NOVA: A visual interface for assessing polarizing media coverage

TL;DR

An interactive visualization system for the public to assess their perception of the mainstream media's coverage of a topic against the data, which combines belief elicitation techniques and narrative structure designs, emphasizing transparency and user-friendliness to facilitate users' self-assessment on personal beliefs.

Abstract

Within the United States, the majority of the populace receives their news online. U.S mainstream media outlets both generate and influence the news consumed by U.S citizens. Many of these citizens have their personal beliefs about these outlets and question the fairness of their reporting. We offer an interactive visualization system for the public to assess their perception of the mainstream media's coverage of a topic against the data. Our system combines belief elicitation techniques and narrative structure designs, emphasizing transparency and user-friendliness to facilitate users' self-assessment on personal beliefs. We gathered $\sim${25k} articles from the span of 2020-2022 from six mainstream media outlets as a testbed. To evaluate our system, we present usage scenarios alongside a user study with a qualitative analysis of user exploration strategies for personal belief assessment. We report our observations from this study and discuss future work and challenges of developing tools for the public to assess media outlet coverage and belief updating on provocative topics.

Figures (11)

• Figure 1: The data transformation process of NOVA. Collected news articles were preprocessed with entity linking and sentiment analysis. Then articles are aggregated by entities and further aggregated by co-occurrences to represent topics. Sentiment scores are generated with descriptive statistics for each topic. The preprocessed article data, co-occurrences data, and topic sentiment data are all stored on a server and requested from the front end. Through user interaction, the aggregated sentiment type of each entity is categorized as neutral, positive, negative, or mixed. Green lines indicate data communication between the server and the front end.
• Figure 2: Sentiment Scatter Plot. Each dot encodes a topic. The color of the dot encodes the number of articles associated with the topic, using a logarithmic color scale. Coordinates encode the two-dimensional sentiment of the topic's coverage, using $score_{pos}$ as the x-axis and $score_{neg}$ as the y-axis. A segmentation controller divides the scatter plot into four regions: neutral, positive, negative, and mixed. Topics that fall into each region are classified accordingly, which are used in other parts of the system. Finally, a fish-eye effect is added to mitigate the cluttering issue.
• Figure 3: Topic Hive first design. Each cell represents a topic. The hive is built around the center cell (United States). Surrounding cells represent the most frequently co-occurring topics with the center cell (topic). Distance of surrounding cells to the center cell encodes the frequency of co-occurrence: closer cells (topics) co-occur more frequently with the center cell (topic). Cell color encodes sentiment, following the sentiment encoding in the Sentiment Scatter Plot.
• Figure 4: Topic Selection stage, users use the Sentiment Scatter Plot and Topic Hive together to discover interesting topics. An Article Threshold can be used to filter topics by the number of associated articles. Selecting a topic from the Sentiment Scatter Plot will trigger the corresponding Topic Hive. Users can choose a pair of co-occurring topics for further investigation in the next stage.
• Figure 5: Belief Elicitation stage, users are prompted to select one from five randomly generated hives that best represent their belief. This process is repeated to two randomly selected outlets. The decisions made by the users are used for extrapolation in the Outlet Comparison stage.