有这么一本Python的书: <<Python 网络数据采集>>

我准备用.NET Core及第三方库实现里面所有的例子.

这是第一部分, 主要使用的是AngleSharp: https://anglesharp.github.io/

(文章的章节书与该书是对应的)

第1章初见网络爬虫

发送Http请求

在python里面这样发送http请求, 它使用的是python的标准库urllib:

在.NET Core里面, 你可以使用HttpClient, 相应的C#代码如下:

            var client = new HttpClient();
            HttpResponseMessage response = await client.GetAsync("http://pythonscraping.com/pages/page1.html");
            response.EnsureSuccessStatusCode();
            var responseBody = await response.Content.ReadAsStringAsync();
            Console.WriteLine(responseBody);
            return responseBody;

或者可以简写为:

                var client = new HttpClient();
                var responseBody = await client.GetStringAsync("http://pythonscraping.com/pages/page1.html");
                Console.WriteLine(responseBody);

其结果如下:

使用AngleSharp解析html源码

python里面可以使用BeautifulSoup或者MechanicalSoup等库对html源码进行解析.

而.NET Core可以使用AngleSharp, Html Agility Pack, DotnetSpider(国产, 也支持元素抽取).等库来操作Html文档.

这里我先使用的是AngleSharp, AngleSharp的解析库可以使用标准的W3C规范来解析HTML, MathML, XML, SVG和CSS. 它支持.NET Standard 1.0.

安装AngleSharp

通过Nuget即可: https://www.nuget.org/packages/AngleSharp/

Install-Package AngleSharp

或者dotnet-cli:

dotnet add package AngleSharp

AngleSharp的一个简单例子

下面这个例子(1.2.2)是把页面中h1元素的内容显示出来.

书中Python的代码:

下面是.NET Core的C#代码:

        public static async Task ReadWithAngleSharpAsync()
        {
            var htmlSourceCode = await SendRequestWithHttpClientAsync();
            var parser = new HtmlParser();
            var document = await parser.ParseAsync(htmlSourceCode);

            Console.WriteLine($"Serializing the (original) document: {document.QuerySelector("h1").OuterHtml}");
            Console.WriteLine($"Serializing the (original) document: {document.QuerySelector("html > body > h1").OuterHtml}");
        }

在这里AngleSharp首先需要创建一个可以循环使用的HtmlParser(Html解析器), 然后使用解析器解析html源码即可: parser.Parse() 或者异步版本 parser.ParseAsync().

解析返回对象的类型是IHtmlDocument, 里面是解析好的DOM. 其中DOM是和AngleSharp里的类这样对应的:

这个图其实是老一点的版本, 新版本的DOM模型是稍微有点不同的, 不过你只要理解这个意思就行...

AngleSharp有很多特点, 但是最重要的特点就是它支持querySelector()和querySelectorAll()方法, 就像DOM的方法一样.

上面这个例子里, 其html的结构大致如下:

所以针对返回的IHtmlDocument对象document, 我们使用document.QuerySelector("h1").OuterHtml, 就可以返回h1的OuterHtml. 而使用document.QuerySelector("html > body > h1").OuterHtml 也是同样的效果, 因为标准的CSS选择器是都支持的.

QuerySelector()返回的是一个/0个元素, 相当于Linq的FirstOrDefault().

其运行结果如下:

异常情况处理

发送Http请求之后, 可能会发生错误, 例如网页不存在(或者请求时出错), 服务器不存在等等.

针对这些情况, .NET Core程序会返回HTTP错误, 可能是404也可能是500等. 但是所有的类型HttpClient都会抛出HttpRequestException, 我们可以这样处理这种异常:

        public static async Task ResponseWithErrorsAsync()
        {
            try
            {
                var client = new HttpClient();
                var responseBody = await client.GetStringAsync("http://notexistwebsite");
                Console.WriteLine(responseBody);
            }
            catch (HttpRequestException e)
            {
                Console.ForegroundColor = ConsoleColor.Red;
                Console.WriteLine("\nException Caught!");
                Console.WriteLine("Message :{0} ", e.Message);
            }
        }

但是即使网页获取成功了, 网页上的内容也并非完全是我们所期待的, 仍可能会抛出异常. 比如说你想要找的标签不存在, 那么就会返回null, 然后再调用改标签的属性, 就会发生NullReferenceException.

所以这种情况可以捕获NullReferenceException, 也可以使用代码判断:

        public static async Task ReadNonExistTagAsync()
        {
            var htmlSourceCode = await SendRequestWithHttpClientAsync();
            var parser = new HtmlParser();
            var document = await parser.ParseAsync(htmlSourceCode);

            var nonExistTag = document.QuerySelector("h8");
            Console.WriteLine(nonExistTag);
            Console.WriteLine($"nonExistTag is null: {nonExistTag is null}");

            try
            {
                Console.WriteLine(nonExistTag.QuerySelector("p").OuterHtml);
            }
            catch (NullReferenceException)
            {
                Console.ForegroundColor = ConsoleColor.Red;
                Console.WriteLine("Tag was not found");
            }
        }

完整的例子:

        public static async Task RunAllAsync()
        {
            Console.ForegroundColor = ConsoleColor.Red;
            async Task<string> GetTileAsync(string uri)
            {
                var httpClient = new HttpClient();
                try
                {
                    var responseHtml = await httpClient.GetStringAsync(uri);
                    var parser = new HtmlParser();
                    var document = await parser.ParseAsync(responseHtml);
                    var tagContent = document.QuerySelector("body > h8").TextContent;
                    return tagContent;
                }
                catch (HttpRequestException e)
                {
                    Console.WriteLine($"{nameof(HttpRequestException)}:");
                    Console.WriteLine("Message :{0} ", e.Message);
                    return null;
                }
                catch (NullReferenceException)
                {
                    Console.WriteLine($"{nameof(NullReferenceException)}:");
                    Console.WriteLine("Tag was not found");
                    return null;
                }
            }

            var title = await GetTileAsync("http://www.pythonscraping.com/pages/page1.html");
            if (string.IsNullOrWhiteSpace(title))
            {
                Console.WriteLine("Title was not found");
            }
            else
            {
                Console.ForegroundColor = ConsoleColor.Green;
                Console.WriteLine(title);
            }
        }

第2章复杂HTML解析

首先我把请求Http返回HTML代码的那部分封装成了一个方法以便复用:

        public static async Task<string> GetHtmlSourceCodeAsync(string uri)
        {
            var httpClient = new HttpClient();
            try
            {
                var htmlSource = await httpClient.GetStringAsync(uri);
                return htmlSource;
            }
            catch (HttpRequestException e)
            {
                Console.ForegroundColor = ConsoleColor.Red;
                Console.WriteLine($"{nameof(HttpRequestException)}: {e.Message}");
                return null;
            }
        }

CSS是网络爬虫的福音, 下面这两个元素在页面中可能会出现很多次:

我们可以使用AngleSharp里面的QuerySelectorAll()方法把所有符合条件的元素都找出来, 返回到一个结果集合里.

        public static async Task FindGreenClassAsync()
        {
            const string url = "http://www.pythonscraping.com/pages/warandpeace.html";
            var html = await GetHtmlSourceCodeAsync(url);
            if (!string.IsNullOrWhiteSpace(html))
            {
                var parser = new HtmlParser();
                var document = await parser.ParseAsync(html);
                var nameList = document.QuerySelectorAll("span > .green");

                Console.WriteLine("Green names are:");
                Console.ForegroundColor = ConsoleColor.Green;
                foreach (var item in nameList)
                {
                    Console.WriteLine(item.TextContent);
                }
            }
            else
            {
                Console.WriteLine("No html source code returned.");
            }
        }

非常简单, 和DOM的标准操作是一样的.

如果只需要元素的文字部分, 那么就是用其TextContent属性即可.

再看个例子

1. 找出页面中所有的h1, h2, h3, h4, h5, h6元素

2. 找出class为green或red的span元素.

        public static async Task FindByAttributeAsync()
        {
            const string url = "http://www.pythonscraping.com/pages/warandpeace.html";
            var html = await GetHtmlSourceCodeAsync(url);
            if (!string.IsNullOrWhiteSpace(html))
            {
                var parser = new HtmlParser();
                var document = await parser.ParseAsync(html);

                var headers = document.QuerySelectorAll("*")
                    .Where(x => new[] { "h1", "h2", "h3", "h4", "h5", "h6" }.Contains(x.TagName.ToLower()));
                Console.WriteLine("Headers are:");
                PrintItemsText(headers);

                var greenAndRed = document.All
                    .Where(x => x.TagName == "span" && (x.ClassList.Contains("green") || x.ClassList.Contains("red")));
                Console.WriteLine("Green and Red spans are:");
                PrintItemsText(greenAndRed);

                var thePrinces = document.QuerySelectorAll("*").Where(x => x.TextContent == "the prince");
                Console.WriteLine(thePrinces.Count());
            }
            else
            {
                Console.WriteLine("No html source code returned.");
            }

            void PrintItemsText(IEnumerable<IElement> elements)
            {
                foreach (var item in elements)
                {
                    Console.WriteLine(item.TextContent);
                }
            }
        }

这里我们可以看到QuerySelectorAll()的返回结果可以使用Linq的Where方法进行过滤, 这样就很强大了.

TagName属性就是元素的标签名.

此外, 还有一个document.All, All属性是该Document所有元素的集合, 它同样也支持Linq.

(该方法中使用了一个本地方法).

由于同时支持CSS选择器和Linq, 所以抽取元素的工作简单多了.

导航树

一个页面, 它的结构可以是这样的:

这里面有几个概念:

子标签和后代标签.

子标签是父标签的下一级, 而后代标签则是指父标签下面所有级别的标签.

tr是table的子标签, tr, th, td, img都是table的后代标签.

使用AngleSharp, 找出子标签可以使用.Children属性. 而找出后代标签, 可以使用CSS选择器.

兄弟标签

找到前一个兄弟标签使用.PreviousElementSibling属性, 后一个兄弟标签是.NextElementSibling属性.

父标签

.ParentElement属性就是父标签.

        public static async Task FindDescendantAsync()
        {
            const string url = "http://www.pythonscraping.com/pages/page3.html";
            var html = await GetHtmlSourceCodeAsync(url);
            if (!string.IsNullOrWhiteSpace(html))
            {
                var parser = new HtmlParser();
                var document = await parser.ParseAsync(html);

                var tableChildren = document.QuerySelector("table#giftList > tbody").Children;
                Console.WriteLine("Table's children are:");
                foreach (var child in tableChildren)
                {
                    System.Console.WriteLine(child.LocalName);
                }

                var descendants = document.QuerySelectorAll("table#giftList > tbody *");
                Console.WriteLine("Table's descendants are:");
                foreach (var item in descendants)
                {
                    Console.WriteLine(item.LocalName);
                }

                var siblings = document.QuerySelectorAll("table#giftList > tbody > tr").Select(x => x.NextElementSibling);
                Console.WriteLine("Table's descendants are:");
                foreach (var item in siblings)
                {
                    Console.WriteLine(item?.LocalName);
                }

                var parentSibling = document.All.SingleOrDefault(x => x.HasAttribute("src") && x.GetAttribute("src") == "../img/gifts/img1.jpg")
                    ?.ParentElement.PreviousElementSibling;
                if (parentSibling != null)
                {
                    Console.WriteLine($"Parent's previous sibling is: {parentSibling.TextContent}");
                }
            }
            else
            {
                Console.WriteLine("No html source code returned.");
            }
        }

结果:

使用正则表达式

"如果你有一个问题打算使用正则表达式来解决, 那么现在你有两个问题了".

这里有一个测试正则表达式的网站: https://www.regexpal.com/

目前, AngleSharp支持通过CSS选择器来查找元素, 也可以使用Linq来过滤元素, 当然也可以通过多种方式使用正则表达式进行更复杂的查找动作.

关于正则表达式我就不介绍了. 直接看例子.

我想找到页面中所有的满足下列要求的图片, 其src的值以../img/gifts/img开头并且随后跟着数字, 然后格式为.jpg的图标.

        public static async Task FindByRegexAsync()
        {
            const string url = "http://www.pythonscraping.com/pages/page3.html";
            var html = await GetHtmlSourceCodeAsync(url);
            if (!string.IsNullOrWhiteSpace(html))
            {
                var parser = new HtmlParser();
                var document = await parser.ParseAsync(html);

                var images = document.QuerySelectorAll("img")
                    .Where(x => x.HasAttribute("src") && Regex.Match(x.Attributes["src"].Value, @"\.\.\/img\/gifts/img.*\.jpg").Success);
                foreach (var item in images)
                {
                    Console.WriteLine(item.Attributes["src"].Value);
                }

                var elementsWith2Attributes = document.All.Where(x => x.Attributes.Length == 2);
                foreach (var item in elementsWith2Attributes)
                {
                    Console.WriteLine(item.LocalName);
                    foreach (var attr in item.Attributes)
                    {
                        Console.WriteLine($"\t{attr.Name} - {attr.Value}");
                    }
                }
            }
            else
            {
                Console.WriteLine("No html source code returned.");
            }
        }

这个其实没有任何难度.

但从本例可以看到, 判断元素有没有一个属性可以使用HasAttribute("xxx")方法, 可以通过.Attributes索引来获取属性, 其属性值就是.Attributes["xxx"].Value.

如果不会正则表达式, 我相信多写的Linq的过滤代码也差不多能达到要求.

第3章开始采集

遍历单个域名

就是几个应用的例子, 直接贴代码吧.

打印出一个页面内所有的超链接地址:

        public static async Task TraversingASingleDomainAsync()
        {
            var httpClient = new HttpClient();
            var htmlSource = await httpClient.GetStringAsync("http://en.wikipedia.org/wiki/Kevin_Bacon");

            var parser = new HtmlParser();
            var document = await parser.ParseAsync(htmlSource);
            var links = document.QuerySelectorAll("a");
            foreach (var link in links)
            {
                Console.WriteLine(link.Attributes["href"]?.Value);
            }
        }

找出满足下列条件的超链接:

在id为bodyContent的div里
url不包括分号
url以/wiki开头

        public static async Task FindSpecificLinksAsync()
        {
            var httpClient = new HttpClient();
            var htmlSource = await httpClient.GetStringAsync("http://en.wikipedia.org/wiki/Kevin_Bacon");

            var parser = new HtmlParser();
            var document = await parser.ParseAsync(htmlSource);
            var links = document.QuerySelector("div#bodyContent").QuerySelectorAll("a")
                .Where(x => x.HasAttribute("href") && Regex.Match(x.Attributes["href"].Value, @"^(/wiki/)((?!:).)*$").Success);
            foreach (var link in links)
            {
                Console.WriteLine(link.Attributes["href"]?.Value);
            }
        }

随机找到页面里面一个连接, 然后递归调用自己的方法, 直到主动停止:

        private static async Task<IEnumerable<IElement>> GetLinksAsync(string uri)
        {
            var httpClient = new HttpClient();
            var htmlSource = await httpClient.GetStringAsync($"http://en.wikipedia.org{uri}");
            var parser = new HtmlParser();
            var document = await parser.ParseAsync(htmlSource);

            var links = document.QuerySelector("div#bodyContent").QuerySelectorAll("a")
                .Where(x => x.HasAttribute("href") && Regex.Match(x.Attributes["href"].Value, @"^(/wiki/)((?!:).)*$").Success);
            return links;
        }

        public static async Task GetRandomNestedLinksAsync()
        {
            var random = new Random();
            var links = (await GetLinksAsync("/wiki/Kevin_Bacon")).ToList();
            while (links.Any())
            {
                var newArticle = links[random.Next(0, links.Count)].Attributes["href"].Value;
                Console.WriteLine(newArticle);
                links = (await GetLinksAsync(newArticle)).ToList();
            }
        }

采集整个网站

首先要了解几个概念:

浅网 surface web: 是互联网上搜索引擎可以直接抓取到的那部分网络.

与浅网对立的就是深网 deep web: 互联网中90%都是深网.

暗网Darknet / dark web / dark internet: 它完全是另外一种怪兽. 它们也建立在已有的网络基础上, 但是使用Tor客户端, 带有运行在HTTP之上的新协议, 提供了一个信息交换的安全隧道. 这类网也可以采集, 但是超出了本书的范围.....

深网相对暗网还是比较容易采集的.

采集整个网站的两个好处:

生成网站地图
收集数据

由于网站的规模和深度, 所以采集到的超链接很多可能是重复的, 这时我们就需要链接去重, 可以使用Set类型的集合:

        private static readonly HashSet<string> LinkSet = new HashSet<string>();
        private static readonly HttpClient HttpClient = new HttpClient();
        private static readonly HtmlParser Parser = new HtmlParser();

        public static async Task GetUniqueLinksAsync(string uri = "")
        {
            var htmlSource = await HttpClient.GetStringAsync($"http://en.wikipedia.org{uri}");
            var document = await Parser.ParseAsync(htmlSource);

            var links = document.QuerySelectorAll("a")
                .Where(x => x.HasAttribute("href") && Regex.Match(x.Attributes["href"].Value, @"^(/wiki/)").Success);

            foreach (var link in links)
            {
                if (!LinkSet.Contains(link.Attributes["href"].Value))
                {
                    var newPage = link.Attributes["href"].Value;
                    Console.WriteLine(newPage);
                    LinkSet.Add(newPage);
                    await GetUniqueLinksAsync(newPage);
                }
            }
        }

(递归调用的深度需要注意一下, 不然有时候能崩溃).

收集整个网站数据

这个例子相对网站, 包括收集相关文字和异常处理等:

        private static readonly HashSet<string> LinkSet = new HashSet<string>();
        private static readonly HttpClient HttpClient = new HttpClient();
        private static readonly HtmlParser Parser = new HtmlParser();

       public static async Task GetLinksWithInfoAsync(string uri = "")
        {
            var htmlSource = await HttpClient.GetStringAsync($"http://en.wikipedia.org{uri}");
            var document = await Parser.ParseAsync(htmlSource);

            try
            {
                var title = document.QuerySelector("h1").TextContent;
                Console.ForegroundColor = ConsoleColor.Green;
                Console.WriteLine(title);

                var contentElement = document.QuerySelector("#mw-content-text").QuerySelectorAll("p").FirstOrDefault();
                if (contentElement != null)
                {
                    Console.WriteLine(contentElement.TextContent);
                }

                var alink = document.QuerySelector("#ca-edit").QuerySelectorAll("span a").SingleOrDefault(x => x.HasAttribute("href"))?.Attributes["href"].Value;
                Console.WriteLine(alink);
            }
            catch (NullReferenceException)
            {
                Console.ForegroundColor = ConsoleColor.Red;
                Console.WriteLine("Cannot find the tag!");
            }

            var links = document.QuerySelectorAll("a")
                .Where(x => x.HasAttribute("href") && Regex.Match(x.Attributes["href"].Value, @"^(/wiki/)").Success).ToList();
            foreach (var link in links)
            {
                if (!LinkSet.Contains(link.Attributes["href"].Value))
                {
                    var newPage = link.Attributes["href"].Value;
                    Console.WriteLine(newPage);
                    LinkSet.Add(newPage);
                    await GetLinksWithInfoAsync(newPage);
                }
            }
        }

不知前方水深的例子

第一个例子, 寻找随机外链:

using System;
using System.Collections.Generic;
using System.Linq;
using System.Net.Http;
using System.Text.RegularExpressions;
using System.Threading.Tasks;
using AngleSharp.Parser.Html;

namespace WebScrapingWithDotNetCore.Chapter03
{
    public class CrawlingAcrossInternet
    {
        private static readonly Random Random = new Random();
        private static readonly HttpClient HttpClient = new HttpClient();
        private static readonly HashSet<string> InternalLinks = new HashSet<string>();
        private static readonly HashSet<string> ExternalLinks = new HashSet<string>();
        private static readonly HtmlParser Parser = new HtmlParser();

        public static async Task FollowExternalOnlyAsync(string startingSite)
        {
            var externalLink = await GetRandomExternalLinkAsync(startingSite);
            if (externalLink != null)
            {
                Console.WriteLine($"External Links is: {externalLink}");
                await FollowExternalOnlyAsync(externalLink);
            }
            else
            {
                Console.WriteLine("Random External link is null, Crawling terminated.");
            }
        }

        private static async Task<string> GetRandomExternalLinkAsync(string startingPage)
        {
            try
            {
                var htmlSource = await HttpClient.GetStringAsync(startingPage);
                var externalLinks = (await GetExternalLinksAsync(htmlSource, SplitAddress(startingPage)[0])).ToList();
                if (externalLinks.Any())
                {
                    return externalLinks[Random.Next(0, externalLinks.Count)];
                }

                var internalLinks = (await GetInternalLinksAsync(htmlSource, startingPage)).ToList();
                if (internalLinks.Any())
                {
                    return await GetRandomExternalLinkAsync(internalLinks[Random.Next(0, internalLinks.Count)]);
                }

                return null;
            }
            catch (HttpRequestException e)
            {
                Console.WriteLine($"Error requesting: {e.Message}");
                return null;
            }
        }

        private static string[] SplitAddress(string address)
        {
            var addressParts = address.Replace("http://", "").Replace("https://", "").Split("/");
            return addressParts;
        }

        private static async Task<IEnumerable<string>> GetInternalLinksAsync(string htmlSource, string includeUrl)
        {
            var document = await Parser.ParseAsync(htmlSource);
            var links = document.QuerySelectorAll("a")
                .Where(x => x.HasAttribute("href") && Regex.Match(x.Attributes["href"].Value, $@"^(/|.*{includeUrl})").Success)
                .Select(x => x.Attributes["href"].Value);
            foreach (var link in links)
            {
                if (!string.IsNullOrEmpty(link) && !InternalLinks.Contains(link))
                {
                    InternalLinks.Add(link);
                }
            }
            return InternalLinks;
        }

        private static async Task<IEnumerable<string>> GetExternalLinksAsync(string htmlSource, string excludeUrl)
        {
            var document = await Parser.ParseAsync(htmlSource);

            var links = document.QuerySelectorAll("a")
                .Where(x => x.HasAttribute("href") && Regex.Match(x.Attributes["href"].Value, $@"^(http|www)((?!{excludeUrl}).)*$").Success)
                .Select(x => x.Attributes["href"].Value);
            foreach (var link in links)
            {
                if (!string.IsNullOrEmpty(link) && !ExternalLinks.Contains(link))
                {
                    ExternalLinks.Add(link);
                }
            }
            return ExternalLinks;
        }

        private static readonly HashSet<string> AllExternalLinks = new HashSet<string>();
        private static readonly HashSet<string> AllInternalLinks = new HashSet<string>();

        public static async Task GetAllExternalLinksAsync(string siteUrl)
        {
            try
            {
                var htmlSource = await HttpClient.GetStringAsync(siteUrl);
                var internalLinks = await GetInternalLinksAsync(htmlSource, SplitAddress(siteUrl)[0]);
                var externalLinks = await GetExternalLinksAsync(htmlSource, SplitAddress(siteUrl)[0]);
                foreach (var link in externalLinks)
                {
                    if (!AllExternalLinks.Contains(link))
                    {
                        AllExternalLinks.Add(link);
                        Console.WriteLine(link);
                    }
                }

                foreach (var link in internalLinks)
                {
                    if (!AllInternalLinks.Contains(link))
                    {
                        Console.WriteLine($"The link is: {link}");
                        AllInternalLinks.Add(link);
                        await GetAllExternalLinksAsync(link);
                    }
                }
            }
            catch (HttpRequestException e)
            {
                Console.WriteLine(e);
                Console.WriteLine($"Request error: {e.Message}");
            }
        }
    }
}