Latent Dirichlet Allocation for Heavy Metal Lyrics
Some time ago, I wrote about simple classification schemes for metal lyrics.
To take this analysis further, and to learn some cool new stuff, I ran LDA on lyrics from a few thousand randomly selected songs from the approximately 200,000 songs in the database. Instead of using gensim, I wanted to get more familiar with Apache Spark’s MLLib.
The goal was to see if the topics were roughly the same as the categories in the other analysis, or if the topics ran along the lines of genre (power metal, black metal, etc.).
Here’s what I found with 10 topics:
36415 words in vocabulary
Top words for topic 0 come, see, know, going, never, need, tell, time, better, one, gates, keep, dead, like, turn, on,, nowhere, power, away, everything
Top words for topic 1 fire, never, life, like, iâm, see, world, end, blood, hell, die,, pray, you,, know, time, take, one, nothing, sky, itâs
Top words for topic 2 like, one, time, new, world, take, made, us, away, breath, never, me,, must, go, life, past, know, end, see, we’re
Top words for topic 3 show, know, come, king, till, see, one, life, time, hey, me., away, make, never, gather, way, people, men, like, og
Top words for topic 4 one, eyes, light, wanna, like, way, wake, time, see, cry, really, never, take, up,, life, make, live, know, feel, pain
Top words for topic 5 death, blood, dead, life, flesh, like, evil, open, black, mind, us, never, time, pain, human, hell, one, left, fear, world
Top words for topic 6 never, day, us, ready, one, see, come, know, give, feel, like, ever, time, shall, stand, long, another, believe, nothing, take
Top words for topic 7 night, time, death, see, life, come, way, turn, world, one, feel, know, black, get, heart, death,, like, say, yeah,, sky
Top words for topic 8 go, could, like, one, come, know, forever, love, never, get, see, light,, free, time, life, would, feel, still, someone
Top words for topic 9 life, got, time, back, feel, want, see, right, know, eyes, things, around, hope, take, never, watch, live, fucking, world, gonna
The code can be found on GitHub.