Platform · Sentiment Analysis

Sentiment analysis for Indian media coverage

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.

Sentiment · last 7 days

71% positive

342 clips · 12% negative · 17% neutral

MonTueWedThuFri
HomeIndiaPlatformSentiment Analysis

Sentiment by source · this month

Economic Times82% pos
YourStory76% pos
Trade blogs54% neutral
Forum threads31% neg
Weighted by article volume

Why not an off-the-shelf model

Neutral news isn't negative — most classifiers get that wrong

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.

  • Brand-attributed tone scored for you, not the whole article's mood
  • Feedback loop every rating you give trains toward a sharper model
  • Feeds AVE sentiment is one of the five factors in your media value
  • Fails safe an uncertain call is surfaced, never silently guessed
  • Private to you your coverage and sentiment data stay confidential to your workspace

How it works

Sentiment built for the way media actually reads

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.

01

Classify

Each article is read in full and scored in its own language — headlines and body, not just keywords.

02

Attribute

Sentiment is tied to your brand specifically, so a negative story about someone else doesn't drag you down.

03

Learn

Your thumbs-up / thumbs-down feedback feeds a flywheel that sharpens the model for your coverage over time.

What you get

Tone you can trust, not a black box

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

Sentiment in Indian languages, not translated guesswork

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.

Translated first, then scored

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.

Tuned for PR, not for social media

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.

Three headlines, three different answers

"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.

Uncertainty is shown, not hidden

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.