Automated sentiment analysis shows a visual representation of sentiment per response.
When viewing responses, the sentiment is hidden by default. You can see the sentiment by clicking the View Sentiment link next to the Export link on the top gray bar.
To be analyzed, a response must have at least 25 characters. Responses with fewer characters will not have sentiment highlighted.
When sentiment is visible, it shows all responses that have sentiment associated with it. When you click on a colored word or phrase, you'll see the exact value of each sentiment type, scaled from -1 to 1.
Sentiment Value Scoring
The machine-learning algorithm used to analyze a response returns a relevance value for each word or phrase it analyzes. The algorithm takes into account the relevance of the sentiment and gives more weight to words/phrases with higher relevance.
Relevance is defined as how relevant the machine-learning algorithm feels the word or phrase is to the entire response.
A higher relevance score means it is more likely to be important in the context of the full response, and is, therefore, weighted heavier in the score breakdown. For example, if a statement has one positive, one neutral and one negative statement, the percent negative could be much higher (such as 70%) if the relevance score for the negative statement is higher than the positive or neutral statement.
Sentiment at the Word/Phrase Level
When you enable sentiment, you'll also see each word or phrase that has sentiment associated with it highlighted by color. The machine-learning algorithm used in Fuel Cycle returns a sentiment score for each word or phrase that it determines has sentiment. Based on that score, all words/phrases with sentiment are highlighted as follows:
- Green highlighted text = positive score
- Yellow highlighted text = neutral score
- Red highlighted text = negative score
Clicking on a green or red-highlighted word will show the sentiment score given to the word. You can click multiple words to open more than one score at a time. Clicking the word a second time will close the popup. Neutral, yellow words are not clickable because the score is always 0 for a neutral word.
Sentiment Data Included in Exports
Whenever you export data from the Qualitative Reporting tool, the sentiment data will also be included in the file as follows:
- Overall sentiment breakdown percentages will be included with each response as columns for % Positive, % Neutral, % Negative
- Sentiment Keywords and Sentiment Entities are new sheets in the Excel file. For each keyword or entity with the associated sentiment, the individual sentiment score and relevance will be included.
Machine Learning and Fuel Cycle
As you work with the Qualitative Reporting tool, you might have questions about the data and how to use it. For information about Fuel Cycle's view on the role of machine learning in sentiment analysis and text analytics, please click here.