2023 Pre-Countdown Thoughts & Reflections

The Hottest 100 of 2023 has been an interesting but somewhat difficult year for 100 Warm Tunas.

The first challenge was that I have been overseas, touring Scandinavia with limited spare time and connectivity for much of the voting period. Fortunately, for what I lacked in spare time and connectivity, I made up for with photos of Auroras:

A beautiful aurora, a short drive from Tromø, Norway

Prior to going on holidays I had even considered taking a break this year, and skipping 100 Warm Tunas of 2023. Of course, the show had to go on, so I powered on. That being said, being on holidays did result in some of the data release timelines to be a little longer than usual.

The second and most major challenge this year was changes that Instagram have made that significantly hinders our ability to collect data from these sources. In the past, 100 Warm Tunas was able to collect hashtag data via the Instagram hashtag “recents” feed for a small set of hashtags.

Recently, Instagram have entirely removed this capability, replacing it with only “Top Posts” and “Recent Top Posts” with no way to view all public posts under a given hashtag:

Instagram's newly limited "Recent Top Posts" and "Top Posts"

In addition to this, Instagram also removed the ability to view stories for a given hashtag.

Whilst it is true that the prevalence of public posts has been diminishing over time (due to private accounts and the uptake of ephemeral posting mediums like Stories), it has historically been a valuable source of (less biased) data:

Year IG Posts IG Stories Website Uploads
2023 188 N/A 3129
2022 871 240 3143
2021 1227 574 N/A
2020 2235 859 N/A
2019 3126 858 N/A

This year, because of the loss of these data sources, we believe there will be a decrease in the diversity of the data collected, and thus, a potential to increase bias in the prediction.

You might be thinking, “that’s OK, the ML model will fix these biases”. However, ML models are only as good as the data they were trained on.

Our ML model is trained on past prediction datasets, using data that is sourced from a different mix of sources compared to the unseen data of 2023 (one that includes more data from Instagram), so significant changes in the balance of sourced data may not be accounted by the adjustments the model makes.

This year is also one of the first years where our predicted winner is vastly different to that offered by the odds at the bookies for “Winning Song”. At the time of writing the odds at three bookmakers currently look like this:

Sportsbet TAB Pointsbet
Paint The Town Red $2.00 Paint The Town Red $1.80 Paint The Town Red $1.80
Rhyme Dust $2.62 Rhyme Dust $2.75 Rhyme Dust $3.20
Rush $3.75 Rush $3.50 Rush $3.75
adore u $13.00 Not Strong Enough $8.00 The Worst Person Alive $10.00
Prada $17.00 The Worst Person Alive $12.00 adore u $17.00

In contrast, (at the time of writing) our prediction of the top 5 looks as follows (with ML Adjustments applied):

Position Song
1 Rush – Troye Sivan
2 Not Strong Enough – boygenius
3 Rumble – Skrillex, Fred again.. & Flowdan
4 What Was I Made For? – Billie Eilish
5 Rhyme Dust – MK & Dom Dolla

Based on the odds presented at the bookmakers, it is indication of significant uncertainty in who will take out the winning song of the Hottest 100 of 2023.

So, that leads me to wrap up with a question; will Troye Sivan win the Hottest 100 of 2023 with Rush as predicted by 100 Warm Tunas? Or will Doja Cat win with Paint The Town Red?

This year is incredibly close and very difficult to know. With the significant shift in data collection biases it could go in any direction. To me, 100 Warm Tunas isn’t just about predicting #1 correctly. In fact, predicting #1 correctly is merely a bonus.

I am mostly interested in the general “bucketed accuracy” of the prediction - that is, how many songs 100 Warm Tunas can predict in specific buckets, ignoring order (e.g. 7 of top 10, 82 of top 100)

Typically, 100 Warm Tunas will reach 7 or 8 out of the top 10, and around 70 to 80 of the top 100. If 100 Warm Tunas can come close to this benchmark, then to me, it will have successfully predicted the Hottest 100 of 2023.

I hope you enjoyed this blog post, and have been enjoying this years prediction. If you want to show your appreciation for the work that I do, you can help contribute to the project by buying me a coffee ☕️.