Measuring emotions in news and social media coverage is essential when investigating trends and differing perceptions of various stakeholder groups. webLyzard detects emotions with affective computing algorithms that classify and sort search results along multiple emotional dimensions such as Joy, Trust, Anger and Fear. Through color coding, these dimensions can be used to enrich visualizations such as tag clouds and geographic maps.
Affective models used for opinion mining differ in their goals and complexity. Sentiment analysis, for example, classifies content streams into positive and negative expressions. It uses sentiment lexicons in conjunction with artificial intelligence-based machine learning methods to determine the polarity of sentences and documents. More complex affective models tend to be based on the work of psychologists. They provide more fine-grained classifications into emotional categories. Many of these models also define sub-categories for a more nuanced representation that considers the intensity of the expressed emotion. The following figure compares two popular affective models that can guide the automated classification of human emotions, Robert Plutchik’s Wheel of Emotions and Erik Cambria’s Hourglass of Emotions.
Tailored Models for Data-Driven Communications
Even the most granular affective models combined with sophisticated machine learning approaches do not address the inherent problem that an organization’s strategic positioning goals will often deviate from existing affective models. Certain emotions such as “Joy” and “Trust” represent desirable associations for most brands. Specific communicative goals, however, are carefully formulated by communications managers and marketing professionals. They are typically not restricted to emotions, but can also include desired associations with concepts such “innovation” and “big data”. At the same time, it is equally important to detect undesired aspects of the coverage. Neither the desired nor the undesired aspects correspond to generic frameworks. They stem from the specific competitive position of an organization. The webLyzard Stakeholder Dialogue and Opinion Model (WYSDOM) takes this specific position into account by complementing emotion detection with a tailored success metric that is regularly updated in line with an organization’s evolving communication goals.
Emotion Detection Research Roadmap
Ongoing work in the research projects ReTV (Re-Inventing TV for the Digital Age) and EPOCH (Event Prediction from Hybrid Datasets) is currently extending the emotion detection component to support multiple affective models. The project experiment with different AI-based approaches to achieve maximum accuracy. Initial results of this work have been applied to the online coverage about the coronavirus pandemic in the form of a Corona Mood Barometer, which sheds light on the prevalent emotions in the public debate. The following screenshot shows how the categories of Plutchik’s Wheel of Emotion have been integrated into the Metadata Sidebar of the webLyzard dashboard.
- Susanto, Y., Livingstone, A.G., Ng, B.C. and Cambria, E. (2020). The Hourglass Model Revisited, IEEE Intelligent Systems, 35(5): Forthcoming.
- Scharl, A., Hubmann-Haidvogel, A., Jones, A., et al. (2016). Analyzing the Public Discourse on Works of Fiction – Automatic Emotion Detection in Online Media Coverage about HBO’s Game of Thrones, Information Processing & Management, 52(1): 129-138 [Best Paper Award – Honorable Mention].
- Plutchik, R. (2001). The Nature of Emotions, American Scientist, 89(4): 344-350.
Last major update with release 2020-06 (Sagebrush Lizard).