Chilean YouTube News Comment Analyzer

Research tool for analyzing bot activity in Chilean political YouTube news comments. Scrapes comments without the YouTube API, performs word frequency analysis, detects suspected bots using multi-signal heuristics, and exports results to CSV.

Setup

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

On first run the NLTK Spanish stopwords corpus is downloaded automatically.

Usage

1. Add YouTube URLs to links.txt (one per line, lines starting with # are ignored).

2. Run the analyzer:

python main.py

CLI options

Flag	Default	Description
-i, --input	links.txt	Path to file with YouTube URLs
-o, --output	output/	Directory for CSV output files
-t, --threshold	0.6	Bot score threshold (0.0 - 1.0)
-l, --limit	no limit	Max comments to fetch per video

Example with options:

python main.py -i links.txt -o output -t 0.5 -l 500

Output

Three CSV files are generated in the output directory:

word_frequency.csv

Column	Description
rank	Position by frequency (1 = most common)
word	The word (lowercased, accents stripped)
count	Number of occurrences across all videos

Spanish stopwords and YouTube noise words are filtered out.

bot_analysis.csv

One row per comment across all videos.

Column	Description
video_url	Source video
author	Comment author display name
channel_id	Author's YouTube channel ID
comment_text	Full comment text
votes	Like count
time	Timestamp
bot_score	Heuristic score 0.0 - 1.0
bot_category	astroturfing, attack_bot, propaganda, spam, mixed, or empty
flags	Pipe-separated heuristic tags that triggered
is_suspected_bot	True if bot_score >= threshold

video_summary.csv

One row per video plus a TOTAL row.

Column	Description
video_url	Video URL
total_comments	Comments scraped
suspected_bots	Count above threshold
bot_percentage	suspected_bots / total_comments * 100
astroturfing_count	Positive astroturfing bots
attack_bot_count	Negative / attack bots
propaganda_count	Copy-paste propaganda bots
spam_count	Generic spam bots
mixed_count	Bots matching multiple categories

Bot detection heuristics

Designed for Chilean political content. Detects five bot archetypes:

• Astroturfing -- generic praise without substance ("buen video", "tiene toda la razon")
• Attack bots -- political insults as entire comments, ALL CAPS, repetitive spam
• Propaganda -- identical text posted by different authors, unusually formal tone
• Spam -- promotional URLs, auto-generated usernames
• Mixed -- triggers signals from multiple categories equally

Signals are grouped into: username patterns, positive astroturfing, negative/attack, propaganda, and cross-video behaviour. Each signal adds to a cumulative score capped at 1.0. See src/bot_detector.py for the full scoring logic and curated word lists.

Project structure

ChileYoutubeNewsAnalyzer/
├── main.py                   # CLI entry point
├── requirements.txt          # Python dependencies
├── links.txt                 # Input: YouTube URLs
├── src/
│   ├── __init__.py
│   ├── scraper.py            # Comment scraping
│   ├── word_analysis.py      # Word frequency analysis
│   ├── bot_detector.py       # Bot detection heuristics
│   └── exporter.py           # CSV export
└── output/                   # Generated CSV files