Turn unstructured HTML pages into structured data. The OpenScraping library can extract information from HTML pages using a JSON config file with xPath rules. It can scrape even multi-level complex objects such as tables and forum posts.
This library is used in production to scrape thousands of pages.
The latest NuGet package is .NET Standard 2.0, which means it can be used both in .NET Core 2.0+ and .NET Framework 4.6.1+ projects.
Create a new console C# project, then install the OpenScraping NuGet package by using the GUI or by using this command in the Package Manager Console:
Install-Package OpenScrapingThen paste and run the following code:
namespace OpenScrapingTest
{
using System;
using Newtonsoft.Json;
using OpenScraping;
using OpenScraping.Config;
class Program
{
static void Main(string[] args)
{
var configJson = @"
{
'title': '//h1',
'body': '//div[contains(@class, \'article\')]'
}
";
var config = StructuredDataConfig.ParseJsonString(configJson);
var html = "<html><body><h1>Article title</h1><div class='article'>Article contents</div></body></html>";
var openScraping = new StructuredDataExtractor(config);
var scrapingResults = openScraping.Extract(html);
Console.WriteLine(scrapingResults["title"]);
Console.WriteLine("----------------------------");
Console.WriteLine(JsonConvert.SerializeObject(scrapingResults, Formatting.Indented));
Console.ReadKey();
}
}
}The output looks like this:
Article title
----------------------------
{
"title": "Article title",
"body": "Article contents"
}
Below is a simple configuration file that extracts an article from a www.bbc.com page.
{
"title": "//div[contains(@class, 'story-body')]//h1",
"dateTime": "//div[contains(@class, 'story-body')]//div[contains(@class, 'date')]",
"body": "//div[@property='articleBody']"
}Here is how to call the library:
// www.bbc.com.json contains the JSON configuration file pasted above
var jsonConfig = File.ReadAllText(@"www.bbc.com.json");
var config = StructuredDataConfig.ParseJsonString(jsonConfig);
var html = File.ReadAllText(@"www.bbc.com.html", Encoding.UTF8);
var openScraping = new StructuredDataExtractor(config);
var scrapingResults = openScraping.Extract(html);
Console.WriteLine(JsonConvert.SerializeObject(scrapingResults, Formatting.Indented));And here is the result for a bbc news article:
{
title: 'Robert Downey Jr pardoned for 20-year-old drug conviction',
dateTime: '24 December 2015',
body: 'Body of the article is shown here'
}Here is how the www.bbc.com page looked like on the day we saved the HTML for this sample:
The sample configuration below is more complex as it demonstrates support for extracting multiple items at the same time, and running transformations on them. For this example we are using a products page from ikea.com.
{
"products":
{
"_xpath": "//div[@id='productLists']//div[starts-with(@id, 'item_')]",
"title": ".//div[contains(@class, 'productTitle')]",
"description": ".//div[contains(@class, 'productDesp')]",
"price":
{
"_xpath": ".//div[contains(@class, 'price')]/text()[1]",
"_transformations": [
"TrimTransformation"
]
}
}
}Here is a snippet of the result:
{
products: [{
title: 'HEMNES',
description: 'coffee table',
price: '$139.00'
},
...
{
title: 'NORDEN',
description: 'sideboard',
price: '$149.00'
},
{
title: 'SANDHAUG',
description: 'tray table',
price: '$79.99'
}]
}Here is how the www.ikea.com page looked like on the day we saved the HTML for this sample:
You can find more complex examples in the unit tests.
In the Ikea example above we used a transformation called TrimTransformation. Transformation modify the raw extracted HTML nodes in some ways. For instance, TrimTransformation just runs String.Trim() on the extracted text before it gets written to the JSON output.
Below are a few of the built-in transformations, with links to example rules. To see how the example rules are tested, please check the code of the test class, as well as the HTML files in the TestData folder.
| Name | Purpose | Example |
|---|---|---|
| AbbreviatedIntegerTranformation | Converts strings like "9k" to the integer 9,000, or "2m" to 2,000,000, or "5B" to 5,000,000,000. | Here |
| CastToIntegerTransformation | Converts strings to the corresponding integer. For example, converts "12450" to the integer 12,450. | Here |
| ExtractIntegerTransformation | Tries to find an integer in the middle of a string. For instance, for the string "Popularity: 1000 views" it extracts the integer 1,000. Note that if the string would have a comma after the 1, it would just extract 1 as the integer. | Here |
| ListTitleTransformation | Tries to find the "title" of the current unordered or ordered HTML list by looking for some text just above the list. | Here |
| RemoveExtraWhitespaceTransformation | Replaces consecutive spaces with a single space. For the string "hello world" it would return "hello world". | Here |
| SplitTransformation | Splits the string into an array based on a separator. | Here |
| TotalTextLengthAboveListTransformation | Tries to determine the length of the text which is above an unordered or ordered HTML list. | Here |
| TrimTransformation | Runs String.Trim() on the extracted text before it gets written to the JSON output. | Here |
| RegexTransformation | Matches text with regular expressions | Here |
| ParseDateTransformation | Converts text to date | Here |
| HtmlDecode | Decodes HTML with WebUtility.HtmlDecode | Here |
| HtmlEncode | Encodes HTML with WebUtility.HtmlEncode | Here |
| UrlDecode | Decodes text with WebUtility.UrlDecode | Here |
| UrlEncode | Encodes text with WebUtility.UrlEncode | Here |
| ExtractTextTransformation | A better way to extract text if you want to preserve white space between adjacent text nodes. For example, if a text node immediately follows a link, then this transformation outputs the extracted text with a space between the anchor text and the adjacent text. Useful when extracting large text articles. | Here |
You can implement custom transformations in your own code. The library will pick them up through reflection. There are two types of transformations, ones that act on incoming HTML (first transformation in the chain), and ones that act on the output of previous transformations. The first kind implement ITransformationFromHtml and the second one implement ITransformationFromObject. You can actually have one transformation implement both interfaces, such as the ParseDateTransformation.
Let's say you want to extract a news article but before the actual extraction you would like to remove some HTML nodes. You can do that in two ways. The first (deprecated) way is to use the the _removeTags setting, where you can list names of HTML tags that need to be removed before we start processing the xPath rules. The second (better) way is setting the _removeXPaths setting, which allows listing XPath rules to find nodes that we want to remove BEFORE we process the normal _xpath extraction rules.
Example HTML:
<!DOCTYPE html>
<html lang="en" xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta charset="utf-8" />
<title>Page title</title>
</head>
<body>
<h1> Article title </h1>
<div id="primary">
<script>alert('test');</script>
<p>Para1 content</p>
<p>Para2 content</p>
<div id="comments">
<p>Comment1</p>
<p>Comment2</p>
</div>
</div>
</body>
</html>JSON config:
{
"_removeTags": [
"script"
],
"title": {
"_xpath": "//h1",
"_transformations": [
"TrimTransformation"
]
},
"body": {
"_removeXPaths": [
"./div[@id='comments']"
],
"_xpath": "//div[@id='primary']",
"_transformations": [
"RemoveExtraWhitespaceTransformation"
]
}
}Result:
{
"title": "Article title",
"body": "Para1 content Para2 content"
}You can use MultiExtractor to load multiple xPath rule JSON config files for different websites, then allow the code to pick the correct rule depending on the URL you provide. This is useful, for example, if you are parsing a network of websites with similar HTML design but with different URLs. For example, check these two JSON config files: stackexchange.com.json and answers.microsoft.com.json. The first config file defines multiple potential URL patterns that can match that config:
{
"_urlPatterns": [
"^https?:\/\/.+\\.stackexchange\\.com\/questions\/[0-9]+\/.+$",
"^https?:\/\/stackoverflow\\.com\/questions\/[0-9]+\/.+$",
"^https?:\/\/serverfault\\.com\/questions\/[0-9]+\/.+$",
"^https?:\/\/superuser\\.com\/questions\/[0-9]+\/.+$",
"^https?:\/\/askubuntu\\.com\/questions\/[0-9]+\/.+$"
],
...
}We can load multiple of these config files into a MultiExtractor, then we can pass in an HTML file and its corresponding URL. MultiExtractor will then goes over the _urlPatterns in each config file, it picks the config file which matches the URL, then applies the corresponding rules.
var multiExtractor = new MultiExtractor(configRootFolder: "TestData", configFilesPattern: "*.json");
var json = multiExtractor.ParsePage(
url: "http://answers.microsoft.com/en-us/windows/forum/windows_10-win_upgrade/i-want-to-reserve-my-free-copy-of-windows-10-but-i/9c3f7f56-3da8-4b40-a30f-e33772439ee1",
html: File.ReadAllText(Path.Combine("TestData", "answers.microsoft.com.html")));To see a full example search for the MultiWebsiteExtractionTest() function in the test class.