Since 2016, a website by the curious name 100 Warm Tunas has been predicting the results of triple j’s Hottest 100 countdown. Over time, the website has garnered a fairly high success rate, and become the go-to for punters seeking advance projections of the annual countdown.
100 Warm Tunas is a side project of Nick Whyte, a frontend engineer based in Sydney and a fan of the Hottest 100. It was inspired in part by a similar but now-defunct project, Warmest 100, but Whyte thought he could build on Warmest 100’s model – especially as users began to share their votes more often as images instead of text, much like how Spotify users flooded Instagram with their year-end Wrapped results as a shareable graphic.
“As years go on, I’m less certain about what’s really going to be Number One”
So how does 100 Warm Tunas do it? Whyte’s algorithm is built on a programming language called Python. First, the program collects images shared from Instagram and Twitter tagged with #Hottest100, as well as the text data from this voting thread on the subreddit /r/triplej. Next, it uses optical character recognition (or OCR) software to extract song and artist names from the images. After some extra data cleaning and processing, the website tally is updated with the new votes.
Whyte says the program has barely needed any human intervention this year. 100 Warm Tunas collected 36,156 votes from 4,026 images, which Whyte estimates to be 1.25 per cent of the entries this year – a disappointing sample size considering previous years (for instance, 3.72 per cent of the total votes were collected in 2016).
Even though 100 Warm Tunas has been able to grab data from multiple social media platforms, the way fans share their votes continues to change. One roadblock Whyte has had to deal with when collecting votes is users’ tendency to share their votes through Instagram stories, which vanish after 24 hours. Additionally, Instagram users tend not to hashtag their stories with ‘#Hottest100’, making it difficult for the program to find them.
“As years are moving on, I guess we’re capturing a little less data despite capturing it across different mediums. For me, I’m a bit more concerned with how accurate it will be, but I think in general it will give a good ballpark of the top ten as it usually does,” Whyte told NME prior to the Hottest 100 of 2020.
“As years go on, I’m less certain about what’s really going to be Number One.”
Following the Hottest 100 on Saturday (January 23), NME spoke with Whyte again to discuss how well 100 Warm Tunas performed. Despite the smaller-than-usual sample size, the program performed on par with previous years, accurately predicting the champion, ‘Heat Waves’ by Glass Animals. However, the tracks leading up to the top spot were certainly more jumbled. While Spacey Jane, Ball Park Music and Flume all made it into the top five as predicted, Tame Impala swooped in with ‘Lost In Yesterday’, bumping out G Flip’s ‘Hyperfine’.
‘WAP’ also made quite the leap from a predicted 14th place to number six. That was quite the surprise for Whyte, who thinks the unexpected result may come from the kinds of people who publicly share their votes. At face value, he says voters who vote for more ‘mainstream’ music such as Cardi B are less likely to share their entries – meaning 100 Warm Tunas can’t reflect their actual popularity. Whyte has observed a similar effect with Flume, whose Hottest 100 rankings over the years have caught him by surprise.
“If you even look at the top three, Ball Park Music [and ‘Cherub’] was predicted to come in at second place, and Flume [with ‘The Difference’ featuring Toro Y Moi] ended up coming in at third,” he says.
“We’ve seen that sort of thing happen before. I don’t think Flume in particular is very representative of the data that is collected. When there is a Flume song in or near the top ten, it usually ranks much higher up than predicted.
“There’s definitely bias in the data and what people vote for, and I think that’s also representative of what happened with ‘WAP’. People sharing their votes are more in tune with the more ‘typical’ music on triple j.”
A full report of how 100 Warm Tunas performed will be published in the coming days. Having already crunched some preliminary numbers, Whyte says 100 Warm Tunas successfully predicted eight of the top ten songs (not including their order), and 75 per cent of the Hottest 100 overall.
“If 100 Warm Tunas didn’t exist, either something else would, or the same people would look elsewhere for other predictions”
Considering the Cardi B anomaly, Whyte still says there are improvements that could be made to the program’s accuracy. He’s looking at integrating new techniques into the algorithm, including machine learning tools and weighting votes depending on factors such as the voter’s age or the genre of the songs. He hopes this might reduce some of the skews in the prediction – for instance, the overrepresentation of metal and heavier music in 100 Warm Tunas compared to the Hottest 100.
“One of the things a lot of people call out everywhere I see people criticising the prediction is it’s very heavily weighted on heavier metal tracks like The Amity Affliction,” he explains. (The Amity Affliction ended up at number 64 on the Hottest 100 with ‘Soak Me In Bleach’.)
“I think that’s just the demographic of where these votes are coming from and what they’re voting for. There’s definitely work to do to remove some of the bias, but that’s not something I’ve ventured into yet.”
Besides issues Whyte’s identified himself, 100 Warm Tunas occasionally faces the odd disgruntled fan who complains that the the prediction spoils the surprise of the Hottest 100. He doesn’t take it to heart – “If they think it’s spoiling the countdown, they can just look the other way”– but he does acknowledge the possibility that 100 Warm Tunas could in theory influence people’s votes.
“I’ve heard from others that the prediction becomes this self-fulfilling prophecy,” he says. “Like, people look at the countdown before they vote to figure out who to vote for, so they still only vote for the songs predicted in the top 100. I think that’s a fair observation to make.
“But also, if 100 Warm Tunas didn’t exist, either something else would, or the same people would look elsewhere for other predictions that people are putting out there, like the bookies.”
Data-driven predictions and music have become more interconnected than ever in recent years, thanks in part to how platforms like Spotify and TikTok have shaped the industry through proprietary algorithms and user-based recommendations. In the grand scheme of things, the code behind 100 Warm Tunas is relatively straightforward in how it predicts the countdown, with each vote counted equally. While Whyte admires Spotify’s all-seeing algorithms, he ultimately still wants to ensure music fans still know how 100 Warm Tunas counts votes and what they’re worth.
“I think that’s sort of the biggest thing I’m going for. I want to make sure the results are transparent, that people are aware it’s a prediction but it’s a prediction driven by real data, it’s not someone’s finger in the air,” Whyte says.
“But then as I move forward, if I do more ranking and analysis, I still want there to be that transparency, where you can go and look at the raw data if that’s something that interests you.”
And in that spirit, Whyte told NME how well he went with his own Hottest 100 votes. With a self-described “broad” music taste, three of his ten votes ended up in the final countdown: Sycco’s ‘Dribble’ (number 29), a Tame Impala track and one by Lime Cordiale. Whyte’s luck – or lack thereof – with his personal votes should convince any lingering skeptics: 100 Warm Tunas’ predictions truly are in the hands of the people, and not the mastermind behind the scenes.