Check out the full paper for NOVA on arxiv: https://arxiv.org/abs/2403.00334
``Echo chamber'' is a common phenomenon found in the general public in the U.S. and can exacerbates the political polarization. Such polarization presents an unresolved challenge in computationally analyzing media bias, in which the public is divided and can hardly reach agreement on media bias, making it hard to find ground truths. We attempt to create a platform, NOVA, where people assess personal beliefs on media biases by finding evidence articles. Such a platform exposes them to both supporting and countering articles, thus mitigates the effect of echo chambers and polarization. We decided to incorporate presentation narrative structure and visual belief elicitation techniques after two design iterations, during which pilot studies and user studies are conducted to refine and evaluate the design to cater to the general public. We report lessons learned from designing and evaluating NOVA for belief elicitation and assessment and identify opportunities and challenges for prospective research.
The data transformation process 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 requested from the frontend.
Green lines indicate data communication between the backend and the frontend.
We design a workflow in which users first externalize their belief through a visualization, then contrast the visualization to the data-driven one, and finally find evidence to explain the discrepancies.
Namely, there are 3 stages: Topic Selection, Belief Elicitation, and Article Review.
In our design, people are allowed to navigate between stages as they please, while the system keeps track of the data and any changes between stages.
We incorporate modals and annotations to help direct user attention and to provide context for what they are assessing.
The system, in the first two stages, adjusts the visualizations to reflect their beliefs allowing them to form a hypothesis based on their existing beliefs and then evaluate them in the third stage.
A screenshot of the Topic Selection stage.
(a) A table of entities and their number of mentioned articles.
(b) The sentiment scatter plot shows two-dimensional sentiment for each topic. The region is divided into four categories: mixed, positive, negative, and neutral.
(c) The utility panel contains a filter and a color scale legend on article frequency.
(d) By choosing a topic from (a) or (b), users can see the statistics for the topic, and they can choose an outlet to inspect further.
A screenshot of the Belief Elicitation stage.
(a) To externalize a user’s belief, the user can drag the topic hexagons to the corresponding sentiment category or click the center hexagon to adjust its sentiment.
(b) After clicking the question mark, NOVA reveals the data hive and highlights the discrepancies between the user’s belief and the data, motivating the user to investigate the conflict.
A screenshot of the Article Reviewer stage.
(a) Article Panel shows positive and negative articles in two columns.
(b) depicts the annotated content of a selected article.
(c) Notes Panel for documenting insights. Clicking the paragraph in (b) creates a “reference” in the note.
(d) The user and data hives are displayed. Users can click a hexagon to inspect its articles. After the exploration, users can click the “Try another” button on the top-right corner to explore another outlet.
Install the packages
npm i
Start
npm run dev