Inspiration
Investors are flooded with unstructured information, earnings reports, financial news, analyst commentary, and public sentiment. While numbers (EPS, revenue, margins) are easy to quantify, the narratives that drive markets are harder to measure.
We were inspired by the idea of transforming qualitative market signals into actionable numbers. With the rise of Large Language Models (LLMs), we saw the opportunity to bridge this gap and give investors an edge.
What it does
Our project converts qualitative data into quantitative investment insights.
- We start each week with trending stocks from quarterly earnings reports.
- We collect and analyze public sentiment from major media outlets.
- Using LLMs, we distill this data into structured signals.
- We generate a final Investment Score for each stock:
This produces real-time, research-backed, easy-to-use insights that keep investors ahead of the curve.
How we built it
- Data Pipeline: Collected quarterly earnings data and scraped major news/media sources.
- Natural Language Processing: Used LLMs to extract tone, sentiment, and recurring themes.
- Quantification: Mapped sentiment to a normalized score
- Dashboard: Built a simple interface to track and visualize weekly scores.
Challenges we ran into
- Data alignment: Matching sentiment timelines with earnings release dates.
- Noise in news: Filtering out repetitive, irrelevant, or biased content.
- Weight balancing: Making sure sentiment, earnings, and coverage had the right influence.
- Model consistency: Keeping LLM sentiment evaluations stable across sources.
Accomplishments that we're proud of
- Built a system that successfully converts narrative-driven signals into numbers.
- Created a weekly investment scoring model that integrates qualitative & quantitative data.
- Designed a user-friendly dashboard for investors to make faster decisions.
What we learned
- Qualitative data (like news and sentiment) can be just as important as quantitative data.
- LLMs are powerful for synthesis but require careful prompting and validation.
- Data preprocessing and alignment can make or break the accuracy of an investment model.
- Backtesting is critical to refining scoring weights.
What's next for Stock News Sentiment
- Deeper data sources: Expanding beyond news into earnings call transcripts, Reddit, and Twitter.
- Backtesting engine: More rigorous historical validation over decades of stock data.
- Personalization: Allowing users to customize their scoring.
- Deployment: Launching a live investor tool that updates automatically each week.
Log in or sign up for Devpost to join the conversation.