Back to projects

Data science / Final project

Bluesky Sentiment and Stock Movement

A modeling pipeline that merges Bluesky sentiment with Yahoo Finance hourly data to test short-term stock movement signals.

Analysis surface

Modeling Pipeline

  1. 01Audit

    Merge Bluesky social sentiment data with Yahoo Finance hourly OHLCV data.

  2. 02Feature

    Aggregate to ticker-hours, split by intraday/overnight, and build rolling features.

  3. 03Model

    Train baseline and tree-based ensembles using strict chronological train/test splits.

  4. 04Evaluate

    Compare F1 and ROC-AUC scores to test for actionable, short-term directional signal.

Problem

The project tests whether social sentiment adds anything useful once timestamps, targets, and train/test splits are handled carefully.

Approach

I structured collection, audit, feature engineering, modeling, evaluation, and dashboard work into reusable notebooks and Python modules.

Result

The repo includes EDA figures, processed feature tables, trained models, and result tables.