Platform · Sentiment Analysis
Every article is scored positive, neutral, or negative by an AI model tuned for business and media language — not a generic off-the-shelf classifier. See tone per article, per source, and over time, in any language.
71% positive
342 clips · 12% negative · 17% neutral
Sentiment by source · this month
Why not an off-the-shelf model
Straightforward reporting (“the company announced results”) is neutral, not bad news. Off-the-shelf models routinely mislabel it. InMedia's prompt is built to read media the way a communications team does.
How it works
Generic sentiment models mistake neutral reporting for negativity and miss the difference between a mention and an endorsement. Ours is tuned for business, PR, and news copy — and it keeps learning from your feedback.
Classify
Each article is read in full and scored in its own language — headlines and body, not just keywords.
Attribute
Sentiment is tied to your brand specifically, so a negative story about someone else doesn't drag you down.
Learn
Your thumbs-up / thumbs-down feedback feeds a flywheel that sharpens the model for your coverage over time.
What you get
Per article
Every clip classified positive, neutral, or negative
Multilingual
Scored natively in the article's own language
Over time
Watch mood shift across days, weeks, and campaigns
Risk-aware
Negative coverage flagged separately as reputational risk
About a third of the Indian coverage we collect is not in English — 60,293 Hindi articles, 36,263 Malayalam, 33,224 Telugu and 21,076 Tamil in a single 30-day window. Sentiment has to work on those, or it scores a slice of your coverage and reports it as the whole.
Every non-English article is translated before it is scored, so a Marathi business story and an English one are judged on the same scale rather than by two different models with two different ideas of what negative means. The original text is kept, so you can always read what was actually published rather than only our version of it.
Generic sentiment tools were trained on product reviews and tweets, and they misread Indian business reporting badly. A factual results story that mentions a decline is not negative coverage. A politely worded column arguing you got it wrong is not positive coverage because it quoted you courteously. Our scoring is tuned for earned media, where the question is what a reader takes away about the brand — not how cheerful the sentence was.
"Company posts 12% revenue growth, misses street estimates" is factual reporting with a negative edge for the brand, not a neutral data point. "Firm denies regulatory lapse" is negative regardless of how measured the denial was, because the reader takes away the allegation. "Executive named among India's top CX leaders" is positive even though nothing about the business changed. Tools trained on consumer reviews get all three wrong in the same predictable direction — they read tone and miss consequence.
Where the model is not confident, the mention is flagged as low-confidence instead of being given a definitive label. You can override any score, and the correction is kept — so the reporting reflects your read of your own coverage. This matters most on the coverage that matters most: the ambiguous, mixed, deniable piece that a dashboard will otherwise quietly file as neutral and drop out of your crisis view.