AION-Sentiment-IN-v3

AION-Sentiment-IN-v3 is an open-source Indian financial news sentiment model with taxonomy-driven market logic and sector-aware signal integration.

What's New (April 2026)

• Full Market Intelligence Layer - Now includes 4-factor event impact engine
• Dynamic Sector Sensitivities - 90-day rolling correlations for 14 NIFTY sectors
• Instrument Classifier - Supports equities, derivatives, commodities, bonds
• Event Impact Rankings - For RBI policy, crude spikes, rupee moves, and VIX shocks

Quick Start

from aion_taxonomy.event_impact_engine import rank_instruments_by_impact

instruments = [
    {"tradingsymbol": "RELIANCE", "segment": "EQ", "exchange": "NSE"},
    {"tradingsymbol": "HDFCBANK", "segment": "EQ", "exchange": "NSE"},
]

ranked = rank_instruments_by_impact("RBI_RATE_HIKE", instruments)
print(ranked)

Model Details

Field	Value
Developer	AION Analytics
Architecture	6-layer transformer encoder, 768 hidden, 12 attention heads
Parameters	~67M
Model Size	268 MB
License	Apache License 2.0

AION Platform Components

Component	Purpose
Sentiment Model	Headline sentiment classification (this model)
Taxonomy Engine	Event detection (136 India-specific events)
Market Intelligence	4-factor event impact engine (NEW)
Sector Map	NSE ticker to sector mapping
VIX Weighting	Confidence adjustment based on India VIX

Benchmark

Model	Accuracy	F1 Score
AION-Sentiment-IN-v3	86.67%	0.865
FinBERT (ProsusAI)	86.67%	0.865
RoBERTa (cardiffnlp)	73.33%	0.725

Citation

@software{aion_sentiment_in_v3,
  author = {AION Analytics},
  title = {AION-Sentiment-IN-v3: Indian Financial News Sentiment Model},
  year = {2026},
  url = {https://github.com/AION-Analytics/aion-sentiment}
}