This library supports the .NET Standard 2.0. The core algorithm is a port of the Mozilla Readability library. The original library is stable and used in production inside Firefox. This way we can piggyback on the hard and well-tested work of Mozilla.
SmartReader also add some improvements on the original library, getting more and better metadata:
- site name
- an author and publication date
- the language
- the excerpt of the article
- the featured image
- a list of images found (it can optionally also download them and store as data URI)
- an estimate of the time needed to read the article
It also allows to perform custom operations before and after extracting the article.
Feel free to suggest new features.
It is trivial using the NuGet package.
PM> Install-Package SmartReader
There are mainly two ways to use the library:
-
The first is by creating a new
Reader
object, with the URI as the argument, and then calling theGetArticle
method to obtain the extractedArticle
-
The second one is by using one of the static methods
ParseArticle
ofReader
directly, to return anArticle
.
Both ways are available also through an async method, called respectively GetArticleAsync
and ParseArticleAsync
.
The advantage of using an object, instead of the static method, is that it gives you the chance to set some options.
There is also the option to parse directly a String
or Stream
that you have obtained by some other way. This is available either with one of the ParseArticle
methods or by using the proper Reader
constructor. In either case, you also need to give the original URI. It will not re-download the text, but it needs the URI to make some checks and fixing the links present on the page. If you cannot provide the original uri, you can use a fake one, like https:\\localhost
.
If the extraction fails, the returned Article
object will have the field IsReadable
set to false
.
If fetching the resource fails, the library will throw an HttpRequestException
, you should handle the exception.
The content of the article is unstyled, but it is wrapped in a div
with the id readability-content
that you can style yourself.
The library tries to detect the correct encoding of the text, if the correct tags are present in the text.
On the Article
object you can call GetImagesAsync
to obtain a Task for a list of Image
objects, representing the images found in the extracted article. The method is async because it makes HEAD Requests, to obtain the size of the images and only returns the ones that are bigger than the specified size. The size by default is 75KB.
This is done to exclude things such as images used in the UI.
On the Article
object you can also call ConvertImagesToDataUriAsync
to inline the images found in the article using the data URI scheme. The method is async. This will insert the images into the Content
property of the Article
. This may significantly increase the size of Content
.
The data URI scheme is not efficient, because is using Base64 to encode the bytes of the image. Base64 encoded data is approximately 33% larger than the original data. The purpose of this method is to provide an offline article that can be fully stored long term. This is useful in case the original article is not accessible anymore. The method only converts the images that are bigger than the specified size. The size by default is 75KB. This is done to exclude things such as images used in the UI.
Notice that this method will not store other external elements that are not images, such as embedded videos.
If fetching an image fails, the library will throw an HttpRequestException
, you should handle the exception.
Using the GetArticle
method.
SmartReader.Reader sr = new SmartReader.Reader("https://arstechnica.com/information-technology/2017/02/humans-must-become-cyborgs-to-survive-says-elon-musk/");
sr.Debug = true;
sr.LoggerDelegate = Console.WriteLine;
SmartReader.Article article = sr.GetArticle();
var images = article.GetImagesAsync();
if(article.IsReadable)
{
// do something with it
}
Using the ParseArticle
static method.
SmartReader.Article article = SmartReader.Reader.ParseArticle("https://arstechnica.com/information-technology/2017/02/humans-must-become-cyborgs-to-survive-says-elon-musk/");
if(article.IsReadable)
{
Console.WriteLine($"Article title {article.Title}");
}
The following settings on the Reader
class can be modified.
int
MaxElemsToParse
Max number of nodes supported by this parser.
Default: 0 (no limit)int
NTopCandidates
The number of top candidates to consider when analyzing how tight the competition is among candidates.
Default: 5bool
Debug
Set the Debug option. If set to true the library writes the data on Logger.
Default: falseAction<string>
LoggerDelegate
Delegate of a function that accepts as argument a string; it will receive log messages.
Default: does not do anythingReportLevel
Logging
Level of information written with theLoggerDelegate
. The valid values are the ones for the enumReportLevel
: Issue or Info. The first level logs only errors or issue that could prevent correctly obtaining an article. The second level logs all the information needed for debugging a problematic article.
Default: ReportLevel.Issuebool
ContinueIfNotReadable
The library tries to determine if it will find an article before actually trying to do it. This option decides whether to continue if the library heuristics fails. This value is ignored if Debug is set to true
Default: trueint
CharThreshold
The minimum number of characters an article must have in order to return a result.
Default: 500bool
KeepClasses
Whether to preserve or clean CSS classes.
Default: falseString[]
ClassesToPreserve
The CSS classes that must be preserved in the article, if we opt to not keep all of them.
Default: ["page"]bool
DisableJSONLD
The library look first at JSON-LD to determine metadata. This setting gives you the option of disabling it
Default: falseDictionary<string, int>
MinContentLengthReadearable
The minimum node content length used to decide if the document is readerable (i.e., the library will find something useful)
You can provide a dictionary with values based on language.
Default: 140int
MinScoreReaderable
The minumum cumulated 'score' used to determine if the document is readerable
Default: 20Func<IElement, bool>
IsNodeVisible
The function used to determine if a node is visible. Used in the process of determining if the document is readerable
Default: NodeUtility.IsProbablyVisiblebool
ForceHeaderEncoding
Whether to force the encoding provided in the response header.
Default: falseint
AncestorsDepth
The default level of depth a node must have to be used for scoring.Nodes without as many ancestors as this level are not counted
Default: 5int
ParagraphThreshold
The default number of characters a node must have in order to be used for scoring
Default: 25
The settings MinScoreReaderable
, CharThreshold
and MinContentLengthReadearable
are used in the process of determining if an article is readerable or if the result found is valid.
The algorithm for scoring assign some score to each valid node, then it determines the best node depending on their relationships, i.e., what score ancestors and descendants of the node have. The settings NTopCandidates
, AncestorsDepth
and ParagraphThreshold
can help you customize this process. It makes sense to change them if you are interested in some sites that uses a particular style or design of coding.
The settings ParagraphThreshold
, MinContentLengthReadearable
and CharThreshold
should be customized for content written in non-alphabetical languages.
A brief overview of the Article model returned by the library.
Uri
Uri
Original UriString
Title
TitleString
Byline
Byline of the article, usually containing author and publication dateString
Dir
Direction of the textString
FeaturedImage
The main image of the articleString
Content
Html content of the articleString
TextContent
The plain text of the article with basic formattingString
Excerpt
A summary of the article, based on metadata or first paragraphString
Language
Language string (es. 'en-US')String
Author
Author of the articleString
SiteName
Name of the site that hosts the articleint
Length
Length of the text of the articleTimeSpan
TimeToRead
Average time needed to read the articleDateTime?
PublicationDate
Date of publication of the articlebool
IsReadable
Indicate whether we successfully find an article
It's important to be aware that the fields Byline, Author and PublicationDate are found independently of each other. So there might be some inconsistencies and unexpected data. For instance, Byline may be a string in the form "@Date by @Author" or "@Author, @Date" or any other combination used by the publication.
The TimeToRead calculation is based on the research found in Standardized Assessment of Reading Performance: The New International Reading Speed Texts IReST. It should be accurate if the article is written in one of the languages in the research, but it is just an educated guess for the others languages.
The FeaturedImage property holds the image indicated by the Open Graph or Twitter meta tags. If neither of these is present, and you called the GetImagesAsync
method, it will be set with the first image found.
The TextContent property is based on the pure text content of the HTML (i.e., the concatenations of text nodes. Then we apply some basic formatting, like removing double spaces or the newlines left by the formatting of the HTML code. We also add meaningful newlines for P and BR nodes.
The library could throw some exceptions, that you should handle.
If you set a value for MaxElemsToParse
larger than 0, the library will throw a standard Exception
if the threshold is passed.
If fetching an HTTP resource fails, the library will throw an HttpRequestException
. This will happen both when trying to fetch the whole article and when trying to fetch an image.
This project has the following directory structure.
Folder | Description |
---|---|
docfx_project/ | Contains the DocFx project that generates the documentation website |
src/ | The main source folder |
src/SmartReader | Source for the SmartReader library |
src/SmartReaderTests | Source for the Tests |
src/SmartReaderConsole | Source for example console project |
src/SmartReader.WebDemo | Source for the demo web project |
You can see the demo web live. So you can test for yourself how effective the library can be for you.
There is also a Docker project for the web demo. You can build and run it with the usual docker commands.
docker build -t smartreader-webdemo .
docker run -it -p 5000:5000 smartreader-webdemo
The second command will forward traffic from port 5000 on your local host to the port 5000 of the docker container. This means that you will be able to access the web demo by visiting http://localhost:5000.
This README contains the info to get started in using the library. If you want to know more advanced options, API reference, etc. read the documentation on the main website.