How to Use No-Code Machine Learning to Optimize your HackerNoon Articles

Improving the reach and success of your writing is no small feat, and machine learning can play a large role in your process.

If you are creating content with the goal of engaging audiences and driving clicks, this article is for you. Modern Machine Learning (ML) tools make it possible to analyze engagement data from past articles so you can make informed decisions about what works and what does not, giving your content a competitive edge.

Let’s explore how to utilize machine learning to predict the success of articles before you even hit "publish."

Gather Your Data

Whether you’re a new writer or an experienced journalist, you’re likely publishing your article on multiple sites, including our beloved Hackernoon.

When writing for a publication that’s not your own, you miss out on all the data and brainstorming you typically have available when submitting to your own blog.

Without access to data and insights from the publication, it may be difficult to determine which topics are most popular on Hackernoon, the number of links required for maximum exposure, or the reputation needed for a successful article.

To get started, we need to find this data.

Unless it’s a very small publication, any modern SEO tool will be able to provide you with the top pages (a.k.a the top articles) for any given website. Let’s use Hackernoon as our main example.

To figure out the top articles on Hackernoon, you can use something like SpyFu, SEMRush, or Ahrefs. Each tool has its own metrics and will serve you well. Beginners may want to try SpyFu for its affordable entry pricing.

I already did that for you and found the top 3000 articles on Hackernoon. If you want to download the list, you can do so here.

Download the list (Datasets for Hackernoon - Google Sheets)

Once you have the source URLs, titles, and engagement statistics, it’s time to move to the next step.

Enrich Your Data

URLs alone won’t give you much information about an article and won’t be enough to teach you how to write for success in 2023. You need more data. You need enrichment.

You can enrich your URLs in any number of ways, but the most valuable ones probably include other SEO Metrics, such as word count, backlink profile, and page authority.

Ask yourself these questions to understand what you are trying to get out of the prediction:

• What are you trying to predict? Is it the title? The length of the article? The number of keywords it will rank for?
  
• What data do you have available right now? Any data that's relevant to your engagement objective can help improve the accuracy rate of your prediction
  
• Keep in mind that once your ML model is ready, you can use it to predict success on the target publication whenever you want. A Hackernoon Prediction Model today will serve you well for a long time unless Hackernoon undergoes a significant audience change (or a new set of hyper-successful articles skew the data).

In this case, our goal is to optimize for the average number of views an article receives on Hackernoon based on the title, backlinks, and keywords.

I found all of this data readily available in MajesticSEO, a popular solution for backlink research. You can use some free alternatives if you’re on a budget, like SEOGlass or Ubersuggest. They won’t get granular, but you’ll still be able to create your dataset.

After the enrichment, the example dataset now features:

• Name of the article
• URL
• Avg number of backlinks (internal and external)
• Citation Flow
• Trust Flow

You can download the enriched dataset for free here. (Datasets for Hackernoon - Google Sheets)

Train Your No-Code Machine Learning Model

Machine learning used to be complicated, and you needed a data scientist to come up with any sort of reliable model. Nowadays, that’s no longer the case.

The pairing of new technologies like AutoML, ChatGPT, and DALL-E2 with visual interfaces, was recognized in 2022 with the emergence of AI-powered tools. This led to the development of no-code machine learning.

Akkio and other companies offer automatic machine learning models that can be quickly and easily trained without any coding or data science knowledge. With these tools, anyone around the world can upload their datasets and predict future trends, forecast data, and detect anomalies.

No-code machine learning is now being used by companies worldwide for various purposes, including detecting defects in electronic products like computers, analyzing customer feedback to improve products, streamlining the research process for new pharmaceuticals, and predicting user churn at the time of product registration.

That stated, let's get back to the task: we have an enriched database of Hackernoon articles, with the relative success metrics we came up with in step 2. If your success metrics are different, you will have to tweak your initial database accordingly.

If you want to follow along and/or train your own model, Akkio and the other tools mentioned above all offer free trials.

1. Upload the dataset

   You can upload the dataset to Akkio through the Google Sheet integration, a direct CSV upload, or any of the built-in CRM and marketing integrations. Let’s do it with a manual upload for now.

This is how the Akkio platform will appear once it’s done:

Together with all your columns, aggregations will help you understand some trends in your data from the get-go. For example, you can see the distribution of SEO clicks throughout the articles, clearly outlining how the vast majority of articles get less than 100 SEO Clicks, while some can be hyper-successful.

You want to be within the top 5%, right? So, let’s train the model to help you get there!

2. Clean up any discrepancies

Proper data preparation is essential for successful machine learning model training. If you prefer using Excel, it can be done there. Alternatively, Akkio offers a comprehensive data preparation toolkit that will remind you of ChatGPT.

This allows you to simply type out your desired edits, and the platform's dataset will reflect them.

In my example, I asked Akkio’s Chat Data Prep to:

• Remove all articles with less than 30 views;
• Remove articles with more than 3200 views;

This is to make sure we avoid anomalies in the dataset. There were only a few articles reaching 3200+ views, and a very large number of likely irrelevant articles with less than 30. Removing these outliers helps the model focus its attention on the driving factors that matter most to our outcome.

You can clean the data as you see fit for your use case. You might need to combine columns and change the format of certain values to have a uniform format throughout the set (eg. dates are a common one).

3. Run the Model

For this use case, we want our machine learning model to predict the success of our new article based on successful articles of the past. This is conveniently referred to as a “prediction” model on the Akkio platform.

A prediction model simply requires you to pick a column, and the algorithm will do the rest. All you need to do is sit back and watch the model train itself.

At the bottom of the screen, you can decide how fast you want the model to get to the bottom line. I suggest you start with “Fast” (it usually requires less than a minute) and move to a longer training time once you are done with any model iteration.

Before you click predict, double-check that you selected the correct value to predict. I want to predict the potential for “SEO Clicks”, so that’s my prediction target.

Please note that Akkio, and other no-code tools, usually offer a wide variety of machine learning models. For a number prediction like our example, the tools will use a process called AutoML to pick the best performing model. These platforms also offer time-series models and other ML tools. Depending on which publication you’re pitching to, and what you’re writing about, you might consider other modeling approaches.

For example, if you want to perform some in-depth research for a new article, you can use these platforms to detect all anomalies in huge datasets (with even millions of records, maybe coming from Wikipedia!), forecast trends (you can see an example of a website forecasting product success on ProductHunt here). Assuming you have some data, the possibilities are pretty much endless.

If you’re interested in learning more about this, Akkio has a full page with supporting documentation for all the major applications. Spoiler alert: one section discusses Twitter Sentiment Analysis and a feature that’s been accessible to only data scientists for years!

4. Analyze the Results

   Once the model is done training, you will see an “Insight Report”, detailing all the incredibly interesting patterns in the data that are driving your selected outcome..

It will let you know what factors are associated with positive and negative outcomes. You will find many, many golden nuggets here. Reading through the insights report and studying the driving factor graphs will help you create your content marketing strategy for the publication!

The report also shows the accuracy level. Because our dataset has a large range of possible click values and is heavily weighted towards the low end of clicks, we get an accuracy that is usually within 55% - but the statistic to pay attention to here is the “off by” of 88 clicks. That's actually pretty good for our purposes!

If your model is not accurate enough you usually need to get more examples or more relevant input columns to improve its accuracy - but almost any model will have a lot of directionally correct information you can leverage to improve your results.

5. Deploy the Model

You can deploy the model in a variety of ways: API, web app, Zapier, and direct integrations of Akkio.

I suggest you start with the web app, as it requires zero setup and can be shipped live in less than 2 minutes.

The web app is auto-generated, and allows for bulk predictions of uploaded Excel or CSV files, or you can manually enter the fields to generate a click prediction. This makes it incredibly easy to use the app to estimate the success of all future articles for the publication based on the metrics you defined in the model!

You can test it out here.

Conclusion

Isn’t it great? In less than 2,000 words, we:

• Generated an awesome dataset of successful articles on Hackernoon;
• Enriched the data with data from SEO tools such as SpyFu and MajesticSEO;
• Manipulated the datasets automagically with GPT-3;
• Shipped a live app that can predict the success of future articles based on the metrics we defined.

Something that previously required data engineers, analysts, and a machine learning expert, is now available to anyone with a computer.

Welcome to the AI revolution, and it’s only going to get better!

In future articles, I will explain even more applications for marketing, sales, and customer service. Machine learning is fascinating and can save you thousands of dollars (and hours!) per week.

Give it a shot with your own dataset. I can promise it’s super fun, easy, efficient, and you will be impressed by the end result! Try it for yourself now.