01 · The thesis
The market lives
beyond the keyword.
A keyword only collects what it can name. We start from how this market actually talks — then let related topics decide the data.
Sentiment out.
Only posts that said the word ever enter the dataset.
the market talks.
Topics and subjects decide the data. Then we read. Then we expand.
Traditional listening starts with a keyword list. Whatever the word cannot name never enters.
A keyword only finds
what it already named.
Keywords, sub-keywords, products, campaigns, events. Channels collect whatever matched. Sentiment splits that same total. Then any sentiment can land on any topic already named — negative on product, positive on competitor. A mention counter cannot see an argument it never collected.
- ListenKeywords, sub-keywords, products, campaigns, events.
Traditional listening starts from a list of words it already knows how to name.
- Collect — the gateOnly posts that matched the list pass.
Channels keep whatever said the word. If the word never appears, the post never enters.
- SentimentPositive, neutral, negative.
A split after the gate has already closed — not a reading of the argument.
- Topics already namedProduct, service, price, competitor.
Any sentiment can land on any of them. The buckets were named in advance — they do not classify the lines 1:1.
- The missing streamTalk without the word never enters.
Unnamed talk is dropped before sentiment or topics exist.
Sub-keywords
Products
Campaigns
Channels keep whatever matched the list. If the word never appears, the post never enters.
Positive, neutral, negative — a split of the same total. Each can land on any topic already named.
Service
Price
Competitor
the dataset never has it.
We start from how this market talks. Context belongs to the category, and it fills as topics return — not a keyword set proposed and confirmed from the start.
A loop, not a list.
Context fills as it turns.
Sentiment out.The list is the gate. Sentiment and topics come after it has already closed.
Understand the market. An agent learns how this category talks — topics, subjects, the current context. That context decides what data belongs.
Collect related. Adjacent conversations enter because they belong to the subject, not because they matched a word.
Read, expand, return. Each post is evidence. New topics go back into understanding. Context fills. The loop runs again.
- 01 · UnderstandMarket, topics, subjects.
How this category talks. That is the current context — not a keyword list locked on day one.
- 02 · CollectRelated topics, adjacent talk.
What belongs to this context now, because the category talks there — not because a word was mapped in advance.
- 03 · ReadVisual · OCR · ASR · sequence.
The whole post as evidence, so nothing is missed because a caption could not say it.
- 04 · ExpandRelated evidence, the next seed.
New topics leave this turn. They do not become another keyword list.
- Topics returnBack to Understand. Context fills.
Each loop adds subjects. The next collect is fuller. Flexible by category — not a set proposed and confirmed from the start.
Reading the whole post is not the product. It is one step of the loop — so nothing is missed because a caption or a keyword could not say it.
Read the whole post,
not just the caption.
ANESSA Mild Milk is the illustration — one post, read in full. Brand and SKU, the on-screen promise, the spoken mechanism, a texture with no text, then the authority close.
OCR · 00:04SMOOTH PROTECT AIR
00:00Brand established
00:01Functional claim
00:08Texture made visible
00:12Authority closeFour evidence frames from one post — brand, functional claim, texture, authority close. Each timestamp is recoverable. ANESSA is the illustration of the Read step, not a category to follow.
The ball gets bigger because topics returned. Understanding grew. That is how the market — and the customer — come into focus.
The snowball is the loop.
Each turn, the context is fuller.
Understand → collect → read → expand → back. Flexible by category — not a set proposed and confirmed from the start.
talks right now
Current subjects, not a keyword list frozen on day one.
whole posts read
Collect what belongs. Read visual, OCR, ASR and sequence so a caption cannot hide it.
The next loop is bigger.
New subjects go back to Understand. The ball is fuller than the turn before.
Not a keyword set locked at the start.
A context that keeps filling.
After the loops, you see which claims are occupied, which are still open, and whether customers confirm or break them. The post was only the illustration.
After it rolls
The category
is now visible.
Occupied, contested, open, and the customer’s own words — not a replay of one creative sequence
What can be done with this?
Decide where to stand: crowded named claims, contested how-it-feels language, or adjacent talk a keyword list never queried.Next decision: occupy an open claim — not another keyword.is crowded
A keyword list would stop here and call it the market.
how it’s used
Wherever creators and customers actually fight — not in a sentiment bucket.
never queried
Adjacent subjects the locked list would never have asked for.
breaks the claim
After use, in their own words — not as positive, neutral, or negative.
Channels enter because they belong to this market — not because a keyword was mapped to them.
The snowball rolls
through the whole field.
Social, publishers, forums, reviews and commerce. A source is in because the category talks there. Every finding stays traceable to its origin.
Kept because the source is still attached.
09 · Next
You bring the category. We start from how that market talks — not from a keyword list.
Bring us one category.
Watch the evidence grow.
See which claims are occupied, which are still open, and which demand a keyword list would never have collected.
Analyze my category ↗Demonstration source: “ANESSA MILD MILK — CHỐNG NẮNG CHUYÊN NGHIỆP CHO DA NHẠY CẢM,” © Anessa Vietnam. Frames are used here to demonstrate the evidence model — visual, OCR, ASR, sequence — not as a category brief.