Machine Learning features & logic

Problem

More actionable

Blogalyzer v1 is a visual & intuitive tool that helps marketers make better decisions around what to write next and how to distribute their content. However, they still need to interpret the data themselves. To make it more actionable, we want to build a recommendation system.
 

More competitive

Today, A.I. or at least Machine learning, gets integrated in a lot of products. Blogalyzer has a lot of data that can train the algorithms, and to stay competitive in the market of content marketing and SEO, we need to offer recommendations.

First steps

Selecting experienced partners

I’ve worked together with an experienced subsidy consultant who specializes in A.I. projects (Lemon Companies) as well as with a consultancy (NLP Town) focused on Natural Language Processing. Together we brainstormed about the plans I had, what was technically possible and how to structure the development of the Machine Learning feature.
 

One model to rule them all

The result is a holistic approach that will look at all the elements that influence the reach and conversion of a content piece: writing effort, distribution effort, page design, domain authority & owned audience. By comparing these elements’ influence to a large set (1.000.000) of blog posts and their traffic and conversion performance, the algorithm will be able to recommend changes to an existing blog post or even recommend first steps for a new article.

Will it work?

Feeding the algorithm

We have a very large set of articles or blogposts, but there's a lot of variables within this set that determines the performance of each item: the length of the blogpost, the category and topic of the blogpost, the readability score, the seo effort, the existing audience, the domain authority, the distribution channels, the website speed,... to name a few. We could easily identify 30 major elements, that can have 1000's of different values. This means 10.000.000 blogposts might not be enough.
 

Will results be statistically significant?

The big challenge is whether the results will be "strong" enough. Meaning: will the model be able to say with a high enough certainty that one or multiple elements can boost the article's performance.
 

Unsure

First we need funds! The document describing the logic and workflow to build the M/L model, with the data collectors and U.I. features is work in progress. To apply for grants, Blogalyzer needs investment first. The full process and development might take 18 months. We'll keep you updated! =)