Document tree flattening as an exploration technique for data mining .xml files (sitemaps, feeds, inventories, raw data, etc)

Tags

Data Mining, Feeds, Flatteners, minerazzi, Sitemaps, tools, URLs Extraction, web feeds, XML Mining

Two of our tools, Web Feed Flattener and Feed URLs Extractor, were updated and now accept files with the .xml extension so we changed their names to indicate this. These tools are available at

http://www.minerazzi.com/tools/flattener/feed-flattener.php

http://www.minerazzi.com/tools/feed-urls/extractor.php

These updates take the tools to a whole new level. Now you can flatten the tree structure of files like sitemaps.xml and similar files and extract URLs. Just submit a  target web address and you are good to go.

I know there are tools out there that can scrape .xml files in order to extract specific pieces of data like URLs, but found them too cumbersome. A major drawback of said design alternatives is that frequently one must know in advance how the document tree was constructed, with all of its tags and nuances, before coding a tool. To top off, if the author of the file changes or edits tags, probably the tool won’t work as expected.

Our approach is different and very flexible. The key here is the flattening of the document tree structure embedded in XML files without even having to know how it was designed or edited. Document tree flattening will unveil this information before you can say: “Give me some soup!”

Of course, we assume that the document tree has no orphan or broken tags (and better, pass validation) which is something to be expected from trusted sources. If it is not valid, well, there are ways of fixing it or ignore the offenders.

With the proposed technique, we can mine all sort of .xml files and build customized tools on top of the flattened results, like derivative tools for mining sitemaps, inventories, raw data, recipes, etc… No need to know in advance anything about the document tree, resource to additional scripting technologies, software, or reinvent the wheel.

Right now we can mine sitemaps all over the Web, including sitemaps hosted at Google, W3C, company sites, etc, and then recrawl the output to grow a microindex. See “Suggested Exercises” sections of the tools for interesting examples. This is a value-added approach for our Maps2Miners ongoing project.

Considering that there are government agencies and organizations facilitating data in .xml format for developers to mine, flattening .xml files and build on top of these is one of those “ah-ha!” ideas.

 

URL Cleaner: Clean URLs from search results and websites

Tags

Data Mining, Search, search engines, tools, URL Cleaner, URL Tools

The URL Cleaner (http://www.minerazzi.com/tools/url-cleaner/muc.php) is our most recent tool.

Clean URLs from search engine result pages and websites, including Google, Bing, Yahoo, Yandex, Wikipedia, and others.

Introduction

• The Problem
  Sometimes collection curators and content developers use web scrapers (Wikipedia, 2018a) to extract URLs from websites and search result pages.If a web scraper is not available or the target search engine reacts against the scraping (Wikipedia, 2018b), URL extraction can still be possible by installing a browser add-on like Copy Selected Links or a similar plugin. Once installed, users can right-click selected text and copy the URL of any links it contains. To copy all links from a page, they just need to press Ctrl + A to select the entire page text, right-click the selected text, and copy all available URLs at once.Regardless of how URLs are collected (with or without web scrapers), the end result might be a list of dirty, ugly records with obscure attribute-value pairs appended by the search engine.Sometimes the list of URLs include entries with:
  • • URLs pointing to social networks. These URLs are often viewed by collection curators as “plastic contamination” in search results suppose to be “organic”. Typical examples are results from Google and similar search engines.
    • URLs about self-promotion. The same search engine might include URLs pointing to unrequested content like its own products, services, partners/ads, links to additional content, etc. Typical examples are results from Google and URLs extracted from Wikipedia webpages.
    • URLs with special characters. For instance, those defining queries (?), fragment identifiers (#), and hash-bangs (#!), among others (Wikipedia, 2018c; 2018d).
    • URLs with some characters encoded.
    • URLs containing mailto:, javascript:, or data:
    • 6-22-2018 Update: URLs obfuscated by shortening services: e.g., bit.ly, goo.gl, is.gd, t.co, and many more. Regardless of their merits, shortened URLs can open the door to all sort of problems (Wikipedia, 2018e). These are frequently viewed by collection curators as unnecessary noise.
• The Solution
  Would it be nice to have a tool that lets users generate by default a list of clean, sorted, and deduplicated URLs, with options for selectively include/exclude some of the above contaminants? This is precisely what our Minerazzi URL Cleaner (MUC) does.
• Unlike other URL cleaners, MUC cleans multiple URLs at once from search engines and websites, and can be used free of charge. Before proceeding any further, lets explain what MUC is and is not. The tool is a data cleaner and a lightweight version of our popular Editor and Curator tool. It is not a web scraper, URL validator, or URL shortener resolver, but can be used to clean results from these.
• In the next section, we describe some uses for MUC, its features and limitations.

What is computed?

• Searches Support
  MUC was designed to edit search results from the following.
  • Google, Bing, Yahoo, Yandex, and DuckDuckGo
  • 100searchengines, HotBot, Ask, and textise.net
  • Google Scholar, and Wikipedia

  The tool is compatible with individual sites and might be so with other search engines. Whenever possible, we are open to add support for other search engines as suggested by users.

• Editing Features
  The tool implements the following edits by default.
  • Social networks
    URLs pointing to Linkedin, Facebook, Twitter, Myspace, Instagram, Pinterest, Snapchat, Youtube, Vimeo, and Tumblr are removed.
  • Self-promotions
    URLs about the supported search engines and pointing to their products, services, and partners/ads, or any additional content are removed.
  • Special characters
    Sections of a URL that start with ? # [ ] @ ! $ & ‘ ( ) * , ; = are removed. Trailing forward slashes (/) are also removed.
  • Special strings
    URLs with mailto:, javascript:, data: are removed.
  • Encoded characters
    URL %-encoded characters are replaced by their unencoded versions.
  • Shorteners (6-22-2018 Update)
    URLs obfuscated by shortening services (nearly 600 of these services) are removed.
  • One or more of the above edits can be disabled by properly checking the corresponding checkboxes.

  Since these features are enabled by default, if a run produces no results it means that either all URLs are fully contaminated or there are no URLs to edit.

• First time users
  We recommend first time users install the Copy Selected Links , or a similar add-on, before proceeding any further. Then do a search in Google and, with the add-on installed, clean URLs, first selectively and then at full blast, MUC.
• Tool limitations
  Up to 5,000 URLs can be submitted at once. We arbitrarily imposed this limit to (a) provide fast responses, (b) minimize browser crashes, and (c) minimize abuses.
• Last but not least, the tool might fail to remove non English, obfuscated, or encrypted characters.

 

The URL Query Parser

The URL Query Parser is our most recent tool for mining URLs. It is available at

http://www.minerazzi.com/tools/url-query/parser.php

What is a URL query?

A URL query is the trailing text after the question mark (?) found in a URL. It consists of attribute-value pairs delimited by ampersands (&). These are also called name-value, key-value, or field-value pairs.

What this tool does

This tool parses URL queries and extracts its name-value pairs.

The tool helps users identify and filter URL queries from a collection or build collections consisting exclusively of URL queries.

With minor modifications, the tool can be converted into a massive URL cleaner. We are currently building another tool that does this, precisely. In this way we may be able to clean up URLs found in Google and Bing search result pages and safely use them in data mining studies.

What is computed

• Up to 5,000 URLs can be parsed. If no query is found in a URL, that record is ignored.
• We have arbitrarily imposed the 5,000 limit for several reasons: to (a) provide fast responses, (b) minimize browser crashes, and (c) minimize abuses.
• Users can opt between two query result modes:
  • individual results (useful for comparing individual URL queries).
  • combined results (useful for comparing specific name-value pairs).

  The latter is the default mode. Since in this mode results are alphabetically sorted, users can easily identify the most common or popular name-value pairs.

Implications to Web Security

This tool can be used by those interested in mining URL queries or conducting studies relevant to Web Security. Why? Please keep reading.

URL queries are used to transmit small pieces of data in the form of name-value pairs. The transmission can be of three types: (a) between web pages, (b) between a web page and a database, or (c) between databases. Real-world applications include access to web services, social profiling, and cloud computing, among others (Kantarcioglu, 2013).

In addition, URL queries are frequently used as vehicles for transmitting session parameters, form data, tracking mechanisms, user names, email addresses, and other data considered sensitive by users.

In a 2014 study, West & Aviv, from Verisign and the US Naval Academy, analyzed over 892 million user-submitted URLs containing 1.3 billion name-value pairs. They found over a quarter-billion plain text pairs involving referral tracking, with more than 10 million pairs potentially revealing some form of demographic, identity-based, or geographical information. Extreme cases involved the facilitation of password authentication credentials, email addresses, and user names (West & Aviv, 2014).

Thus, the development of tools designed to mining URL queries is something relevant to Web Security.

Suggested Exercises

• Do a search in several search engines or public databases. Collect a set of URL queries and submit this set to our tool. Compare results.
• Analyze a set of URL queries obtained from a public forum, social networks, or groups (e.g. Google Groups).
• For this exercise you need to install a browser add-on to facilitate collection of URLs. In Firefox for instance you can install an add-on plugin that lets you selects multiple links and copy their URLs. Do a search in Google or similar search engines and with said add-on collect search result URLs. Submit these URLs to our tool. Compare results. This is a nice way of grabbing intelligence from URL queries relevant to specific search terms. In addition, since Google lets you do advanced field-specific searches (e.g., inurl, intitle, etc), this is a nice way of mining URL queries driven by advanced searches.

References

• Kantarcioglu, M. (2013). BigSecret: A Secure Data Management Framework for Key-Value Stores. See also these Researchgate or Github entries.
• West, A. G. & Aviv, A. J. (2014). On the Privacy Concerns of URL Query Strings. In W2SP’14: Proceedings of the 8th Workshop on Web 2.0 Security and Privacy. San Jose, CA, USA. May 2014.