🚨 Introducing: Narrative Threat Score

Narrative Threat Score allows users to prioritize and act swiftly by providing a comprehensive threat assessment based on several key risk factors.

Key Features

• Customizable Risk Factors: Our score takes into account factors like sentiment, growth rate, and volume threshold. Users can customize them through account settings.

  • Sentiment Score: Aggregates mention scores at the narrative level.

  • Growth Rate Score: Measures period-over-period changes.

  • Volume Threshold Score: Assesses the total volume of narratives.

• Narrative Feed Enhancements:

  • Sort By Threat Score: Easily sort narratives from most to least threatening.
  • Filter Options: Focus solely on threatening narratives.
  • Score Visibility: Display the threat score directly on narrative cards.

• Customizable Weights:

  • Customized Weight: Users can customize the weighting for each of the scores at both the account level (default) and at the Workspace level.

Out of Scope

For now, we won't track threat score changes over time or display arrows to indicate score changes. The focus is on assessing current threats and enabling prompt action.

How are Threat Scores calculated?

The default Threat Score is composed of four main components:

1. Narrative Size (33.3% of the score)
2. Negative Sentiment (33.3% of the score)
3. Trend Score (16.7% of the score)
4. Virality Score (16.7% of the score)

1. Narrative Size

This measures how large the narrative is compared to others in your workspace. Larger narratives generally have a better chance of influencing perceptions.

2. Negative Sentiment

This component evaluates how negative the sentiment is within the narrative. It considers both the percentage of negative mentions and the probability of negative sentiment.

3. Trend Score

This measures how quickly the narrative is growing compared to its past growth. It helps identify narratives that are gaining traction unusually fast.

Calculation:

1. Calculate the gradient (rate of growth) over the past 5 days.
2. Calculate the gradient over the past 30 days.
3. Divide the 5-day gradient by the 30-day gradient.
4. Map this ratio to a 0-1 range using a sigmoid function, where 0.5 represents equal short-term and long-term growth.

Key points:

• A score above 0.5 indicates faster recent growth compared to the long-term trend.
• A score below 0.5 indicates slower recent growth compared to the long-term trend.
• New narratives (less than 5 days old) will have a default score of 0.5 until enough data is available.

4. Virality Score

This captures day-to-day changes in narrative volume, helping to identify sudden spikes in activity.

Breakdown of the virality score calculation:

1. Daily Change Calculation: The virality score starts by calculating the percentage change in mention volume from day to day.

2. Smoothing Function: To moderate the sensitivity to daily swings, an exponential smoothing function is applied over a 5-day window.

3. Smoothing Details:

   • This smoothing approach means that the most recent day's data has the most significant impact on the score.
   • However, about 30% of the calculated value still comes from the previous 4 days' raw values.

4. Range Transformation:

   • The smoothed values theoretically range from -1 (100% decrease) to positive infinity.
   • These values are then mapped to a 0-1 range

5. Midpoint:

   • A 0% change between time periods will result in a virality score of 0.5.
   • Increases in volume will result in scores above 0.5.
   • Decreases in volume will result in scores below 0.5.

Key points to note:

• The score is designed to capture short-term, day-to-day changes in narrative volume.
• It's more sensitive to recent changes due to the exponential smoothing, but still considers the past few days.
• A score of 0.5 represents no change in volume.
• Scores above 0.5 indicate growing volume, with higher scores representing faster growth.
• Scores below 0.5 indicate shrinking volume, with lower scores representing faster decline.

This approach allows the virality score to capture sudden spikes or drops in activity while smoothing out some of the noise that could come from looking at single-day changes in isolation.

Interpreting the Threat Score

• Scores closer to 10 indicate higher potential threat or impact
• Scores closer to 0 indicate lower potential threat or impact
• Compare scores between narratives to prioritize your attention
• Look at changes in a narrative's score over time to spot emerging issues

Example: How Components Affect the Overall Score

• A large narrative with mostly neutral sentiment might have a moderate Threat Score
• A small narrative with highly negative sentiment could have a surprisingly high Threat Score
• A narrative showing sudden growth or virality might see a quick increase in its Threat Score

Important Notes

• The Threat Score is a tool to help prioritize, not a definitive measure of risk
• Always consider the score in context with your own expertise and knowledge
• Scores may fluctuate as narratives evolve over time

Customization

While not currently available to users, future versions may allow customization of:

• Time windows for trend and virality calculations

We're continuously working to improve the Threat Score system. Customer feedback is valuable in helping us refine the tool.