 50% negative, and 0% neutral. Sentiment distribution can be especially useful when the data contains no numeric ratings or too few ratings to support meaningful score analysis.
Concepts
The Concepts column lists the words, phrases, or topics included in the current visualization. Use the plus icon to add a concept to your active concepts. A concept that has already been added may appear with its assigned color. You can also use the search field to locate a specific concept.
Exact matches
Exact matches shows the number of records containing the concept itself, including basic word-form variations.
The percentage beside the count shows how much of the selected data contains an exact match. Its denominator depends on whether the visualization is configured to use the currently filtered records or all records in the text analysis project. This percentage measures the concept’s reach. It is different from the positive and negative sentiment percentages.
Negative matches
The negative percentage shows the share of qualifying matches classified as negative.
A result of 75% negative means that 75% of the qualifying sentiment classifications for that concept were negative. The result describes the sentiment classifications for that concept rather than the feelings of everyone represented in the text analysis project.
Positive matches
The blue and red bars make it easier to compare positive and negative sentiment across concepts. A larger bar represents a larger share of the concept’s sentiment classifications, rather than a larger number of people or records.
Always consider the match count alongside the percentages.
Neutral matches
Neutral matches are contexts that are neither clearly positive nor clearly negative.
Neutral is not displayed as a primary bar because the main visualization emphasizes concepts with stronger positive or negative sentiment. You can view neutral percentages and examples by expanding or hovering over a concept and in the exported results.
Positive and negative percentages may total less than 100% because of neutral classifications and rounding.
Expand a concept to review the evidence
Select a concept row to open examples of how it appears in positive, negative and neutral contexts.
The expanded view shows:
The sentiment assigned to the match
The original record text
The matched concept highlighted in context
Relevant metadata associated with the record
Review examples from each sentiment category rather than relying on one record. This helps you determine whether the classification reflects a consistent pattern or whether the concept is being used in several different ways.
Select Download more matches when you need a larger set of supporting records for review, reporting, or further analysis.
Recognize mixed sentiment
A concept with a relatively balanced positive and negative distribution may indicate:
Different experiences across customer groups
Different expectations or preferences
Inconsistent service or product performance
A topic with both benefits and drawbacks
Language whose meaning depends heavily on context
Mixed sentiment can reflect a clear pattern in which the experience varies across the text analysis project. It can be a valuable signal for identifying differences across groups, situations, or expectations.
Apply filters and review the matching records to determine whether the difference is connected with a particular product, region, channel, time period, score, or other metadata field.