Cluster url pattern automatically.

cluster, pattern, regular-expression, url
pip install os-urlpattern==0.1.11


os-urlpattern PyPI - Python Version PyPI

This package is used for unsupervised URLs clustering. Furthermore, it generate URL patterns(RegEx) from clusters for matching purpose. It is a pure python package tested under python2.7 python3.6, pypy can also be used for performance(4x-8x). Command line tools are provided for standalone clustering and matching, APIs are also convenient. Several extra packages can be installed for additional features. Under CPython 1cpu, 100 thousand URLs clustering cost almost 1min and 200M memory. Built-in matching strategy is efficient enough in most use cases(4k/s, depend on patterns complexity).

$ pip install -U os-urlpattern
$ wget -qO- '' | pattern-make


Similar URLs

  • URLs with the same URL structure.
  • Components of the parsed URLs at the same position are in the same character space.
  • Different types of charactors may be in the same order in most cases.

URL structure

Typically, URL can be parsed into 6 components:


Because different sites may have similar URLs structure and <params> is rare, so <schema> <netloc> and <params> are ignored, <path> <query> <fragment> are used to define URL structure.

If the URLs have the same path levels, same query keys(also keys order) and with the same fragment existence, their URL structure should be the same.

URL structure:
path levels: 2
query keys: k1, k2
have fragment: True

Character space

Consider RFC 3986 (Section 2: Characters), URL with the following characters would be legal:


There are three major character space: lower-case letters(a-z), upper-case letters(A-Z), number letters(0-9). Other symbols are in their own character space.


character space: a-z A-Z 0-9 !

Order consideration

Split a string by character, consecutive character space can be joined. In most cases, order is a distinguished feature.


split into: HELLO word 666 !

character space order: A-Z a-z 0-9 !


Complex consecutive major character space can be mixed, order is less important.


split into: H ell W orld 666 !

major join: HellWorld666 !

character space order: A-Za-z0-9 !

Because of URL quote, '%' can be mixed with major character space.


split into: % E 4 % BD % A 0 % E 5 % A 5 % BD !

major join: %E4%BD%A0%E5%A5%BD !

character space order: A-Z0-9% !

URL pattern

URL pattern is used to express each cluster. It is normal regex string. Each URL in the same cluster can be matched with the pattern.

pattern examples:


The built-in matching strategy is strict, it can't tolerate incomplet matching.

letter: helloword

pattern01: [a-z0-9]+  # not match, because no number in the letter
pattern02: [a-z]+ # match


Install with pip

$ pip install os-urlpattern

Install extra packages

subpackage install command enables
memory pip install os-urlpattern[memroy] Show memory useage
ete-tree pip install os-urlpattern[ete-tree] Enable ete pattern tree formatter


Command line

  • pattern-make

    Load urls, cluster and dump patterns.

    $ pattern-make -h
    usage: pattern-make [-h] [-v] [-i INPUTS [INPUTS ...]]
                        [-l {NOTSET,DEBUG,INFO,WARN,ERROR,FATAL}] [-c CONFIG]
                        [-f {PATTERN,CLUSTER,JSON,ETE,INLINE,NULL}]
    optional arguments:
      -h, --help            show this help message and exit
      -v, --version         show program's version number and exit
      -i INPUTS [INPUTS ...], --inputs INPUTS [INPUTS ...]
                            input files to be processed (default: stdin)
                            log level (default: NOTSET)
      -c CONFIG, --config CONFIG
                            config file
                            output formatter (default: CLUSTER)

    Dump clustered URLs with patterns:

    $ cat urls.txt | pattern-make -L debug > clustered.txt

    Only generate URL patterns:

    $ cat urls.txt | pattern-make -L debug -F pattern > patterns.txt

    Generate pattern tree from URLs(ete installed):

    $ cat urls.txt | pattern-make -L debug -F ete
  • pattern-match

    Load patterns, dump URLs matched results.

    $ pattern-match -h
    usage: pattern-match [-h] [-v] [-i INPUTS [INPUTS ...]]
                         [-l {NOTSET,DEBUG,INFO,WARN,ERROR,FATAL}] -p
                         PATTERN_FILES [PATTERN_FILES ...] [-a]
    optional arguments:
      -h, --help            show this help message and exit
      -v, --version         show program's version number and exit
      -i INPUTS [INPUTS ...], --inputs INPUTS [INPUTS ...]
                            input files to be processed (default: stdin)
                            log level (default: NOTSET)
                            pattern files to be loaded
      -a, --all-matched     all matched patterns

    Match URLs:

    $ cat urls.txt | pattern-match -L debug -p patterns.txt


  • Cluster and generate URL patterns:

    from os_urlpattern.formatter import pformat
    from os_urlpattern.pattern_maker import PatternMaker
    pattern_maker = PatternMaker()
    # load URLs(unicode)
    for url in urls:
    # cluster and print pattern
    for url_meta, clustered in pattern_maker.make():
        for pattern in pformat('pattern', url_meta, clustered):
            # do whatever you want
  • Match URLs:

    from os_urlpattern.pattern_matcher import PatternMatcher
    pattern_matcher = PatternMatcher()
    # load url_pattern(unicode)
    for url_pattern in url_patterns:
        # meta will bind to matched result
        pattern_matcher.load(url_pattern, meta=url_pattern)
    # match URL(unicode)
    for url in urls:
        matched_results = patterm_matcher.match(url)
        # the best matched result:
        # sorted(matched_results, reverse=True)[0]
        patterns = [n.meta for n in matched_results]
  • Low-level APIs:

    It is necessary to use low-level APIs for customizing processing procdure, especially for parallel computing or working on an distributed cluster(hadoop).

    Key points: same fuzzy-digest same maker and same matcher.

    Use os_urlpattern.parser.fuzzy_digest to get fuzzy digest from URL, URL pattern or URLMeta and parsed pieces/patterns.

    A brief All-In-One example:

    from __future__ import print_function, unicode_literals
    from os_urlpattern.formatter import pformat
    from os_urlpattern.parser import fuzzy_digest, parse
    from os_urlpattern.pattern_maker import Maker
    from os_urlpattern.pattern_matcher import Matcher
    urls = ['' % i for i in xrange(0,10)]
    makers = {}
    matchers = {}
    # Init makers from URLs(unicode).
    for url in urls:
        url_meta, parsed_pieces = parse(url)
        # same digest same maker
        digest = fuzzy_digest(url_meta, parsed_pieces)
        if digest not in makers:
            makers[digest] = Maker(url_meta)
    # Iterate makers, do clustering, generate URL pattern and init matchers.
    for maker in makers.values():
        for clustered in maker.make():
            for pattern in pformat('pattern', maker.url_meta, clustered):
                # init matchers
                url_meta, parsed_patterns = parse(pattern)
                digest = fuzzy_digest(url_meta, parsed_patterns)
                if digest not in matchers:
                    matchers[digest] = Matcher(url_meta)
                matchers[digest].load(parsed_patterns, pattern)
    # Match URLs(unicode).
    for url in urls:
        url_meta, parsed_pieces = parse(url)
        # same digest same matcher
        digest = fuzzy_digest(url_meta, parsed_pieces)
        if digest in matchers:
            matched = [n.meta for n in matchers[digest].match(parsed_pieces)]
            print(url, *matched, sep="\t")
        else: # no matched at all

Unit Tests

$ tox


MIT licensed.