A classifier for detecting soft 404 pages
A “soft” 404 page is a page that is served with 200 status, but is really a page that says that content is not available.
pip install soft404
The easiest way is to use the soft404.probability function:
>>> import soft404 >>> soft404.probability('<h1>Page not found</h1>') 0.9736860086882132
You can also create a classifier explicitly:
>>> from soft404 import Soft404Classifier >>> clf = Soft404Classifier() >>> clf.predict('<h1>Page not found</h1>') 0.9736860086882132
Classifier is trained on 120k pages from 25k domains, with 404 page ratio of about 1/3. With 10-fold cross-validation, PR AUC (average precision) is 0.990 ± 0.003, and ROC AUC is 0.995 ± 0.002.
Getting data for training
Install dev requirements:
pip install -r requirements_dev.txt
Run the crawler for a while (results will appear in pages.jl.gz file):
cd crawler scrapy crawl spider -o gzip:pages.jl -s JOBDIR=job
First, extract text and structure from html:
./soft404/convert_to_text.py pages.jl.gz items
This will produce two files, items.meta.jl.gz and items.items.jl.gz. Next, train the classifier:
Vectorizer takes a while to run, but it’s result is cached (the filename where it is cached will be printed on the next run). If you are happy with results, save the classifier:
./soft404/train.py items --save soft404/clf.joblib
License is MIT.
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.