Ideas/suggestions for language processing beyond spin?

Sup BHW,

I'm potentially interested in developing NLP tools for blackhat SEO and I just wanted to hear your opinions, suggestions or requests. I'm talking about the language-y side of things, can't really do much about scraping itself or your spouse's looks I'm afraid.

Modern-day approaches to AI make it possible to, say, create an algorithm which would write novel blog or website comments in the style of the target website which could pass the Turing test... but the question is, would anyone actually need this? Autoblogging, for example, is apparently dead, Penguin is wrecking havoc, etc. I've had a look at existing tools and there's plenty of interesting stuff, 99% is only spin though (I admit WordAI is the only tool to really impress me in this area), some content building (Article Builder etc.), plus bots and YT comment posters that don't really do very much about original content.

In particular, I'm wondering about the usefulness of stuff like the following:

1. Document generation: creating an entirely new article resembling a given set of existing articles, whether websites, product reviews, YT comments or tweets. Content Builder and WordAI kind of do that, but they're both limited to (respectively) a fixed set of keywords and a specific input text.
2. Keyword insertion: given article A, rewriting A to include keywords X,Y,Z in an inconspicuous way. There are some tools that do this, but they all produce unthinkable cludge. Alternatively, given keywords X,Y,Z, generating a new unique text built around them, whatever they are.
3. Alternatives to spin(tax): instead of modifying text, it's very well possible to (re)construct sentences from the bottom up using elements of existing sentences.

Any of this sounds worth investing in, or do you have any better ideas for language processing tools that feel missing?

On a final note: I'm a bit baffled how many devs use only Copyscape for evaluation. Let me assure you, there are way more efficient widely used methods of spotting plagiarism than looking at word uniqueness.

contextual and human readable sentence generation should be priority.
NLP/NLG is not yet matured to generate human readable articles.

contextual and human readable sentence generation should be priority.

Click to expand...

Thanks for the reply. Where would it fit into SEO tactics, though?

NLP/NLG is not yet matured to generate human readable articles.

Click to expand...

That's definitely true for meaning-to-text (where you try to express an exact abstract meaning), but not necessarily text-to-text methods (where you generate text from existing data based on a set of constraints or principles).

It's well possible to simulate a coherent article, though it also depends on the type of content. I mentioned making a script for that in another thread, so have an example:

In 1969, Black Sabbath single-handedly created Heavy Metal with their self-titled debut album. And so, ?Paranoid? saw the light of day and remains to this day, one of the greatest Heavy Metal album of all time. Every second of this album is perfect. Super classic ?War Pigs? is amazing, as is ?Paranoid?, ?Planet Caravan? and ?Hand of Doom?. They reach down into the very pits of their souls - and summon a work of art so majestic, it defies suitable descriptions. Suddenly you realise how well the band?s different musicians complemented each other, creating a drum/bass/organ/guitar/vocal noise which was exciting, an adrenaline rush like no other.

The above "article" is entirely strung together from sentences taken from different Amazon reviews (you can google individual sentences to see). It's far from perfect, but generating random convincing short text that adheres to a certain structure isn't all that hard.