For, against or neutral toward the government, and how strongly
Each post is read for its stance toward the government and any other body you name, using the names people write for it in each language.
The answer for each post is for, against or neutral, with an intensity and a mark when it is sarcastic. Across a topic, those answers become the shares that are for, against and neutral.
Three piles
Posts arrive in one stream and are sorted by what they say about the body you track.

01
One post, one target
A post is read once for each body you track, with that body’s names and the other words used for it, such as "the Centre" and its Hindi equivalents. A post on a fuel price announcement can be for the government and neutral toward another body.
02
For, against or neutral
Each reading gives the stance and its intensity, from mild to strong.
03
Sarcasm
Praise that means the opposite is read as such and marked as sarcasm.
04
The shares for a topic
A topic’s shares weigh each post by its reach and take its strongest stance. Where accounts are scored for authenticity, posts from likely inauthentic accounts count for less, and the unweighted shares are kept beside the weighted ones.
What is read
01
Post text
The words of the post itself.
02
Text in images
The words in memes and screenshots, read from the image.
03
Speech
The words spoken in video and audio. A post with a two-minute speech clip contributes what is said in it.
04
Twelve languages
The language is identified first, including Hindi written in Latin letters and mixed Hindi and English. Each part of a post keeps its own reading, so a caption and a spoken clip can differ.
05
Sarcasm
Marked for each body the post is read against.
A second reading for wide-reach posts
Posts that reach 100,000 people or more are read a second time by a language model, within a daily limit. If that second reading fails, the first one stands and the record says so.
In each language your analysts label a set of posts by hand for stance, intensity and sarcasm. A language counts as checked only when its readings have been tested against at least 300 labelled posts held back for the test. Until then its stance is shown as not validated.
External reports, including their stance figures, are released only by an approver.
Where stance shows up
| Where | What it shows |
|---|---|
| Timeline | A strip of for, against and neutral on each topic’s lane, and a plain sentence saying whether the posts are mostly for, mostly against or mixed, with the share |
| Reports | The day’s top trends with their stance, beside their summaries and their rise and fall. |
| Alert rules | Rules on the share against, weighted by reach: for example, a topic with a share against of four in ten and reach above 100,000. |
Limits
Stance needs a label set of 2,000 posts in each language and a model tuned on it. Until then stance is shown as not validated, and Timeline leaves the stance cell empty instead of estimating it.
The first label sets are in Hindi, English and Hindi in Latin letters. Stance in the other languages is read and shown as not validated until its own set exists.
Stance is read toward the bodies you name, with their names and the other words used for them. A body that is not named is not tracked.
Questions
Stance toward whom?
Toward the bodies you name, usually the government, each with the names and other words people use for it in each language.
In which languages?
Posts in all twelve are read. A language is shown as validated only once its readings have passed the test against posts labelled by your analysts.
Where does a person decide?
Your analysts label the posts each language is tested on, and an approver releases any report that leaves your organisation.
Can we be alerted when opinion turns?
Yes. Alert rules can be written on the share of posts that are against, together with reach.
Name the bodies you track and the words people use for them, and see how stance would be read on your topics.
Request a demo/platform/stance · Stance