What is Sentiment Analysis?

Language is something we use everyday. Research shows that words matter and the emotion and intent in those words matter even more. At AdLabs, we have created a tool to explore these concepts in your everyday verbiage between you and your customers. We are happy to announce, Sentiment Analysis, our newest insights tool for analyzing and optimizing ad text.  This tool can help you better understand the language you use in your Bing or Google Ads accounts and how it can correlate with performance. Sentiment Analysis offers perspective on how your words are connecting with your target audience on a level like never before.

How Sentiment Analysis Works

The tool uses open source affective analysis technology to classify ad text by emotion. To categorize emotion, we use an affective categorization model called the Hourglass of Emotions.  This model uses four independent ranges to model 24 emotions.  Each range is divided into six levels of emotions.  In figure 1 (below), we see a function representing the pleasantness range with the six emotions labeled ecstasy, joy, serenity, pensiveness, sadness & grief.  The x-axis in figure 1 corresponds to the pleasantness value that then computes from your ad text, the value on the y-axis of this function determines the emotion of the text.  Each of the four ranges sensitivity, aptitude, pleasantness, and attention use the same function to classify their values into emotional buckets.  The Hourglass model is a useful tool for visualizing these four ranges together. 

The hourglass model assumes that weak emotions are more similar to each other than stronger emotions.  In figure 2, we see weaker emotions are close together at the neck of the hourglass, while strong emotions are at opposite ends.  For example, on the pleasantness range serenity and pensiveness are more similar to each other than serenity is to joy or sadness is to pensiveness.  

The Sentiment Analysis tool is based on years of research. The tool was carefully developed for AdLabs customers to reach their audience on a deeper level. A variety of perspectives on the effects of language can be revolutionary for clicks and impressions in a digital competitive environment. For more information on the research behind sentiment analysis please refer to the references at the end of this article.

Connect & Add Value to Your Ads

In the AdLabs platform, Sentiment Analysis is already being applied to our active customers. Once logged into the platform, you can find the tool on the left hand side under “Insights”. Next click the “Sentiments” tab and there you will see a bar graph for each of the four ranges with weaker emotions near the center and stronger emotions on the far left and right.  The graph also includes average performance data for each emotion. This helps make identifying which emotions may correlate with negative or positive performance a simple task. Additionally, you are also now able to observe what the associated words are for each sentiment and upon clicking the associated words you can view some of the top ads where these associated words are found – see figure 3 below for an example.

Figure 1 – We see a function representing the pleasantness range with the six emotions labeled ecstasy, joy, serenity, pensiveness, sadness & grief.

Figure 2 – The hourglass model flattened and labeled. The range labels are on the ends of the hourglass and emotional states are labeled on the flattened face of the hourglass. 

Figure 3 – Shows the interactive “Sentiment Insights” tool, which enables an Advertiser to observe a sentiment’s associated words & the text ads which they derive from, in the AdLabs platform.


  • Cambria, E., Poria, S., Hazarika, D., & Kwok, K. (2018). SenticNet 5: Discovering Conceptual Primitives for Sentiment Analysis by Means of Context Embeddings. AAAI.
  • Erik Cambria and Amir Hussain. 2015. Sentic Computing: A Common-Sense-Based Framework for Concept-Level Sentiment Analysis (1st. ed.). Springer Publishing Company, Incorporated.