Hacker News is a coveted spot on the Internet to share your technology content. Everyone wants to share content about their business or industry or themselves and make it to the front page. Upvotes divided by the power of time help determine how your content ranks on Hacker News. You may be like the many people who have Googled “how do you get to the front of Hacker News” and want to know how to get upvotes. Many factors play into the success of content on Hacker News (as with every platform!) and we wanted to explore if AI could help us determine what those are.
An AI application to determine the success of Hacker News posts has been built before. One we’re going to look at was made by Voice and Machine Learning Technical Project Manager, Jean-Frederic Plante. From his research, Jean-Frederic Plante determined that posting date and time, story language, reposting, and class of topic all impacted upvotes on Hacker News. We will be using the same Kaggle dataset as he did, just preparing it for use on our platform and the specific use case of titles versus upvotes. We are going to determine if the title has an impact on the success of your article going live on Hacker News and how many points certain titles should experience between: none, some, and high.
Using a pre-existing dataset to build a sentiment analysis AI that will determine if an article will be successful from titles
This is a great example of a time where a custom dataset isn’t really necessary - you don’t need to build a dataset of your own data. Instead, we want all the Hacker News data we can possibly scrape. Thankfully, Hacker News has their dataset of posts and associated data uploaded to Kaggle open source! For this dataset, we will want to use titles versus number of upvotes. We will divide the dataset by the upvotes: >10 points would be classified as none, 11-99 points would be classified as some, and +100 points would be classified as high. You will find this pre-prepared dataset in the Metaranx Marketplace.
Turning this AI application into a business on Metaranx
This is a great application to add to the marketplace to let others use! You can create this AI application and either host it on the Marketplace for a fee (or for free) so other users can have access to it or you can build a website around it. For example, use Webflow or other no code website building platforms, create a splash page with text like “Will your title get ranked on Hacker News?” where people can paste in their potential titles and get a ranking from the AI! You will be charged a few cents for each call or use of the AI, so be sure to add a payment feature if you’d like to profit off other people using your creation. Please note that our API/plugin features and marketplace are coming soon.
Let's build a fraudulent email sorter using no code on Metaranx
- Log in to your Metaranx Console and navigate to the Canvas
- Click "Build/Train an AI"
- Select "Language" then "Sentiment Analysis" task
- Select the Hacker News Posts dataset that has been prepared for Metaranx from the Marketplace in your Console (or upload your own dataset)
- As this data is already formatted, you can name your AI application then click "Train AI"
- When you receive a notification that training is finished, it's time to test! Create a title for content you want to publish on Hacker News and enter it into the AI. The AI will presumably determine whether it will be upvotes/receive points by telling you the likelihood: none, some, or high.
An AI application that judges title versus likelihood of success could be recreated using a dataset of Medium article titles and their claps or any number of publications and their rating systems. Though it may not be inclusive of all types of data: such as time and date of posting, author clout, content, and more, it’s a pretty fun indicator of whether your title will first drive that initial click or interest from an audience to hopefully gain points. Have fun writing, publishing, and testing out titles!