Analysis types

← Rings · Start · What you get →

Analysis is layers, not one pass. It runs after we keep what is relevant in a Collection run.

First: should we keep this? Then the words. Then pictures and video (faces, on-screen text, places). Then stories across many items. Then the people. Then the write-ups on the dashboards. Then what to collect next.

flowchart LR
  g["1. Gate — keep or hide"] --> t["2. Text — mood + names"]
  t --> m["3. Media — image / video"]
  m --> s["4. Stories — narratives"]
  s --> p["5. People — type + blurb"]
  s --> d["6. Dashboards — write-ups"]
  s --> k["7. Next collect — keywords"]

1. The gate

Pass	What it is	What you see
Relevance filter	First cheap pass: is this about you? Own accounts and entity mentions are kept automatically.	Kept items go on. Hidden items stay stored. Explore can show why.

2. The words — every kept item

Pass	What it is	What you see
Sentiment	Mood of this one item: positive, negative, mixed, or neutral, plus a score.	Insight charts, community mood, narrative mood.
Entities	People, places, organisations, products named in the text.	Search, filters, Mind Map links.
Relevancy score	0–1: how useful is this for understanding the conversation?	Only high enough scores grow the next [[Rings

3. Pictures and video

Full detail: Image and video.

When a post, comment, or reply has a photo or video, we save the file first, then read it. Captions are not enough — people, places, and slogans often live only in the media.

Pass	What it is	What you see
Scene	What the picture or frame shows (product, crowd, logo, protest, meal…).	Search by what is shown, not only the caption.
Face presence	Whether people appear, roughly how many.	Filter to posts that show people.
Face ↔ author	Faces compared to the author’s public profile photo.	Tell “the account owner is in this post” from “this is a photo of someone else”.
Text in the image / on-screen	Words on packaging, signs, memes, overlays, burned-in captions.	Searchable with the post text, including @handles that never appear in the caption.
Place tagging	Platform location tags, plus places we can see or read (landmark, shop front, street sign).	Map and filter by place even when the caption is silent.
Mentions	People and brands named in the caption, on screen, or spoken.	One mention list across text and media.
Safety	Whether the media looks unsafe.	Held for review. Does not delete the post.
Video transcript	What was said. Frames still get the image passes above.	Spoken words, not just the caption.

4. Stories — across many items

Pass	What it is	What you see
Narratives	Recurring storylines: title, description, mood of the story, supporting items.	Public Opinion. Similar stories are merged.
Tags	Categories on items and stories.	Insights tags, filters.
Keywords	Search phrases worth using later.	Fed into the paid-search step.

5. People

Pass	What it is	What you see
Author profile	Type + blurb from their profile and posts. Runs when you open the person.	Community member page. Types are listed on Rings > How Community classifies a person.
Community summary	Health of the whole community: findings, influencers, risks, recommendations.	Community insights.

6. Dashboard write-ups

Pass	What it is	What you see
Sentiment digest	One line on what is driving positive vs negative talk.	Insights → Sentiment.
Chart takeaway	One line under a chart.	Insight charts.
Assistant	Answers in plain language, grounded in collected data.	Assistant. Says so when the data does not support an answer.

7. Planning the next collection

Pass	What it is	What you see
Entity search terms	For org-map people/brands with no social account.	Suggested on the org map; used as keyword targets.
Keyword pick	Which candidate terms are worth paying to search.	Approval panel before paid search. Asks first.

Notes steer several passes

Project notes on Ingestion → Analysis change what “relevant” means. Changing them can mark old results stale. Re-running analysis is an explicit, pausable step because it costs money.

See What you get for where this lands in the product.