Read the market's mood before it moves.
FinSentiment is an app that reads global financial market news with transformer-based AI and turns it into clear, explainable sentiment signals — helping investors and small businesses make better-informed decisions.
Company reports quarterly revenue above expectations and raises full-year guidance.
Central bank signals interest rates will stay higher for longer as inflation persists.
Company files its scheduled annual report with the market regulator.
Markets react to news in seconds. No person can read it all — and simple keyword tools miss what the words really mean.
Too much news
Earnings calls, filings and economic briefings arrive constantly, across every market.
Keyword tools miss context
Lexicon-based methods count words; they can't tell "beat expectations" from "missed expectations" in context.
Small players lose out
Large institutions have analyst teams. Individual investors and small businesses usually don't.
From headline to signal — sentence by sentence.
- 01 / COLLECT
Gather market news
Financial news from global stock markets flows into the app.
Global news - 02 / UNDERSTAND
Transformer reading
Finance-tuned language models read each sentence in context, using domain-adaptive embeddings.
FinBERTRoBERTa-fin - 03 / CONTEXT
Events & time
Financial event flags and temporal sentiment tracking capture what kind of news it is and how mood shifts.
Event flagsTemporal tracking - 04 / EXPLAIN
Signal you can trust
A positive, neutral or negative signal — with the words that drove it highlighted.
SHAPAttention maps
Tested across the news that moves markets.
The model was evaluated across multiple classes of financial events.
Earnings calls
What management says about results and outlook — and how confident it sounds.
Regulatory disclosures
Official filings and announcements that companies are required to publish.
Macroeconomic briefings
Economic news — interest rates, inflation, growth — that moves whole markets.
High precision, real-time speed.
Sentiment classification accuracy, consistently above this level.
Balanced precision and recall across sentiment classes.
Peak correlation between detected sentiment and price movement.
Average training time per epoch — fast enough for real-time use.
Better than traditional methods
Compared with lexicon-based and conventional machine learning models.
Bar shows the improvement range on a 0–10 percentage-point scale.
Compared against
Plus significant gains in interpretability through SHAP value tracing and attention-based visual diagnostics.
Source: Almarzoog et al., IEEE ICCR 2025. doi:10.1109/ICCR67387.2025.11291723
The company cited strong demand and record margins, though supply delays may ease next quarter.
Never a black box.
Every signal shows its reasoning, so users can see exactly which words drove the sentiment.
- SHAP value tracingMeasures how much each word or feature contributed to the final sentiment.
- Attention visualsShows where the transformer "looked" in the text when it decided.
- Fine-grained, sentence-levelInterprets each sentence on its own, even under market volatility.
One architecture. Any market.
The framework scales without architectural reconfiguration.
From local exchanges to global markets.
Stocks today, other assets with the same engine.
Designed to extend to news in other languages.
Market intelligence for everyone, not just Wall Street.
Individual investors
Understand the mood behind the headlines before making a move.
Small businesses
Track market and economic news that affects costs, demand and financing.
Analysts & advisors
Screen large volumes of news quickly and focus on what matters.
Finance expertise meets language AI.
FinSentiment was founded in 2025 by a research team that combines knowledge of financial markets with computer science and natural language processing.
Administrative & Financial Sciences
Defines which financial events matter, how sentiment relates to prices, and what investors need.
Computer Science
Builds the transformer models, the explainability layer and the app itself.
Frequently asked questions
What does FinSentiment do?
It analyses financial market news with AI and classifies its sentiment as positive, neutral or negative, with an explanation of which words drove each result.
Which AI models does it use?
Transformer-based financial language models — FinBERT and RoBERTa-fin — trained on annotated financial text, with event flags and temporal sentiment tracking.
How accurate is it?
In research testing, sentiment classification accuracy was consistently above 96.8%, with an F1-score of 0.94 — 5.0–5.9% more accurate than lexicon-based and traditional machine learning models.
Does sentiment really relate to prices?
The research found a peak correlation of 0.71 between detected sentiment and price movement. Correlation is a useful signal, but it is not a guarantee of future prices.
Is this investment advice?
No. FinSentiment provides information to support decisions. It is not financial advice, and investment decisions remain the user's own responsibility.
Backed by peer-reviewed IEEE research.
Read the study behind FinSentiment's transformer-based financial sentiment engine.
Almarzoog, K. H. A., Furaijl, H. B., Kadhum, N., Srayyih, F. H., Raheem, M. J., Faisal, F. G., et al. (2025). Advanced financial sentiment analysis of global stock market news using transformer-based deep language understanding models. In 2025 3rd International Conference on Cyber Resilience (ICCR) (pp. 1–8). IEEE.
https://doi.org/10.1109/ICCR67387.2025.11291723