Star Home: Data Analysis of Myanmar Scam-Related Missing Persons

Translation note: This English version follows the structure, data, figures, and references of the corresponding Chinese post. It was reviewed and synchronized on 2026-08-03.

Cover illustration for the Star Home analysis

Summary: This article uses a recent public mutual-aid spreadsheet to structurally analyze and visualize missing-person cases related to Myanmar scams.

Summary: A concise profile of victims and regional distribution based on a public mutual-aid archive.

Author: Cao Yueyang
Affiliation: Department of Data and Systems Engineering, The University of Hong Kong
Role: Independent analyst (data collection, labeling, visualization, reporting)
Contact: Please use the public contact links on this site.
Time: January 2025

1. Goals and Conclusions

Using a public “Star Home” mutual-aid archive, I extracted structured information and quickly summarized the reasons, time distribution, geography, and demographic profile of the missing persons. The project also tested the efficiency and cost advantages of LLM-assisted labeling for small-scale social issues.

Data source and compliance

Source: a public Excel sheet collected on 2025-01-14
Processing: aggregation and anonymized presentation only; no personally identifiable information was included
Usage note: for public-interest and research discussion only

Key findings

Profile: 95% male; 80% between 18 and 35
Geography: 613 records from Yunnan, with 72.53% concentrated in Xishuangbanna
Cause: 88.49% were lured away by “high salary” promises

2. Project Overview

On 2025-01-14, after the widely discussed “Wang Xing was tricked into Thailand” incident, a public online document called the “Star Home Mutual-Aid Archive” circulated online. I downloaded it and carried out an independent exploratory analysis.

Analysis method

Data cleaning (Python)
- Read the Excel file with pandas
- Removed redundant columns and normalized dates to a YY.MM format
- Exported the result into a structured TXT file for copy-paste into the LLM
LLM-assisted labeling (Gemini Web)
- Pasted the TXT data into Gemini Pro 2.0 in batches
- Prompted the model to return fields such as cause, time, and region
- Iterated until the output was CSV-friendly
Statistics and visualization (Excel + Tableau)
- Used Excel for counting and filtering
- Used Tableau for maps, bar charts, and pie charts

Method summary

Strength: flexible, low-cost, and fast for small public-interest datasets
Limitation: some ambiguity and occasional hallucinations, so human review is still needed

3. Full Workflow

Step 1: Local data cleaning

I first exported the Excel sheet into a TXT file with commas so it could be pasted into the LLM more easily.

import pandas as pd

def excel_to_txt_with_commas(excel_file, txt_file):
    """
    Export every cell in an Excel workbook into a comma-separated TXT file.
    """
    try:
        df = pd.read_excel(excel_file)
    except FileNotFoundError:
        print(f"Error: Excel file '{excel_file}' was not found.")
        return

    # Remove columns whose names start with Unnamed.
    df = df.loc[:, ~df.columns.str.startswith('Unnamed')]

    with open(txt_file, 'w', encoding='utf-8') as f:
        # Wrap each header in corner brackets and separate headers with commas.
        header = ",".join([f"【{col}】" for col in df.columns])
        f.write(header + "\n")

        # Write each row as a comma-separated line.
        for _, row in df.iterrows():
            row_values = [str(value) for value in row.values]
            f.write(",".join(row_values) + "\n")

    print(f"Excel file '{excel_file}' was exported to TXT successfully.")

excel_file_path = 'initial-cleaned-data.xlsx'
txt_file_path = 'initial-cleaned-data.txt'

excel_to_txt_with_commas(excel_file_path, txt_file_path)

Step 2: LLM labeling

Because the dataset was small, I did not need an API. I pasted the structured text into Google AI Studio’s Gemini Pro 2.0 and refined the prompt repeatedly until the output could be exported as CSV.

A representative prompt first asked the model to classify the reasons for being deceived and return a CSV-friendly text file with fields such as disappearance number, disappearance date, and reason. I then asked it to make the reasons more abstract while retaining useful details, and finally requested a complete row-by-row output.