I’ll keep this short and sweet. I did an analysis of The Killers using some NLP techniques in my last blog post. I scraped both The Killers and Lana Del Rey lyrics previously for some topic model analysis, but it wasn’t so successful. To not waste my scraped data, here are the same results for Lana Del Rey. For details on methodology, you can go check that out.
A preamble. I find Lana Del Rey to be such an intriguing artist. She has this very cinematic and old-fashioned quality to her, but she manages to sound fresh and contemporary at the same time. I enjoyed her Ultraviolence album, which I consider her best work, and disliked Honeymoon, which I consider to be quite boring.
Word Frequency Analysis
It’s quite funny, but not entirely surprising, to see that love, life and baby appear in the top words across all albums. Well, Lana Del Rey is a torch singer, so it’s to be expected.
Lyrics Embedding Analysis
Brite Lights, Sad Girl, Flipside, Money Power Glory appear to be in the periphery, and can be considered to be the most unusual songs, semantically speaking. Looking per album, songs from Born to Die seems to be the closest ones overall, while AKA Lizzy Grant and Ultraviolence are all over the place.
Overall, it’s quite difficult to interpret the embedding results, but it could be due to the fact that we’re using a very crude way of turning word vectors into song vectors, and so a lot of information is lost. A Part II would focus more attention on this bottleneck.