I will be posting a postmortem report on this project here in early May. (I'm delaying because of ongoing discussions and continuing trickle of improvements to the 68 targeted sections and articles. I want to be accurate about the magnitude of (human!) improvements for readers spurred by the project.) Here are the Village Pump discussion, the original fifth grade reading level summaries, the subsequently recommended ninth grade level summaries, and the eighth grade and seventh grade reading level summaries with sources. Cramulator (talk) 01:24, 28 April 2025 (UTC)
These files are made available under the Creative Commons CC0 1.0 Universal Public Domain Dedication.
The person who associated a work with this deed has dedicated the work to the public domain by waiving all of his or her rights to the work worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law. You can copy, modify, distribute and perform the work, even for commercial purposes, all without asking permission.
popularTechnical.txt
|
|---|
# popularTechnical.txt copied from: # https://pageviews.wmcloud.org/massviews/?platform=all-access&agent=user&source=category&range=latest-20&subjectpage=0&subcategories=1&sort=views&direction=1&view=list&target=https://en.wikipedia.org/wiki/Category:Wikipedia%20articles%20that%20are%20too%20technical # Category:Wikipedia articles that are too technical 1 Malaysia 176,115 8,386 / day 2 DeepSeek 159,276 7,585 / day 3 Edward Snowden 121,710 5,796 / day 4 Wayback Machine 96,298 4,586 / day 5 List of TCP and UDP port numbers 79,959 3,808 / day 6 Occam's razor 73,617 3,506 / day 7 Dionysus 69,671 3,318 / day 8 Multiple myeloma 62,123 2,958 / day 9 Existentialism 50,531 2,406 / day 10 The Jackson 5 48,319 2,301 / day 11 Psychosis 47,105 2,243 / day 12 5.56×45mm NATO 43,267 2,060 / day 13 World Trade Organization 40,079 1,909 / day 14 DisplayPort 36,291 1,728 / day 15 Eigenvalues and eigenvectors 34,156 1,626 / day 16 Avicenna 32,268 1,537 / day 17 Rorschach test 31,424 1,496 / day 18 Android 16 31,202 1,486 / day 19 BlackBerry Limited 27,678 1,318 / day 20 Astigmatism 27,440 1,307 / day 21 CAN bus 27,233 1,297 / day 22 Madonna–whore complex 26,262 1,251 / day 23 Coulomb's law 26,151 1,245 / day 24 Ø 25,367 1,208 / day 25 Wi-Fi 6 25,007 1,191 / day 26 Sports betting 24,857 1,184 / day 27 Pituitary gland 24,482 1,166 / day 28 Confidence interval 24,239 1,154 / day 29 Waste management 23,647 1,126 / day 30 Cotton 23,530 1,120 / day 31 Synthetic-aperture radar 23,519 1,120 / day 32 Diesel engine 23,182 1,104 / day 33 Elasticsearch 22,714 1,082 / day 34 Theropoda 21,941 1,045 / day 35 Brownian motion 21,885 1,042 / day 36 Carpal tunnel syndrome 19,926 949 / day 37 Psychoanalysis 18,922 901 / day 38 Tuned mass damper 18,539 883 / day 39 Nominal Pipe Size 18,484 880 / day 40 Finite-state machine 18,421 877 / day 41 Fatal insomnia 18,015 858 / day 42 Jacques Derrida 17,849 850 / day 43 Miami-Dade County, Florida 17,548 836 / day 44 Time crystal 17,515 834 / day 45 Microphone 17,396 828 / day 46 Carnian pluvial episode 17,367 827 / day 47 Long short-term memory 17,293 823 / day 48 1. FSV Mainz 05 17,128 816 / day 49 OLED 17,067 813 / day 50 Fire extinguisher 16,807 800 / day 51 Fossil 16,774 799 / day 52 Spinal cord 16,656 793 / day 53 Hypovolemic shock 16,534 787 / day 54 Magnet URI scheme 16,466 784 / day 55 Code-switching 16,283 775 / day 56 Minimum wage 16,080 766 / day 57 USB4 15,910 758 / day 58 Invertible matrix 15,557 741 / day 59 Liquid breathing 15,168 722 / day 60 MIDI 14,917 710 / day 61 Thermal conductivity and resistivity 14,792 704 / day 62 OBD-II PIDs 14,377 685 / day 63 Theory of mind 14,317 682 / day 64 Enmeshment 14,275 680 / day 65 Greensleeves 14,100 671 / day 66 Null hypothesis 13,864 660 / day 67 Spectral density 13,676 651 / day 68 Ku Klux Klan titles and vocabulary 13,619 649 / day 69 List of Internet phenomena 13,595 647 / day 70 Effect size 13,570 646 / day 71 Gram-negative bacteria 13,488 642 / day 72 Consumer price index 13,420 639 / day 73 Glycine 13,361 636 / day 74 Water polo 12,715 605 / day 75 Filioque 12,426 592 / day 76 Swift (programming language) 12,404 591 / day 77 ISO metric screw thread 12,348 588 / day 78 Gravastar 11,967 570 / day 79 Berberine 11,908 567 / day 80 FFmpeg 11,877 566 / day 81 Type I and type II errors 11,783 561 / day 82 Busy beaver 11,749 559 / day 83 Rainbow table 11,670 556 / day 84 Latex 11,627 554 / day 85 AV1 11,615 553 / day 86 Problem solving 11,608 553 / day 87 Paralegal 11,565 551 / day 88 Strategy 11,510 548 / day 89 Bipolar junction transistor 11,260 536 / day 90 Is-a 11,246 536 / day 91 Vancomycin 11,176 532 / day 92 Eigendecomposition of a matrix 11,153 531 / day 93 Carl Rogers 11,130 530 / day 94 Younger Dryas impact hypothesis 10,984 523 / day |
parse_articles.py
|
|---|
import re
import json
import requests
from typing import List, Dict, Optional, Tuple
def parse_technical_template(wikitext: str) -> Tuple[bool, bool, Optional[int]]:
"""
Parse the technical template from the wikitext.
Returns a tuple of (needs_summarization, is_section, template_position)
where:
- needs_summarization: Whether the article needs summarization
- is_section: True if the template is for a section, False if it's for the entire article
- template_position: Position of the template in the wikitext (for section extraction)
"""
# Find all Technical templates (using case-insensitive flag)
template_matches = list(re.finditer(r'{{Technical(\|.*?)?}}', wikitext, re.IGNORECASE))
for match in template_matches:
template_pos = match.start()
template_text = match.group(0)
# Check if this is a section template by looking for the 'section' parameter
# anywhere in the template parameters (case-insensitive)
is_section = bool(re.search(r'\|section\b', template_text, re.IGNORECASE))
if is_section:
return True, True, template_pos
else:
# It's a general technical template
return True, False, None
# No technical template found
return False, False, None
def get_article_wikitext(title: str) -> str:
"""Fetch the wikitext of an article using the MediaWiki API"""
params = {
'action': 'parse',
'page': title,
'prop': 'wikitext',
'format': 'json'
}
api_url = 'https://en.wikipedia.org/w/api.php'
response = requests.get(api_url, params=params)
data = response.json()
if 'error' in data:
print(f"Error fetching {title}: {data['error']['info']}")
return ""
return data['parse']['wikitext']['*']
def extract_section_wikitext(wikitext: str, template_position: int) -> Tuple[str, str]:
"""
Extract the wikitext for the section containing the template.
This function finds the section heading (==) preceding the template
and includes all content until the next heading of same or lower level.
Returns:
Tuple[str, str]: A tuple of (section_wikitext, section_name)
"""
if template_position is None:
return "", ""
# Find the section heading preceding the template
section_text = wikitext[:template_position]
heading_matches = list(re.finditer(r'^(=+)\s*([^=].*?)\s*\1\s*$', section_text, re.MULTILINE))
if not heading_matches:
return "", "" # No section found
# Get the last heading before the template
last_heading = heading_matches[-1]
heading_level = len(last_heading.group(1)) # Number of = signs
section_name = last_heading.group(2).strip() # Extract the actual section name
section_start = last_heading.start()
# Find the next heading of same or lower level
rest_of_text = wikitext[template_position:] # Start search from the template position, not section start
next_heading_pattern = r'^(=+)\s*([^=].*?)\s*\1\s*$'
next_headings = list(re.finditer(next_heading_pattern, rest_of_text, re.MULTILINE))
if next_headings:
for heading in next_headings:
next_level = len(heading.group(1))
if next_level <= heading_level:
section_end = template_position + heading.start()
return wikitext[section_start:section_end].strip(), section_name
# If we don't find a next heading, include everything until the end
return wikitext[section_start:].strip(), section_name
def read_article_list(file_path: str) -> List[str]:
"""Read the list of articles from the file"""
articles = []
with open(file_path, 'r', encoding='utf-8') as file:
lines = file.readlines()
# Skip the first three lines as mentioned
for line in lines[3:]:
parts = line.strip().split('\t')
if len(parts) >= 2:
# The article title is the second column
articles.append(parts[1])
return articles
def analyze_articles(article_titles: List[str]) -> List[Dict]:
"""Analyze articles and determine if they need summarization"""
results = []
for title in article_titles:
print(f"Processing article: {title}")
wikitext = get_article_wikitext(title)
if not wikitext:
continue
needs_summarization, is_section, template_position = parse_technical_template(wikitext)
if needs_summarization:
result = {
"title": title,
"needs_summarization": True
}
if is_section:
# Only summarize the specific section
section_text, section_name = extract_section_wikitext(wikitext, template_position)
result["section_name"] = section_name
result["wikitext"] = section_text
else:
# Summarize the entire article
result["section_name"] = "entire article"
result["wikitext"] = wikitext
results.append(result)
else:
# Article doesn't need summarization
results.append({
"title": title,
"needs_summarization": False
})
return results
def main():
file_path = "popularTechnical.txt"
output_file = "technical_articles.json"
# Read the list of articles
article_titles = read_article_list(file_path)
print(f"Found {len(article_titles)} articles to analyze")
# Analyze the articles
results = analyze_articles(article_titles)
# Save the results to a JSON file
with open(output_file, 'w', encoding='utf-8') as f:
json.dump(results, f, indent=2, ensure_ascii=False)
print(f"Results saved to {output_file}")
print(f"Articles needing summarization: {sum(1 for r in results if r.get('needs_summarization', False))}")
if __name__ == "__main__":
main()
|
obtain_summaries.py
|
|---|
import json
from os import environ
from google import genai
from google.genai import types
# Global constants for file paths
INPUT_FILE = "technical_articles.json"
OUTPUT_FILE = "summaries.json"
# Template for summary prompts
SUMMARY_PROMPT_TEMPLATE = """Please provide a one paragraph summary at grade {grade_level} reading level for the following wikitext {section_text}English Wikipedia article "{title}". Format your response as raw wikitext (not in a code block) with the following requirements:
1. Use an encyclopedic tone that avoids first and second person pronouns (no "you", "we", "I")
2. Maintain a neutral point of view using precise language
3. Present facts without embellishment in third person, present tense where appropriate
4. Include citations using <ref> tags for key facts (e.g., <ref>Source details</ref>)
5. Use wiki formatting for emphasis (''italics'', '''bold''') instead of markdown
6. If linking to other Wikipedia articles, use [[article name]] format
Your response should be directly usable in Wikipedia without further formatting changes.
{wikitext}"""
def load_existing_summaries():
"""Load existing summaries from file if it exists"""
try:
with open(OUTPUT_FILE, "r", encoding="utf-8") as f:
return json.load(f)
except (FileNotFoundError, json.JSONDecodeError):
return {}
def save_summaries(summaries):
"""Save summaries to file"""
with open(OUTPUT_FILE, "w", encoding="utf-8") as f:
json.dump(summaries, f, indent=2, ensure_ascii=False)
def create_summary_prompt(title, section_name, wikitext, grade_level=9):
"""Create a summary prompt for the API"""
# Base prompt template
section_text = f"from the {section_name} section of the " if section_name != "entire article" else "from the "
prompt = SUMMARY_PROMPT_TEMPLATE.format(
grade_level=grade_level,
section_text=section_text,
title=title,
wikitext=wikitext
)
return prompt
def get_summary(title, section_name, wikitext, grade_level=9):
"""Get summary from Google Gemini API"""
client = genai.Client(api_key=environ["GEMINI_API_KEY"])
# Build the prompt
prompt = create_summary_prompt(title, section_name, wikitext, grade_level)
response = client.models.generate_content(
model="gemini-2.5-pro-exp-03-25",
config=types.GenerateContentConfig(temperature=0),
contents=prompt)
return response.text
def main():
# Check if API key is set
if "GEMINI_API_KEY" not in environ:
print("Error: GEMINI_API_KEY environment variable not set")
print("Get one for free at: https://aistudio.google.com/app/apikey")
print("Please set it with: export GEMINI_API_KEY=your_api_key")
return
# Load existing summaries
summaries = load_existing_summaries()
# Load articles that need summarization
try:
with open(INPUT_FILE, "r", encoding="utf-8") as f:
articles = json.load(f)
except FileNotFoundError:
print(f"Error: {INPUT_FILE} not found")
print("Please run parse_articles.py first")
return
# Reading level to use (can be easily changed in one place)
grade_level = 7
# Count articles that need summarization
to_summarize = [a for a in articles if a.get("needs_summarization", False) and a["title"] not in summaries]
total = len(to_summarize)
print(f"Found {total} articles that need summarization")
print(f"Using grade {grade_level} reading level for summaries")
# Generate summaries for articles that need it
for i, article in enumerate(to_summarize, 1):
title = article["title"]
section_name = article.get("section_name", "entire article")
wikitext = article["wikitext"]
print(f"Generating summary {i}/{total}: {title} [{section_name}]")
try:
summary = get_summary(title, section_name, wikitext, grade_level)
summaries[title] = {
"title": title,
"section_name": section_name,
"summary": summary
}
# Save after each successful summary to avoid losing work
save_summaries(summaries)
print(f"✓ Summary saved for {title}")
except Exception as e:
print(f"× Error generating summary for {title}: {str(e)}")
print(f"Completed. {len(summaries)} total summaries in {OUTPUT_FILE}")
if __name__ == "__main__":
main()
|
prepare_summaries.py
|
|---|
import json
import re
import html
def load_summaries():
"""Load summaries from file"""
try:
with open("summaries.json", "r", encoding="utf-8") as f:
return json.load(f)
except (FileNotFoundError, json.JSONDecodeError):
print("Error: summaries.json not found or invalid")
return {}
def markdown_to_wikitext(text):
"""Convert markdown formatting to wikitext"""
# Convert *bold* to ''italic''
text = re.sub(r'\*([^*]+)\*', r"''\1''", text)
# Convert [text](link) to [link text]
text = re.sub(r'\[([^\]]+)\]\(([^)]+)\)', r'[\2 \1]', text)
return text
def create_talk_page_message(title, section_name, summary):
"""Create a talk page message for the article"""
# Clean up the summary and convert markdown to wikitext
summary = markdown_to_wikitext(summary.strip())
# Create the talk page URL
talk_url = f"https://en.wikipedia.org/w/index.php?title=Talk:{title}&action=edit§ion=new&preloadtitle=Proposed%20summary%20for%20technical%20prose"
# Create the message based on whether it's a section or the entire article
if section_name == "entire article":
message = f"""I've been using Google's Gemini 2.5 Pro Experimental [[large language model]] to create summaries for the most popular articles with {{{{tl|Technical}}}} templates. This article, {title}, has such a template above the entire article. Here is the paragraph summary at grade 5 reading level which Gemini 2.5 Pro suggested:
:{summary}
While I have read and may have made some modifications to that summary, I am not going to add it to the article because I want other editors to review, revise if appropriate, and add it instead. This is an experiment with a few dozen articles initially to see how these suggestions are received, and after a week or two, I will decide how to proceed. Thank you for your consideration. ~~~~"""
else:
message = f"""I've been using Google's Gemini 2.5 Pro Experimental [[large language model]] to create summaries for the most popular articles with {{{{tl|Technical}}}} templates. This article, {title}, has such a template in the "[[{title}#{section_name}|{section_name}]]" section. Here is the paragraph summary at grade 5 reading level which Gemini 2.5 Pro suggested for that section:
:{summary}
While I have read and may have made some modifications to that summary, I am not going to add it to the section because I want other editors to review, revise if appropriate, and add it instead. This is an experiment with a few dozen articles initially to see how these suggestions are received, and after a week or two, I will decide how to proceed. Thank you for your consideration. ~~~~"""
# Escape HTML special characters in the message for the text area
escaped_message = html.escape(message)
# Create HTML with a copy button
html_output = f"""<div class="talk-page-message">
<h2>{title} - {section_name}</h2>
<p><a href="{talk_url}" target="_blank">Add a new talk page section</a></p>
<div class="copy-container">
<textarea class="message-text" rows="15" cols="100" readonly>{escaped_message}</textarea>
<button class="copy-button" onclick="copyText(this)">Copy</button>
</div>
</div>"""
return html_output
def generate_html_page(summaries):
"""Generate a complete HTML page with all talk page messages"""
html_head = """<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Wikipedia Talk Page Messages</title>
<style>
body {
font-family: Arial, sans-serif;
line-height: 1.6;
margin: 0;
padding: 20px;
background-color: #f5f5f5;
}
h1 {
color: #333;
text-align: center;
}
.talk-page-message {
background-color: white;
border-radius: 8px;
padding: 15px;
margin-bottom: 20px;
box-shadow: 0 2px 4px rgba(0,0,0,0.1);
}
h2 {
margin-top: 0;
color: #0366d6;
}
.copy-container {
position: relative;
margin-top: 10px;
}
.message-text {
width: 100%;
border: 1px solid #ddd;
border-radius: 4px;
padding: 10px;
font-family: monospace;
resize: vertical;
}
.copy-button {
position: absolute;
top: 5px;
right: 5px;
background-color: #0366d6;
color: white;
border: none;
border-radius: 4px;
padding: 5px 10px;
cursor: pointer;
}
.copy-button:hover {
background-color: #0353a8;
}
.success {
background-color: #28a745;
}
</style>
</head>
<body>
<h1>Wikipedia Talk Page Messages</h1>
"""
html_messages = ""
for title, data in summaries.items():
section_name = data.get("section_name", "entire article")
summary = data.get("summary", "")
html_messages += create_talk_page_message(title, section_name, summary) + "\n"
html_foot = """<script>
function copyText(button) {
const container = button.parentElement;
const textarea = container.querySelector('.message-text');
textarea.select();
document.execCommand('copy');
// Show success state
button.textContent = 'Copied!';
button.classList.add('success');
// Reset after 2 seconds
setTimeout(() => {
button.textContent = 'Copy';
button.classList.remove('success');
}, 2000);
}
</script>
</body>
</html>"""
return html_head + html_messages + html_foot
def main():
# Load summaries
summaries = load_summaries()
if not summaries:
print("No summaries found.")
return
# Generate HTML
html_content = generate_html_page(summaries)
# Write to file
with open("talk_page_messages.html", "w", encoding="utf-8") as f:
f.write(html_content)
print(f"Generated talk page messages for {len(summaries)} articles.")
print("Output saved to talk_page_messages.html")
if __name__ == "__main__":
main()
|