These were the guidelines given to the volunteers that labelled the UK-2006 dataset.

General definition of Web spam:
«any deliberate action that is meant to trigger an unjustifiably favorable [ranking], considering the page's true value» (Gyöngyi and García Molina 2005).

When classifying, ask this question: are there aspects of this page that are mostly to attract and/or redirect traffic?. We are interested in detecting Web sites that employ spam techniques.

Typical Cases of Web Spam (tag as: SPAM)
Pages that are full of keywords, even if they include actual contents.
	Just links with many different keywords.	Just links with many different keywords.	The archive of a mailing list, copied to produce more keywords.	Bottom: hidden text (white text on white background, use CTRL-A to select all text, to see this type of spam).
Pages that are only advertising, with very little content. This includes automatically generated pages designed to sell advertising. This also includes sites that offer catalogs of products that are actually redirecting to other merchants, without providing extra value.
	Left: ads. Right: keywords.	Left: ads, with ratings probably auto-generated. Right: keywords.	Left: ads. Right: more links to other pages fulls of ads.	Fake search engine that display always the same results, no matter what the query is.
	Top and right: ads. Bottom: outdated "news" copied from news sources.	Fake search engine showing only ads for every query.	Another fake search engine showing only advertising.
	Left: automatically generated links. Main: one line of text (probably copied) about the topic, and a page full of advertising.	Top navigation, left navigation, and basically all of the contents of this page are advertising.
Pages with unrelated links, or exchanging links with too many different, unrelated, partners. Some of them may also be classified as BORDERLINE if they provide some extra value.
	Top: keywords and links. Bottom: links.	Bottom: links to Web sites made by this company. This may also be considered borderline, depending on the ratio of original content vs repetitive linking.	Left: links to unrelated sites. Bottom: keywords. This may also be considered borderline if the page also offers useful content.	Repetitive linking to the internal pages of the site.
Borderline cases (tag as: BORDERLINE)
Pages that are heavily optimized for search engines or for selling advertising, but that also provide some content.
	Links to several of its partner sites, but all the sites are topically related. Judge based on the (content for humans/content for search engines) ratio.	Heavily optimized porn site (but not all porn sites should be considered spam, unless they use spamming tricks).	One paragraph of text; the rest is only ads. The text is stolen from Wikipedia (if you detect this, mark as SPAM), but as this is not easy to check, you can tag this page as BORDERLINE.	One paragraph of text and the rest of the page is only advertising. The paragraph provides little added value and it might be argued that the Web site is spam. Tag this type of site as borderline if you are suspicious about the content/ads ratio.
Pages offering search engine optimization services, random link exchange, affiliate links, etc.
	Page offering to participate in a link exchange program, as it is very likely part of a link farm.	Page offering original content, a few links to unrelated sites (this is a mild link-farm), and a link to affiliate programs (this is hard to detect, can be tagged as normal).	Left: links to various unrelated sites, plus links to participate in affiliate programs. This can be normal, borderline or spam depending on the actual content/link ratio.
Web pages that are not spam (tag as: NORMAL)
Pages that do not use Web spam tricks. Quality is not an issue here: these can be high-quality or low-quality resources. The important thing is that they do not use Web spam techniques.
	Website with contents. Even when the title has many keywords, they are all related to news.	On-line forum. Even when this has lots of pages and little content per page, it provides a service to users.	Shopping catalog, as long as the products are sold by the same merchant (like Amazon does), or by different merchants (like eBay does). However, if the page is always acting as a facade/redirect to one specific store, then it is borderline or spam.	On-line directory. This has many links but this is OK if the links are carefully, manually selected by human editors.
Web pages that cannot be classified (tag as: DON'T KNOW)
	Web sites that you cannot access, or require a password. Web sites that you do not know how to classify. "Parked" domains with a single page (sometimes they show a "this domain is for sale" message with lots of links).

• We are interested in finding hosts that use spamming techniques, so if you find one such a page in a host it is correct to classify the entire host as SPAM or BORDERLINE.
• If you are unsure if something is SPAM or BORDERLINE, choose BORDERLINE. If you are unsure if something is BORDERLINE or NORMAL, choose BORDERLINE.
• In the case of mirrors (copies of a host), give the same rating to the mirror as to the main (target) host.

What search engines consider as Web spam

Google

[Source] Webmasters should ...

• Avoid hidden text or hidden links.
• Don't employ cloaking or sneaky redirects.
• Don't load pages with irrelevant words.
• Don't create multiple pages, sub-domains, or domains with substantially duplicate content.
• Avoid "doorway" pages created just for search engines, or other "cookie cutter" approaches such as affiliate programs with little or no original content.
• If your site participates in an affiliate program, make sure that your site adds value. Provide unique and relevant content that gives users a reason to visit your site first.

Yahoo!

[Source] Spam pages are ...

• Pages that harm accuracy, diversity, or relevance of search results
• Pages that seem deceptive, fraudulent, or provide a poor user experience
• Pages dedicated to redirecting the user to another page
• Pages in great quantity, automatically generated or of little value
• Pages using methods to artificially inflate search engine ranking
• The use of text that is hidden from the user
• Excessively cross-linking sites to inflate a site's apparent popularity
• Pages built primarily for search engines
• Misuse of competitor names/products

MSN Search

[Source] Webmasters should avoid ...

• Loading pages with irrelevant words in an attempt to increase a page's keyword density. This includes stuffing ALT tags that users are unlikely to view.
• Using hidden text or links. You should use only text and links that are visible to users.
• Using techniques to artificially increase the number of links to your page, such as link farms.

Help for the classification

The interface for the classification is divided into three panels: the left panel is for classifying pages, the center panel is for displaying information about the selected host, and the right panel is for previsualization.

Left panel

These are the controls for the left panel:

Center panel

In the center panel, the following information is shown:

• The number of pages in the host and a list of a few sample pages.
• The in-links with the link-based PageRank contribution (PageRank/degree) of every host linking to the target host, plus the judgments on the linking hosts (if any).
• The out-links, same as above.

Also, the following extra information is shown. You should use this extra information with care, do not base your decision solely on Google's PageRank or Alexa's TrafficRank as those can be deceived by the Web spammers we want to detect:

• The link-based PageRank of the current host.
• Google's PageRank.
• Alexa's TrafficRank.
• WHOIS information about the person or company that registered the corresponding domain name.
• A thumbnail of the site as seen by thumbshots.org, when available.

Right panel

On the right panel, a preview of the Web site is shown. We display the home pages of each host as they were seen at crawling time. If you continue navigating from there you'll be taken to the current version of the pages.

Hints