bomboradata/dragnet

Name: dragnet

Owner: Bombora

Description: Just the facts -- web page content extraction

Created: 2017-08-04 22:56:31.0

Updated: 2017-08-09 15:30:33.0

Pushed: 2018-04-24 04:51:17.0

Homepage: null

Size: 347042

Language: Python

GitHub Committers

User	Most Recent Commit	# Commits

Other Committers

User	Email	Most Recent Commit	# Commits

README

Dragnet

Dragnet isn't interested in the shiny chrome or boilerplate dressing of a web page. It's interested in… 'just the facts.' The machine learning models in Dragnet extract the main article content and optionally user generated comments from a web page. They provide state of the art performance on variety of test benchmarks.

For more information on our approach check out:

• Our paper Content Extraction Using Diverse Feature Sets, published at WWW in 2013, gives an overview of the machine learning approach.
• A comparison of Dragnet and alternate content extraction packages.
• This blog post explains the intuition behind the algorithms.

This project was originally inspired by Kohlschütter et al, Boilerplate Detection using Shallow Text Features and Weninger et al CETR – Content Extraction with Tag Ratios, and more recently by Readability.

GETTING STARTED

Depending on your use case, we provide two separate functions to extract just the main article content or the content and any user generated comments. Each function takes an HTML string and returns the content string.

rt requests
 dragnet import extract_content, extract_content_and_comments

tch HTML
= 'https://moz.com/devblog/dragnet-content-extraction-from-diverse-feature-sets/'
requests.get(url)

t main article without comments
ent = extract_content(r.content)

t article and comments
ent_comments = extract_content_and_comments(r.content)

We also provide a sklearn-style extractor class(complete with fit and predict methods). You can either train an extractor yourself, or load a pre-trained one:

 dragnet.util import load_pickled_model

ent_extractor = load_pickled_model(
        'kohlschuetter_readability_weninger_content_model.pkl.gz')
ent_comments_extractor = load_pickled_model(
        'kohlschuetter_readability_weninger_comments_content_model.pkl.gz')

ent = content_extractor.extract(r.content)
ent_comments = content_comments_extractor.extract(r.content)

A note about encoding

If you know the encoding of the document (e.g. from HTTP headers), you can pass it down to the parser:

ent = content_extractor.extract(html_string, encoding='utf-8')

Otherwise, we try to guess the encoding from a meta tag or specified <?xml encoding=".."?> tag. If that fails, we assume “UTF-8”.

Installing

Dragnet is written in Python (developed with 2.7, with support recently added for 3) and built on the numpy/scipy/Cython numerical computing environment. In addition we use lxml (libxml2) for HTML parsing.

We recommend installing from the master branch to ensure you have the latest version.

Installing with Vagrant:

This is the easiest method to install Dragnet and builds a Vagrant virtual machine with Dragnet and it's dependencies.

1. Install vagrant.
2. Install Virtual Box.
3. Clone the master branch: git clone git@github.com:seomoz/dragnet.git
4. Bring up the vagrant box: vagrant up
5. Log into the vagrant box:

ant ssh
ese should now pass
ke test

Installing without Vagrant

1. Install the dependencies need for Dragnet. The build depends on numpy, Cython and lxml (which in turn depends on libxml2). We use provision.sh to provision the Vagrant VM so you can use it as a template and modify as appropriate for your operation system.
2. Clone the master branch: git clone git@github.com:seomoz/dragnet.git
3. Install the requirements: sudo pip install -r dragnet/requirements.txt
4. Build dragnet

 dragnet
do make install
ese should now pass
ke test

Contributing

We love contributions! Open an issue, or fork/create a pull request.

More details about the code structure

The Extractor class encapsulates a blockifier, some feature extractors, and and a machine learning model.

A blockifier implements blockify that takes a HTML string and returns a list of block objects. A feature extractor is a callable that takes a list of blocks and returns a numpy array of features (len(blocks), nfeatures). There is some additional optional functionality to “train” the feature (e.g. estimate parameters needed for centering) specified in features.py. The machine learning model implements the scikits-learn interface (predict and fit) and is used to compute the content/no-content prediction for each block.

Training/test data

The training and test data is available at dragnet_data.

Training content extraction models

1. Download the training data (see above). In what follows ROOTDIR contains the root of the dragnet_data repo, another directory with similar structure (HTML and Corrected sub-directories).

2. Create the block corrected files needed to do supervised learning on the block level. First make a sub-directory $ROOTDIR/block_corrected/ for the output files, then run:

    dragnet.data_processing import extract_all_gold_standard_data
   dir = '/path/to/dragnet_data/'
   act_all_gold_standard_data(rootdir)
   

   This solves the longest common sub-sequence problem to determine which blocks were extracted in the gold standard. Occasionally this will fail if lxml (libxml2) cannot parse a HTML document. In this case, remove the offending document and restart the process.

3. Use k-fold cross validation in the training set to do model selection and set any hyperparameters. Make decisions about the following:

   • Whether to use just article content or content and comments.
   • The features to use
   • The machine learning model to use

   For example, to train the randomized decision tree classifier from sklearn using the shallow text features from Kohlschuetter et al. and the CETR features from Weninger et al.:

    dragnet.blocks import TagCountNoCSSReadabilityBlockifier
    dragnet.extractor import Extractor
    dragnet.model_training import train_model
    sklearn.ensemble import ExtraTreesClassifier
   
   dir = '/path/to/dragnet_data/'
   
   ures = ['kohlschuetter', 'weninger', 'readability']
   
   xtract = 'both'   # or 'content'
   
   l = ExtraTreesClassifier(
   n_estimators=10,
   max_features=None,
   min_samples_leaf=75
   
   _extractor = Extractor(TagCountNoCSSReadabilityBlockifier,
   features=features,
   to_extract=to_extract,
   model=model
   
   
   actor = train_model(base_extractor, rootdir)
   

   This trains the model and, if a value is passed to output_dir, writes a pickled version of it along with some some block level classification errors to a file in the specified output_dir. If no output_dir is specified, the block-level performance is printed to stdout.

4. Once you have decided on a final model, train it on the entire training data using dragnet.model_training.train_models.

5. As a last step, test the performance of the model on the test set (see below).

Evaluating content extraction models

Use evaluate_models_predictions in model_training to compute the token level accuracy, precision, recall, and F1. For example, to evaluate a trained model run:

 dragnet.compat import train_test_split
 dragnet.data_processing import prepare_all_data
 dragnet.model_training import evaluate_model_predictions

dir = '/path/to/dragnet_data/'
 = prepare_all_data(rootdir)
ning_data, test_data = train_test_split(data, test_size=0.2, random_state=42)

_blocks, test_labels, test_weights = extractor.concatenate_data(test_data)
n_blocks, train_labels, train_weights = extractor.concatenate_data(training_data)

actor.fit(train_blocks, train_labels, weights=train_weights)
ictions = extractor.predict(test_blocks)
es = evaluate_model_predictions(test_labels, predictions)