Understanding the "candidate brand" through narrative mapping
(Maps viewable on desktop only).
Through LLM-based narrative analytics, we’ve mapped 12,000 tweets mentioning either Joe Biden or Donald Trump in 2021, as a demonstration of how a candidate's narrative footprint on social media can be measured. By hovering your cursor over the dots below, you can browse the content within an easy-to-understand narrative map.
The initial map: At 3 different levels of detail. The 5 cluster-level shows the axes. Now go to the detailed 10-cluster level, then the 33-cluster level - by moving the slider on the lower right of of the map. This reveals further detail
Is the data "clean"? The impact of one prolific creator on sentiment
Trolls and bots are a reality. Analysts should make an informed call on what data to include - or not - for sentiment measurement. This illustrates the impact of a single creator "creator B" on this dataset. In most sentiment studies, it is not easy to get a line of sight on the data upon which the sentiment is based. So, who knows what the score really means?
Narrative analytics makes it easy. First, the dots on the lower right are separated by a white "gap" which suggests that - narratively speaking - they they are dissimilar to the rest of the conversation on twitter. Below, we can see that "creator B" posted a large majority of all posts on the lower right, as shown in green below - which means that he is essentially 'yelling alone' - e.g. not attracting others into this narrative space. Analysts in this case may choose to delete this data, then run the clustering again - as we have for purposes of this demonstration.
Move slider to adjust level of detail
The "cleaned" map - revealing the narratives without the noise
Axes of the map: Analysis vs Judgement; The political fight vs the personalities & motives. The 5-cluster level shows the general “shape” of the content. The center cluster is important – revealing that the essential battle for credibility and relevance. None of the clusters or axes are pre-set – they are unique to this dataset and empirically formed, factoring trillions of possible cluster/axis combinations. The is a key feature of narrative analytics: the ability to treat large samples of unstructured language - data that would normally be interpreted subjectively - and understand it empirically.
Now go to the detailed 29-cluster level - by moving the slider on the lower right of of the map, and read on.
Comparing/contrasting narrative footprint: Biden vs Trump
Taking the most-mentioned narratives from the 29 cluster level of the above map, we can compare the number of tweets mentioning Biden, Trump, or both. Hover your cursor over the bars in the graph to learn more. This is an "outside in" way to measure the candidate brand, based on categories that are set dynamically by the people doing the talking (and the voting) rather than measuring rigid, pre-selected attributes which may leave blindspots.
Narrative traction - what narratives earn more retweets (versus fewer)
Narrative traction is an essential measurement - a way to understand how 'sticky" one narrative is versus another. Measured via weighted retweets (to normalise for the impact of some creators having more followers than others), this reveals that the narrative with the highest traction - by far - is "the political fight" - the top part of the map within the 5-cluster view. By contrast, analytical tweets earn far fewer re-tweets. Note: you can not measure this with existing research or via social listening approaches.