Principles & Vision
The market has feelings. We measure them.
Why sentiment index matters
Markets have always run on two things: numbers and feelings. For decades, we've had excellent tools for the numbers. The feelings part? Not so much.
That's changing. Hedge funds and quant firms have quietly built sentiment analysis into their models for years — scraping social media data, parsing earnings calls, measuring the emotional temperature of the internet. It works. But those signals stay locked behind institutional paywalls, buried in proprietary systems that cost six figures a year to access.
Meanwhile, the noise problem is getting worse. More platforms, more posts, more opinions, more bots. The raw signal — what people actually think and feel about a company, a sector, an idea — is harder than ever to extract from the flood.
QuantSenti exists to fix both of those problems at once.
Project Overview
QuantSenti started with a simple idea: make quantified market sentiment available to everyone. Whether you are a professional analyst, an independent researcher, or simply curious about how markets feel — this platform is designed for you. No complex setup, no account required. Just open the dashboard and read a single, intuitive 0–100 score.
We believe that when public social data is transformed into transparent, reproducible indicators, it is useful for all market participants.
Our vision is to build an open sentiment analysis platform covering equities, commodities,
or anything else. Clean charts, plain-language labels, and zero configuration make it
simple from the very first visit.
We do not give investment advice, we simply present the data itself.
1 · Data Acquisition
QuantSenti processes only open public discourse. We transform global market signals into insights through a privacy-first lens. It automatically anonymizes all content to ensure no personal identifiers or digital footprints are ever recorded.
2 · BERT-family Inference
Each post is split into individual sentences before scoring. Financial assets are evaluated with FinBERT, a domain-adapted model pre-trained on financial news and analyst reports; general and social topics use models from the RoBERTa family. Every sentence receives three probability scores — positive, negative, and neutral — plus a majority label. The exact model version is stored alongside every record.
3 · Aggregation & Index Calculation
Sentence-level scores are aggregated into a single Sentiment Index over a rolling time window. Three mechanisms ensure the result is stable and meaningful:
Time decay --- recent sentences carry more weight via exponential decay; a post from the last hour outweighs one from yesterday.
Laplace smoothing (α = 5) --- adds five virtual neutral observations, preventing a handful of early posts from producing extreme readings.
Confidence grading --- results backed by fewer than 5 sentences are flagged low; 20 + sentences reach high confidence.
S = Sentiment Index | E = Entity | T = Time window | W = Engagement-based weight | Npos / Nneg = Positive / negative sentence count | α = Laplace smoothing constant (default 5), prevents cold-start volatility. Result is linearly mapped to 0–100.
4 · Live & Backtest Data
QuantSenti offers two data modes: live snapshots updated in real time, and historical backtest aggregates over any custom date range. Both datasets are available for download as CSV.
Tian
Founder of QuantSenti
To observe nature, we need visible light.
To observe humanity, we need emotion.