Kogann/stockmatch-synthetic

StockMatch Synthetic Dataset & EDA Overview

Dataset Overview

This dataset contains 12,500 synthetic stock profiles designed for building an AI-powered stock recommendation system. It combines key numerical financial metrics with LLM-generated business summaries (Qwen2.5).

Exploratory Data Analysis (EDA) & Cleaning Summary

During the EDA process, we systematically analyzed and refined the dataset based on the following findings:

1. 1.Irrelevant Columns Removal: We removed non-essential identifiers such as internal row index (stock_id) and full company names (company_name), keeping ticker as the primary distinct identifier.
2. 2.Missing Values Inspection: We checked for missing values across all features and found 0 missing values. This aligns with market expectations, as fundamental stock market data is fully public and universally available.
3. 3.Outliers & Extremes: We analyzed numerical distributions for anomalies. While natural market extremes were present (e.g., high-growth tech stocks vs. low-beta utilities), no completely impossible or corrupted values were found.
4. 4.Duplicate Inspection: We inspected duplicate records. While minor overlaps appeared in auto-generated company names, every row possessed a unique ticker symbol, ensuring distinct stock entities.

Exploratory Data Visualizations

Clustering

Embedding Model Comparison

Included Files

• —data_science_FP_updated.ipynb: Full notebook containing Data Generation, EDA, Visualizations, Embedding Benchmark, FAISS Indexing, and Generation Pipeline.