We utilize various techniques to estimate text sentiment, focusing on the terms being used and how they are used in combination with one another.
Social listening sentiment numbers tend to be questionable, largely because social sentiment from short-form content like Tweets is difficult to evaluate. Automated sentiment evaluation on longer-form content is better, although still not perfect. One of the reasons we focus our sentiment analysis on reviews and articles is that sentiment and topic modeling tends to be more accurate.
Our sentiment analysis is focused on reviews and articles so that the context within the sentence can be used in assessing sentiment. We have found sentiment evaluation based on brand reviews to be a reliable metric because the data is cleaner and focused; it tends to be specific about the brand and is presented in a (comparatively) more reliable format than many other social channel discussions. For the employee support metrics, we’re currently including only numerical data (number of reviews/numeric ratings given by employees) – presently sentiment isn’t being assessed through text content.
Specifically, you will see a greater level of correlation between the Themes and Keyword sentiment found on our Content Analysis section of BlueOcean Brand Navigator and how they directly tie to the movement of your frequency and sentiment scores within Subfactors of Different, Memorable, Clear, Needed, Desirable and Esteemed.
Since brand content is (generally) positive by default, we are only scoring sentiment for audience content.
Although automated sentiment scoring accuracy varies by context, in our testing we’ve found accuracy rates of about 75%, which is approximately industry average.
