It's #data time.
We've analyzed all HackerNews post title from 2006 to late 2019 and tried to find out what titles performed the best.
We found that the statistically best HN post title of all time was "JMAP (YC S10) Linux Inside is hiring" 🤷♂️
Here is how we did it.
A thread
We've analyzed all HackerNews post title from 2006 to late 2019 and tried to find out what titles performed the best.
We found that the statistically best HN post title of all time was "JMAP (YC S10) Linux Inside is hiring" 🤷♂️
Here is how we did it.
A thread
First, a few general information about HN posts.
Here is the number of posts per quarter since the launch of the website.
Surprisingly flat since 2014.
Here is the number of posts per quarter since the launch of the website.
Surprisingly flat since 2014.
Then we analyzed, on average, how an HN post performs.
Unsurprisingly, not so well.
Median score: 2 with 66% of all posts ending up with less than 3 points.
Unsurprisingly, not so well.
Median score: 2 with 66% of all posts ending up with less than 3 points.
Title length distribution.
HN posts title can't be longer than 80 characters.
When counting words (including stop words) we find that the distribution is almost perfectly centered on 7.
HN posts title can't be longer than 80 characters.
When counting words (including stop words) we find that the distribution is almost perfectly centered on 7.
Let's now find if title length and score are correlated.
At first sight, it is not.
Which makes sense, because again, 66% of all submissions never reach more than 3 points.
At first sight, it is not.
Which makes sense, because again, 66% of all submissions never reach more than 3 points.
But if we only take into account posts with more than 5 points, the trend is totally different.
So, no matter your title length, your post won't perform well, but in the case it does, you'd want your title to be as short as possible
So, no matter your title length, your post won't perform well, but in the case it does, you'd want your title to be as short as possible
Some 1-word submission that did well:
- 2048
- Hyperloop
- Atom
We then analyzed categories that performed the best:
- 2048
- Hyperloop
- Atom
We then analyzed categories that performed the best:
And again, when discarding posts with less than 5 points, results are a bit different.
And now the best part, best performing words.
For every post in the dataset, we've split its title in single words, removing stop words (a, the, an, etc..).
We ended up with a big array where each line was a word of one post's title associated with the score of said post.
For every post in the dataset, we've split its title in single words, removing stop words (a, the, an, etc..).
We ended up with a big array where each line was a word of one post's title associated with the score of said post.
Finally, we aggregated everything and we were able to find the top 50 words that had the best score.
For example, the word “zwiebelfreunde” appeared in 7 titles and the median score of all those submissions was 121 points. Impressive.
Interactive graph plotly.com/~daolf/61/
For example, the word “zwiebelfreunde” appeared in 7 titles and the median score of all those submissions was 121 points. Impressive.
Interactive graph plotly.com/~daolf/61/
You can see that Jmap (an IMAP successor) and LinuxInsides (an open-source book about the Linux kernel) are top performers.
They appeared in more than 20 posts and yet managed to have a median post score above 60!
They appeared in more than 20 posts and yet managed to have a median post score above 60!
Hence the best HN title of all time: "JMAP (YC S10) Linux Inside is hiring".
Unfortunately, our model might need some improvements 😬
news.ycombinator.com/item?id=226748… (only 14 points)
Bonus:
Unfortunately, our model might need some improvements 😬
news.ycombinator.com/item?id=226748… (only 14 points)
Bonus:
As requested by @paulg, here are the most common programming languages on HN titles
And finally, the trend of some interesting words:
