This is a pre-production deployment of Warehouse, however changes made here WILL affect the production instance of PyPI.
Project Description

MaybeDont is a library that helps avoid downloading pages with duplicate content during crawling. It learns which URL components are important and which are not important during crawling, and tries to predict if the page will be duplicate based on it’s URL.

The idea is that if you have a crawler that just follows all links, it might download a lot of duplicate pages: for example, for a forum there might be pages like /view.php?topicId=10 and /view.php?topicId=10&start=0 - the only difference is added start=0, and the content of this pages is likely duplicate. If we knew that adding start=0 does not change content, then we would avoid downloading the page /view.php?topicId=10&start=0 if we have already fetched /view.php?topicId=10, and thus save time and bandwidth.

Duplicate detector

maybedont.DupePredictor collects statistics about page URLs and contents, and is able to predict if the new URL will bring any new content.

First, initialize a DupePredictor:

from maybedont import DupePredictor
dp = DupePredictor(
    texts_sample=[page_1, page_2, page_3],
    jaccard_threshold=0.9)  # default value

texts_sample is a list of page contents. It can be ommited, but it is recommended to provide it: it is used to learn which parts of the page are common for a lot of site’s pages, and excludes this parts from duplicate comparison. This helps with pages where the content is small relative to the site chrome (footer, header, etc.): without removing chrome all such pages would be considered duplicates, as only a tiny fraction of the content changes.

Next, we can update DupePredictor model with downloaded pages:

dp.update_model(url_4, text_4)
dp.update_model(url_5, text_5)

After a while, DupePredictor will learn which arguments in URLs are important, and which can be safely ignored. DupePredictor.get_dupe_prob returns the probability of url being a duplicate of some content that has already been seem:


Runtime overhead should be not too large: on a crawl with < 100k pages, expected time to update the model is 1-5 ms, and below 1 ms to get the probability. All visited urls and hashes of content are stored in memory, along with some indexing structures.


pip install MaybeDont

Spider middleware

If you have a Scrapy spider, or are looking for an inspiration for a spider middleware, check out maybedont.scrapy_middleware.AvoidDupContentMiddleware. First, it collects an queue of documents to know better which page elements are common on the site, in order to exclude them from content comparison. After that it builds it’s DupePredictor, updates it with crawled pages (only textual pages are taken into account), and starts dropping requests for duplicate content once it gets confident enough. Not all requests for duplicates are dropped: with a small probability (currenty 5%) requests are carried anyway. This makes duplicate detection more robust against changes in site URL or content structure as the crawl progresses.

To enable the middleware, the following settings are required:

DOWNLOADER_MIDDLEWARES['maybedont.scrapy_middleware.AvoidDupContentMiddleware'] = 200

Middleware is only applied to requests with avoid_dup_content in request.meta.

Optional settings:

  • AVOID_DUP_CONTENT_THRESHOLD = 0.98 - minimal probability when requests are skipped.
  • AVOID_DUP_CONTENT_EXPLORATION = 0.05 - probability of still making a request that should be dropped
  • AVOID_DUP_CONTENT_INITIAL_QUEUE_LIMIT = 300 - number of pages that should be downloaded before DupePredictor is initialized

How it works

Duplicate detection is based on MinHashLSH from the datasketch library. Text 4-shingles of words are used for hashing, not spanning line breaks in the extracted text.

Several hypotheses about duplicates are tested:

  1. All URLs with a given URL path are the same (have the same content), regardless of query parameters;
  2. All URLs which only differ in a given URL query parameter are the same (e.g. session tokens can be detected this way);
  3. All URLs which have a given path and only differ in a given URL query parameter are the same;
  4. All URLs which have a given path and query string and only differ in a single given query parameter are the same;
  5. URLs are the same if they have same path and only differ in that some of them have a given param=value query argument added;
  6. URLs are the same if they have a given path and only differ in a given param=value query argument;

Bernoulli distribution is fit for each hypothesis.


License is MIT

Release History

Release History


This version

History Node

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

Show More

Download Files

Download Files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

File Name & Checksum SHA256 Checksum Help Version File Type Upload Date
MaybeDont-0.1.0-py2.py3-none-any.whl (12.3 kB) Copy SHA256 Checksum SHA256 3.4 Wheel Jun 9, 2016
MaybeDont-0.1.0.tar.gz (8.4 kB) Copy SHA256 Checksum SHA256 Source Jun 9, 2016

Supported By

WebFaction WebFaction Technical Writing Elastic Elastic Search Pingdom Pingdom Monitoring Dyn Dyn DNS Sentry Sentry Error Logging CloudAMQP CloudAMQP RabbitMQ Heroku Heroku PaaS Kabu Creative Kabu Creative UX & Design Fastly Fastly CDN DigiCert DigiCert EV Certificate Rackspace Rackspace Cloud Servers DreamHost DreamHost Log Hosting