Why song recommendations are harder

Yes. Both of those are major reasons. And there are a few more.

Why song recommendations are harder

1. Songs are tiny

A novel gives you hours of signal:

• plot

• prose

• characters

• ideas

• humor

• pacing

• payoff

A song gives you 3–5 minutes.

So a book has many more dimensions available to discriminate taste. A song can be “great” yet still miss because one tiny thing is wrong: vocal timbre, production sheen, chorus shape, lyric phrasing, etc.

2. Repetition changes the object

Books are usually consumed once, maybe reread occasionally.

Songs are consumed:

• once

• 10 times

• 100 times

• for decades

So liking a song is not just:

“Was this good?”

It is also:

“Did this survive repetition?”

That’s a very different target.

3. Songs become autobiographical

This may be the biggest difference.

A song can acquire value from:

• a girlfriend

• a car

• senior year

• a concert

• a TV scene

• a movie

• a bar

• a specific summer

That context can become half the song.

A newly recommended song has none of that history yet.

4. Familiarity itself is pleasurable

There’s a huge mere-exposure / recognition component.

That’s why the “big Billboard songs” experiment suddenly worked better. Even when you didn’t love them, they often had a glimmer.

The obscure recommendations had no foothold.

5. Song taste is more discontinuous

You can love:

• one Cyndi Lauper song

• reject another

• love one Def Leppard-era rock song

• hate Pour Some Sugar on Me

• love one REO song much more than another

So artist/genre-level similarity is weaker.

6. The positive target is rarer

With books, “this is probably a 7–8/10 for Peter” is useful.

With songs, you’re hunting:

the one that becomes part of your life

That is a much narrower target.

7. Your negatives are easier

This is why our PASS model got good quickly.

It’s easier to identify:

• novelty

• lightweight

• certain MOR

• groove without enough substance

• culturally wrong era

• emotionally empty-for-you

than to predict:

“This specific song will become top tier.”

So the real target isn’t “song quality”

It’s closer to:

Intrinsic fit × familiarity × attachability × repetition durability × emotional context

That is way harder than “will Peter enjoy this novel?”

And the funny part is: once we have your iTunes plays/ratings + YouTube list + explicit YES/PASS examples, we actually have enough data to start modeling those factors separately.