1
$\begingroup$

Given a webpage url and the extracted article text from the given page, I want to calculate a probability of the article to be content marketing.

I consider content marketing as differentiated from spam, because it is to admit, content marketing makes some serious efforts to deliver real information to the reader. But the true reason behind the effort of writing an article in the context of content marketing is to sooner or later guide the reader to a newsletter subscription, a product landingpage or to contact the author. In the language of content marketers this kind of qualified reader is considered a "lead".

1. call to actions

I could by using Knn or Bayes simply test the scentences in the article text against a training set of known CTAs.

Okay, this where NLP comes in and this is all nothing else than spam detection.

2. link structure

I possibly can find strong evidence on content marketing motivation, in the outgoing links of a certain page or article. If it links to a landing page styled product page (some heuristics to identify these kind of webpage are needed as well) and a lot of other sites of the website link to a "near by" page.

By "near by" I mean text similarity, optimized for a (and the same) keyword.

question So what do you think about this? How could scoring look like. How should a probability model look like? Are there aspects, that I miss? Does anyone have some experiences in this field? Is a paper out there? Does anyone know some resources for this?

$\endgroup$
1
$\begingroup$

If you have a broad set of testing data, I think this is feasible. I've had luck in getting basic models to identify abstract concepts like someone's politics or real news vs fake news, so I think this could work if people write content marketing different from normal news.

I just used this link in another post, but here is a tutorial in Python of taking unstructured articles, passing them through an NLP pipeline, and classifying them into multiple groups.

$\endgroup$
  • $\begingroup$ Thank you. But unfortunately, the NLP part is not the big issue. The probability part for the link structure is it. $\endgroup$ – Chris Pillen Jun 5 '17 at 21:27
  • $\begingroup$ Are you saying you have a list of certain sites that are highly correlated with the class you want to predict? It sounds like those should be extracted as features. $\endgroup$ – CalZ Jun 6 '17 at 11:14
  • $\begingroup$ No, the main idea is, that a website will have several kinds of pages. One of these are lead pages. A lead page is a page which produces leads. A landing page, a product page, a contact page. A page that promotes something. This page is different from an article page. This page contains an blog oder newspaper styled article. If the article however links to such a lead page it is likely, that the article is part of sales funnel, which in return is part of inbound marketing. Now, a both site on the same domain? No? How near are they in terms of linkage? $\endgroup$ – Chris Pillen Jun 7 '17 at 13:13
  • $\begingroup$ That all together should lead to probability. Bayes? $\endgroup$ – Chris Pillen Jun 7 '17 at 13:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.