By-poll & Campaign

Source: TamilTick Trends, 24 Sep–23 Oct 2026, 1,384 articles / 98 outlets, 3,570 videos / 51 channels · tamiltick.com/trends/topic/bypoll-election#chart

modelled Tone of the headlines (modelled) · 24 Sep–23 Oct · 4,954 headlines scored, 0 not scored (awaiting a model or the positive re-check).

Positive 13.4%Neutral 58.3%Negative 28.3%Net tone −14.9 pts (±2.1)

Modelled by an AI model from headlines, not measured sentiment. 4,954 of 4,954 headlines in this window scored; bars under 30 scored are not drawn. How we count

• Positive
• Negative
• Neutral: the rest
• net tone (positive − negative), with the ±2.1-pt hand-label error band
• not enough scored headlines (<30)
• faded bar = partial day / week

The ±2.1-pt band is the gap between the model and hand labels on 200 headlines sampled across the state; it was not measured for any one leader, party or issue. Only positive calls get the second check; negative and neutral calls do not. Measured effect on 200 hand-labelled headlines (round 2): no positive/negative flips; 6 negative headlines, mostly denials and rebuttals, were read as neutral and 9 neutral ones as negative; net tone (positive − negative) came out within 2.1 points of the hand labels (Tamil Nadu 2.1 points more negative). Per class on those 200 headlines: positive precision 0.89, recall 0.94; negative precision 0.92, recall 0.94; neutral precision 0.90, recall 0.86.