Understand what the market is saying,
then turn it into factors.
Unicode Node is an AI research assistant built for quantitative research teams. It reads public financial news, stock forums, investor communities and social media, understands sentiment, attention and events, and turns them into testable factors for researchers to review.
Not counting keywords. Reading every sentence.
The same phrase can be bullish or bearish depending on context. Unicode Node uses large language models to read context, catch sarcasm, hedging and mixed signals, and highlight the passages that drive each judgment.
[Company A] Q3 orders beat expectations, and capacity is still expanding — any pullback is a chance to buy in
- Entity
- [Company A] → [Ticker]
- Event
- Earnings outlook raised
- Noise check
- Not an ad, normal account
Rationale: orders and capacity are fundamental signals and carry the most weight; the last clause expresses clear buying intent.
One agent, from raw text to tested factors
- 01CollectIncrementally gather public news, forum, community and social posts, indexed by publish time.
- 02CleanDe-duplicate, filter ads and suspected bot accounts, and map company names to tickers.
- 03UnderstandA language model scores sentiment, events and direction for each item and records its rationale.
- 04ConstructCombine signals into interpretable factors aligned to trading days, with no look-ahead.
- 05TestCompute IC, quantile returns and turnover, decorrelate against existing factors, and report for review.
Built for professional research teams
Quant research teams
Add sentiment, attention and event-driven alternative-data factors to multi-factor models.
Asset managers
Systematically track news and discussion related to holdings to support the research process.
Research analysts
Quickly see what the market is discussing about a company or theme, and how sentiment shifts.
AI finds the leads. Researchers make the call.
Public data only
We only collect publicly accessible information, respect each platform’s terms and rate limits, and do not collect personal data.
Human in the loop
Outputs are research candidates, not trading instructions. Your research team decides whether and how to use them.
Traceable judgments
Every factor value can be traced back to the source text and the passages the model highlighted.
Not investment advice
Unicode Node is a research tool. It does not provide investment advice or execute trades on your behalf.
About Unicode Node
Unicode Node is built and operated by a team focused on quantitative research and data engineering. A great deal of valuable market information lives in unstructured text, and large language models make it possible, for the first time, to read it systematically.
Our goal is to let researchers spend their time on judgment, not on collecting and cleaning data.
Frequently asked questions
How is Unicode Node different from sentiment data feeds?
Data feeds provide raw material. Unicode Node delivers processed factor candidates and test results, with the rationale behind every judgment.
Why use a language model instead of a sentiment dictionary?
Dictionaries only count keywords and are easily fooled by sarcasm, negation and slang. Language models read the whole sentence in context.
How do you guard against overfitting?
Unicode Node separates in-sample and out-of-sample data, logs the number of trials per candidate, and favors interpretable factors. Passing screening is a starting point; researchers still validate independently.
Can we bring our own data?
Yes. Besides public sources, you can connect your own text data, which is used only within your workspace.