From 92e50a676c27ba29f3168f1720b11b4ba1258b0d Mon Sep 17 00:00:00 2001 From: chungyau97 Date: Wed, 5 Jul 2023 16:47:01 +0800 Subject: [PATCH] add web crawl --- .../nodes/documentloaders/Cheerio/Cheerio.ts | 48 ++++++----- packages/components/package.json | 1 + packages/components/src/utils.ts | 84 +++++++++++++++++++ 3 files changed, 113 insertions(+), 20 deletions(-) diff --git a/packages/components/nodes/documentloaders/Cheerio/Cheerio.ts b/packages/components/nodes/documentloaders/Cheerio/Cheerio.ts index 9e113505..10eff77e 100644 --- a/packages/components/nodes/documentloaders/Cheerio/Cheerio.ts +++ b/packages/components/nodes/documentloaders/Cheerio/Cheerio.ts @@ -2,7 +2,7 @@ import { INode, INodeData, INodeParams } from '../../../src/Interface' import { TextSplitter } from 'langchain/text_splitter' import { CheerioWebBaseLoader } from 'langchain/document_loaders/web/cheerio' import { test } from 'linkifyjs' -import { getAvailableURLs } from '../../../src' +import { webCrawl } from '../../../src' class Cheerio_DocumentLoaders implements INode { label: string @@ -35,19 +35,20 @@ class Cheerio_DocumentLoaders implements INode { optional: true }, { - label: 'Web Scrap for Relative Links', - name: 'webScrap', + label: 'Web Crawl for Relative Links', + name: 'boolWebCrawl', type: 'boolean', optional: true, additionalParams: true }, { - label: 'Web Scrap Links Limit', + label: 'Web Crawl Links Limit', name: 'limit', type: 'number', default: 10, optional: true, - additionalParams: true + additionalParams: true, + description: 'Set 0 to crawl all relative links' }, { label: 'Metadata', @@ -62,7 +63,7 @@ class Cheerio_DocumentLoaders implements INode { async init(nodeData: INodeData): Promise { const textSplitter = nodeData.inputs?.textSplitter as TextSplitter const metadata = nodeData.inputs?.metadata - const webScrap = nodeData.inputs?.webScrap as boolean + const boolWebCrawl = nodeData.inputs?.boolWebCrawl as boolean let limit = nodeData.inputs?.limit as string let url = nodeData.inputs?.url as string @@ -71,25 +72,32 @@ class Cheerio_DocumentLoaders implements INode { throw new Error('Invalid URL') } - const cheerioLoader = async (url: string): Promise => { - let docs = [] - const loader = new CheerioWebBaseLoader(url) - if (textSplitter) { - docs = await loader.loadAndSplit(textSplitter) - } else { - docs = await loader.load() + async function cheerioLoader(url: string): Promise { + try { + let docs = [] + const loader = new CheerioWebBaseLoader(url) + if (textSplitter) { + docs = await loader.loadAndSplit(textSplitter) + } else { + docs = await loader.load() + } + return docs + } catch (err) { + if (process.env.DEBUG === 'true') console.error(`error in CheerioWebBaseLoader: ${err.message}, on page: ${url}`) } - return docs } - let availableUrls: string[] let docs = [] - if (webScrap) { - if (!limit) limit = '10' - availableUrls = await getAvailableURLs(url, parseInt(limit)) - for (let i = 0; i < availableUrls.length; i++) { - docs.push(...(await cheerioLoader(availableUrls[i]))) + if (boolWebCrawl) { + if (process.env.DEBUG === 'true') console.info('Start Web Crawl') + if (!limit) throw new Error('Please set a limit to crawl') + else if (parseInt(limit) < 0) throw new Error('Limit cannot be less than 0') + const pages: string[] = await webCrawl(url, parseInt(limit)) + if (process.env.DEBUG === 'true') console.info(`pages: ${JSON.stringify(pages)}, length: ${pages.length}`) + for (const page of pages) { + docs.push(...(await cheerioLoader(page))) } + if (process.env.DEBUG === 'true') console.info('Finish Web Crawl') } else { docs = await cheerioLoader(url) } diff --git a/packages/components/package.json b/packages/components/package.json index 4555a883..81f963ed 100644 --- a/packages/components/package.json +++ b/packages/components/package.json @@ -24,6 +24,7 @@ "@qdrant/js-client-rest": "^1.2.2", "@supabase/supabase-js": "^2.21.0", "@types/js-yaml": "^4.0.5", + "@types/jsdom": "^21.1.1", "axios": "^0.27.2", "cheerio": "^1.0.0-rc.12", "chromadb": "^1.4.2", diff --git a/packages/components/src/utils.ts b/packages/components/src/utils.ts index c247ebc2..63bb3969 100644 --- a/packages/components/src/utils.ts +++ b/packages/components/src/utils.ts @@ -2,6 +2,7 @@ import axios from 'axios' import { load } from 'cheerio' import * as fs from 'fs' import * as path from 'path' +import { JSDOM } from 'jsdom' import { BaseCallbackHandler } from 'langchain/callbacks' import { Server } from 'socket.io' import { ChainValues } from 'langchain/dist/schema' @@ -201,6 +202,89 @@ export const getAvailableURLs = async (url: string, limit: number) => { } } +function getURLsFromHTML(htmlBody: string, baseURL: string): string[] { + const dom = new JSDOM(htmlBody) + const linkElements = dom.window.document.querySelectorAll('a') + const urls: string[] = [] + for (const linkElement of linkElements) { + if (linkElement.href.slice(0, 1) === '/') { + try { + const urlObj = new URL(baseURL + linkElement.href) + urls.push(urlObj.href) //relative + } catch (err) { + if (process.env.DEBUG === 'true') console.error(`error with relative url: ${err.message}`) + continue + } + } else { + try { + const urlObj = new URL(linkElement.href) + urls.push(urlObj.href) //absolute + } catch (err) { + if (process.env.DEBUG === 'true') console.error(`error with absolute url: ${err.message}`) + continue + } + } + } + return urls +} + +function normalizeURL(urlString: string): string { + const urlObj = new URL(urlString) + const hostPath = urlObj.hostname + urlObj.pathname + if (hostPath.length > 0 && hostPath.slice(-1) == '/') { + // handling trailing slash + return hostPath.slice(0, -1) + } + return hostPath +} + +export async function crawl(baseURL: string, currentURL: string, pages: string[], limit: number): Promise { + const baseURLObj = new URL(baseURL) + const currentURLObj = new URL(currentURL) + + if (limit !== 0) if (pages.length === limit) return pages + + if (baseURLObj.hostname !== currentURLObj.hostname) return pages + + const normalizeCurrentURL = baseURLObj.protocol + '//' + normalizeURL(currentURL) + if (pages.includes(normalizeCurrentURL)) { + return pages + } + + pages.push(normalizeCurrentURL) + + if (process.env.DEBUG === 'true') console.info(`actively crawling ${currentURL}`) + try { + const resp = await fetch(currentURL) + + if (resp.status > 399) { + if (process.env.DEBUG === 'true') console.error(`error in fetch with status code: ${resp.status}, on page: ${currentURL}`) + return pages + } + + const contentType: string | null = resp.headers.get('content-type') + if ((contentType && !contentType.includes('text/html')) || !contentType) { + if (process.env.DEBUG === 'true') console.error(`non html response, content type: ${contentType}, on page: ${currentURL}`) + return pages + } + + const htmlBody = await resp.text() + const nextURLs = getURLsFromHTML(htmlBody, baseURL) + for (const nextURL of nextURLs) { + pages = await crawl(baseURL, nextURL, pages, limit) + } + } catch (err) { + if (process.env.DEBUG === 'true') console.error(`error in fetch url: ${err.message}, on page: ${currentURL}`) + } + return pages +} + +export async function webCrawl(stringURL: string, limit: number): Promise { + const URLObj = new URL(stringURL) + const modifyURL = stringURL.slice(-1) === '/' ? stringURL.slice(0, -1) : stringURL + return await crawl(URLObj.protocol + '//' + URLObj.hostname, modifyURL, [], limit) +} + /** * Custom chain handler class */