- Upload a post-level file (CSV) and select a delimiter
- Identify the header of the column containing text to classify
- Adding timestamps/authors will extend the dashboard
Which general emotions were detected the most?
This horizontal stacked bar chart shows the general moods that were detected at the most basic level. The results show the accumulation of classified emotion probabilities that were highest for each post. These ‘core emotions’ are Joy, Empowered, Excitement, Fear, Anger and Sadness. The corresponding colors are consistent throughout the rest of the dashboard. The ordering of emotions in this band are from most positive (left) to negative (right). Note: Excited and Empowered are included in our updated typology based on research showing their dominance within millions of online conversations (Zimmerman, et al 2015).
Show the average mini-bar chart?
The small horizontal stacked bar chart shows the average moods detected across all datasets analysed by emotionVis.
Which specific feelings were detected? How did they lead to core emotion classifications?
This detailed bar chart shows the emotional accumulation of 27 more specific feelings that can be detected. Rather than displaying a percentage of probabilities total classifications, here the sum of all granular levels detected from posts are revealed for each discrete feeling. Colors correspond to core emotion group that these feelings contributed to in the chart above it. This allows you to compare aggregate overviews of emotions with the accumulation over time in graphs below. Note: if a core emotion does not appear above this cart, then the fine grain feelings were not dominant enough to win core emotion classifications of posts.
When were levels of arousal and sentiment (valence) the highest?
This dual-axis chart shows conversation volumes over time. The post volume is represented by gray bars in the background so you can see when people were talking the most. This is overlaid with two lines that represent arousal (blue) and sentiment (green) levels. These layers combined give an overview of heartbeats in discussion; when the conversation is high/low in volume, aroused/calm, positive/negative. A time slider below lets you manipulate the chart and drill-down to specific windows in time for a more careful look at these ‘emotional trajectories’ or specific events.
When were each of the emotions detected?
This stacked area chart shows the breakdown of a conversation (by core emotions) over time and how quickly these reverberations happened in your dataset. The color encoding and vertical order of emotions is the same as previous bar charts. The distribution shows shifts in dominance of core emotions, as well your emotional resilience over time to negative periods. You can pinpoint values for all core emotions by mousing-over the chart. As with other charts, the labels are also buttons that can be used to remove individual emotions and isolate a stream of interest (i.e. click on Joy to remove joy from the stream). It is also possible to change chart type from the expanded view and instead show a stream graph, for example.
What was said? What is the overall emotional footprint of the conversation?
This two-dimensional scatterplot lets you see all posts in your dataset as they are mapped by emotion dimensions. Mouse-over any post to see who said what, and what values of sentiment (from -1 to 1) and arousal (from 0 to 1) were detected. Posts have been colored by their winning emotional classification. The size of bubbles corresponds to how decisive the classification was (strength of probability score). You can also filter for each emotion by clicking on their label at the top right, and refer to distributions in the strip plots along each axis.
Who has been expressing the most emotion?
This series of bar charts ranks people in the conversation who express the highest levels of each core emotion, on average. These small multiples serve as ways to find people who may not post the most, or have the highest reach, but may instead have extreme levels of a given emotion. If you find people who are very joyous or extremely angry, you can then engage/defend these loyalists/activists who are flagged as having particular emotions in your conversation.
Why were certain emotions expressed?
This treemap displays term frequencies and allows you to see which terms are most associated with which emotion. As emotions circulate in your conversation, particular feelings can get ‘stuck’ to certain bodies, spaces, situations, etc. Negative emotions may be attached to a specific topic or person, for example. These keywords offer clues suggesting reasons why emotions were expressed from the crowd at large.
You can export this dataset through this link.