Glossary
Narrative Clustering
The 3–8 real debates in a comment section — made visible.
Definition
Narrative clustering groups comments by the underlying argument or theme they represent, rather than by keyword or sentiment. The result is a map of the distinct debates actually happening in a comment section — typically 3 to 8 canonical narratives that account for the majority of engagement.
How it works
- 1
Each comment is first classified for themes and stance by the AI.
- 2
All theme labels are collected and deduplicated across the comment set.
- 3
A capable language model groups related themes into 5–8 canonical clusters.
- 4
Comments are mapped to clusters, and each cluster is scored by total engagement weight.
- 5
A one-sentence summary is generated per cluster to label the narrative.
Why it matters
A post with 2,000 comments does not contain 2,000 distinct opinions. It contains a small number of recurring arguments, each held by different proportions of the audience. Narrative clustering collapses the noise into a clear structure, letting analysts see which arguments are driving the conversation rather than reading individual comments.
Related distinctions
Narrative clustering vs topic modelling
Topic modelling (e.g. LDA) identifies statistical co-occurrence of words across documents. Narrative clustering identifies the argumentative positions being expressed. The output of narrative clustering is a set of named debates, not word clouds.
Narrative clustering vs sentiment analysis
Sentiment analysis produces a single polarity score. Narrative clustering produces a structured map of the arguments in play — which is substantially more useful for understanding why an audience holds a particular view.
Frequently asked questions
What is narrative clustering in social media analysis?
Narrative clustering is a technique that groups social media comments by the underlying argument they express. The output is a set of canonical narratives — typically 3 to 8 — that represent the distinct debates happening in a comment section.
How many narratives does a typical comment section contain?
Most comment sections — even large ones with thousands of comments — contain between 3 and 8 dominant narratives that account for the majority of engagement. The remainder are noise, off-topic, or minor variations on the main themes.
How is narrative clustering different from topic modelling?
Topic modelling identifies statistical word co-occurrence patterns. Narrative clustering identifies argumentative positions. The practical difference is that narrative clustering produces named, readable debates rather than word lists.
Can narrative clustering work on small comment sets?
Yes. Narrative clustering is useful from around 20–30 comments upwards. On small sets, the AI produces fewer clusters and the distinctions between them may be finer-grained.
See narrative clustering in practice
Narativ applies stance analysis, narrative clustering, and engagement weighting to live comment sections — from £1 per post.