This new approach seems remarkably resistant to existing pagerank manipulation techniques. Given a preference-vector, it stays “on topic” remarkably well.
That’s unsurprising. But is it resistant against cosine similarity manipulation techniques?
PageRank is insensitive to the topics of linking sites, and is only concerned with popularity. This made comment spam and sales of high-profile links work well.
But if you also require links to come from specific sets of on-topic sites, then random blogspam won’t work any more. A janky content farm linking to garbage may end up clustered with garbage.
I’d be interested in learning more on this new approach. This article only seemed to discuss weaknesses of the previous approach, but perhaps my scan was too cursory?
That’s unsurprising. But is it resistant against cosine similarity manipulation techniques?
It might be harder to manipulate?
PageRank is insensitive to the topics of linking sites, and is only concerned with popularity. This made comment spam and sales of high-profile links work well.
But if you also require links to come from specific sets of on-topic sites, then random blogspam won’t work any more. A janky content farm linking to garbage may end up clustered with garbage.
I’d be interested in learning more on this new approach. This article only seemed to discuss weaknesses of the previous approach, but perhaps my scan was too cursory?