Experiment to use ChatGPT to summarize page sections.
/* jshint browser: true, jquery: true */
/* globals $, mw, OO */
// ==UserScript==
// @name Wikipedia ChatGPT section summaries
// @namespace http://tampermonkey.net/
// @version 0.1
// @description Experiment to use ChatGPT to summarize page sections.
// @author tonythomas01
// @license GPL-3.0-or-later
// @match https://*.wikipedia.org/*
// @icon https://doc.wikimedia.org/oojs-ui/master/demos/dist/themes/wikimediaui/images/icons/robot.svg
// @grant none
// ==/UserScript==
(function () {
const GPTModel = "gpt-3.5-turbo";
const HuggingFacesModelsMap = {
"FacebookBartLargeCNN": "facebook/bart-large-cnn",
};
function main() {
$('.mw-headline').each(function () {
var sectionHeading = $(this);
const editLink = sectionHeading.siblings('.mw-editsection');
const immediateParent = sectionHeading.parent();
if (immediateParent.is('h2') || immediateParent.is('.mw-heading2')) {
var summarizeLink = $('<span>')
.addClass('mw-summarysection')
.addClass('mw-editsection')
.css('margin-left', '0.5em')
.append(
$('<span>')
.addClass('mw-editsection-bracket')
.text('[')
)
.append(
$('<a>')
.attr('href', '#')
.text('summarize')
.click(function (e) {
e.preventDefault();
summarizeSection(sectionHeading);
})
)
.append(
$('<span>')
.addClass('mw-editsection-bracket')
.text(']')
);
editLink.after(summarizeLink);
}
});
}
function loadMain() {
if (!window.$) {
return setTimeout(loadMain, 50);
}
$(main);
}
loadMain();
function fetchSummaryUsingOpenAPI(fixedPromptForChatGPT, openAPIKey, sectionText, callback) {
const gptQuery = fixedPromptForChatGPT + sectionText;
console.log("To GPT: ", gptQuery);
$.ajax({
url: "https://api.openai.com/v1/chat/completions",
method: "POST",
headers: {
"Content-Type": "application/json",
"Authorization": "Bearer " + openAPIKey,
},
data: JSON.stringify({
model: "gpt-3.5-turbo",
messages: [
{
role: "user",
content: gptQuery
}
],
temperature: 0.7
}),
success: function (response) {
const responseContent = response.choices[0].message.content;
callback(null, responseContent);
},
error: function (error) {
callback(error);
}
});
}
function fetchSummaryUsingHuggingFacesModel(apiKeySecret, modelName, sectionText, callback) {
$.ajax({
url: "https://api-inference.huggingface.co/models/" + modelName,
method: "POST",
headers: {
"Content-Type": "application/json",
"Authorization": "Bearer " + apiKeySecret,
},
data: JSON.stringify({
inputs: sectionText
}),
success: function (response) {
const responseContent = response[0].summary_text;
callback(null, responseContent);
},
error: function (error) {
callback(error);
}
});
}
function onSummaryFetch(error, responseContent) {
if (error) {
console.log(error);
} else {
console.log("LLM model responded: " + responseContent);
}
}
function getSectionTextUnderHeading(namespace, sectionParent) {
if (sectionParent.parent().is('.mw-heading')) {
return getSectionText( sectionParent );
} else {
return $.Deferred().resolve(
$('<div>').append(
sectionParent.nextUntil('h2').clone()
).prop('innerText')
);
}
}
var discussionToolsInfo;
/**
* @param {jQuery} $heading
* @return {jQuery.Promise<Array>}
*/
function getSectionData( $heading ) {
var dataPromise;
if ( discussionToolsInfo ) {
dataPromise = $.Deferred().resolve( discussionToolsInfo );
} else {
dataPromise = mw.loader.using( [ 'mediawiki.api' ] ).then( function () {
return new mw.Api().get( {
action: 'discussiontoolspageinfo',
page: mw.config.get( 'wgPageName' ),
prop: 'threaditemshtml',
format: 'json',
formatversion: 2,
} );
} ).then( function ( data ) {
discussionToolsInfo = data;
return data;
} );
}
var sectionId = $heading.find( '.mw-headline' ).data( 'mw-thread-id' );
var sectionContent = [];
var processReplies = function ( reply ) {
if ( reply.type === 'comment' ) {
sectionContent.push( { type: 'comment', level: reply.level, author: reply.author, text: getCommentTextFromHtml( reply.html ) } );
} else if ( reply.type === 'heading' ) {
sectionContent.push( { type: 'heading', level: reply.level, headingLevel: reply.headingLevel, text: getCommentTextFromHtml( reply.html ) } );
for ( var i = 0; i < reply.replies.length; i++ ) {
processReplies( reply.replies[i] );
}
} else {
console.log( 'Unexpected type: ' + reply.type, reply );
}
};
return dataPromise.then( function ( data ) {
for ( var i = 0; i < data.discussiontoolspageinfo.threaditemshtml.length; i++ ) {
var section = data.discussiontoolspageinfo.threaditemshtml[i];
if ( section.id === sectionId ) {
sectionContent.push( { type: 'heading', level: 0, headingLevel: section.level, text: getCommentTextFromHtml( section.html ) } );
for ( var j = 0; j < section.replies.length; j++ ) {
processReplies( section.replies[j] );
}
}
}
return sectionContent;
} );
}
/**
* @param {string} html
* @return {string}
*/
function getCommentTextFromHtml( html ) {
return $.parseHTML( '<div>' + html + '</div>' ).map( el => el.innerText || '' ).join( '' );
}
/**
* @param {jQuery} $heading
* @return {jQuery.Promise<string>}
*/
function getSectionText( $heading ) {
return getSectionData( $heading ).then( function ( data ) {
var sectionText = '';
for ( var i = 0; i < data.length; i++ ) {
var item = data[i];
if ( item.type === 'heading' ) {
sectionText += '\t'.repeat( item.level ) + '='.repeat( item.headingLevel ) + item.text + '='.repeat( item.headingLevel ) + '\n\n';
} else if ( item.type === 'comment' ) {
sectionText += ( item.author + ': ' + item.text ).replace( /^|\n/g, '$&' + '\t'.repeat( item.level ) );
}
sectionText += '\n\n';
}
return sectionText;
} );
}
var apiKey = null;
/**
* @return {string}
*/
function getApiKey() {
if ( apiKey ) {
return apiKey;
}
apiKey = localStorage.getItem( 'LLMApiKey' );
if ( apiKey ) {
return apiKey;
}
apiKey = prompt( 'Please enter your OpenAI API key from https://platform.openai.com/account/api-keys' );
if ( !apiKey ) {
throw new Error( 'Section summary requires an API key!' );
}
localStorage.setItem( 'LLMApiKey', apiKey );
}
function summarizeSection(sectionHeading) {
const sectionParent = sectionHeading.parent();
const selectedLLMModel = GPTModel;
const LLMApiKey = getApiKey();
var fixedPromptForChatGPT = "Summarize the following section in less than 50 words: ";
const namespace = mw.config.get("wgCanonicalNamespace");
if (namespace === "Talk") {
fixedPromptForChatGPT = "Summarize the following section in less than 50 words. See that each row represents a " +
"reply from a user with the Username presented right before (talk). Use the usernames when summarizing. \n";
}
getSectionTextUnderHeading(namespace, sectionParent).then(function(sectionText) {
console.log("Found Section Text:", sectionText);
switch (selectedLLMModel) {
case GPTModel:
fetchSummaryUsingOpenAPI(fixedPromptForChatGPT, LLMApiKey, sectionText, onSummaryFetch);
return;
default:
fetchSummaryUsingHuggingFacesModel(LLMApiKey, HuggingFacesModelsMap[selectedLLMModel], sectionText, onSummaryFetch);
return;
}
});
}
})();