Recall in search, we could “manipulate” the algorithm by including the same word count lots of times or using many synonyms to increase the chance of getting a search term.

Thus, we use the link text (and surrounding text) as terms instead of the actual page contents.

Forum/Comment Spam

The above approach has the issue that you can go to pages that allow posting and link your webpage (i.e. YouTube, Facebook, CBC News, etc.). As these pages have high rank, you also have high rank ⇒ you can choose terms that your web page gets.

Spam Farming

Recall that “spider traps” will accumulate rank. Random jumps prevent them from accumulating all rank, but it is still boosted by the topology.

Spam Farming involves the following technique:

• Page you want to promote has millions of hidden links to “farm pages” - accumulate random jump weight
• “Farm pages” link back to the page you want to promote (rank sent back to page to be boosted)

• more websites you can insert links ⇒ better
• Consider how the rank of a page grows via this technique:
  • $x =$ total contribution of accesible pages to $t$
  • $y =$ page rank of $t$ (boosted page)
  • $z =$ page rank of a farm p
  • We have that $z = \frac{\beta y}{M} + \frac \alpha N$, so $y = x + M\beta z + \frac \alpha N$
  • Substituting $z$, we have that $y = \frac{x}{1 - \beta^2} + \frac{\beta M}{(1 + \beta )N}$ - rank can essentially grow linearly with $M$, $\alpha / N$ ignored.

As a start to solving this issue, we ignore links tagged as “nofollow”, and convince forums, new sites, etc. to insert “nofollow” to all links in comments.

• This has an added benefit of allowing a researcher (i.e. on university website, high rank) to link to a page and not boost its ranking
• This does not entirely solve the problem, as the rank is still directly proportional to $M$ (number of farm pages)

Trying to identify the “farm pages” is difficult (i.e. small pages not allowed ⇒ can be made large enough to count, force small pages to include more).

Personalized Page Rank