Calling all Tame Impala Fans…
An similarity analysis of Tame Impala’s top three songs
Tame Impala is a very unique “music project” known for its psychedelic take on music. It is a one-man band which makes music under the Alternative or Indie Rock genre. My goal in my analysis was to identify which other artists are most similar to Tame Impala in order to understand where the band lies in relation to other musicians in their genre of music. This insight could help inform decisions of streaming platforms such as Spotify or Youtube Music. These platforms rely on recommending music to a user which matches certain criteria based on the user’s music interests. Understanding which songs are most similar to certain Tame Impala songs could help fans branch out and find other bands that are likely to enjoy. Also, streaming services can use this algorithm to help recommend artists/songs based on similarity in certain features.
I worked with the Spotify Music Analysis dataset which provides information on each song’s acousticness, danceability, duration, energy, instrumentalness, key, tempo, and more. Tame Impala songs are known for their mellow, yet “groovable” songs, their unique tempo, and their mesmerizing instrumental solos. Thus, I decided to extract the danceability, tempo, and instrumentalness features for my analysis using cosine similarity. I queried the dataset to find the top 10 most similar songs to “Let It Happen”, “The Moment”, and “Feels Like We Only Go Backwards”. Since each set of features contained numerical values, I used cosine similarity to measure the similarity between two sequences of numbers.
Let It Happen: https://gist.github.com/mpothen1/71bffffc0cb719ea0214b410a77e8b86
Feels Like We are Only Going Backwards:
Let It Happen, Tame Impala’s most famous song is most similar to Can We Go Up?, by Paper Diamond. This was interesting because this artist is known for creating electronic music, which is not Tame Impala’s genre. Front to Back by Buku was most similar to the second query song, The Moment. Interestingly, Buku is an artist also known for electronic music. The third query song, Feels Like We Only Go Backwards, was most similar to Emperor Jones by Joe Lavano. This is a completely instrumental saxophonist song. These results perfectly capture Tame Impala’s music style: diverse. The band has songs which are similar to pop or electronic music and other songs that are more closely related to old school jazz songs. Deathcore bands such as WhiteChappel and gospel bands such as Cathedrals were in the list of artists that produce similar music to the queried songs. This demonstrates the range in Tame Impala’s music.
I used the sklearn library to calculate the distance metric, cosine similarity. This facilitated my analysis by allowing me to capture the measure between the angle of the sequence vectors. I also used the numpy library in order to calculate the cosine similarity score. The formula I used was:
cosine = np.dot(target_song,this_song)/(norm(target_song)*norm(this_song))Here is an example of a query for a song examining similarity on three features:
Finally, I used flourish to create a table depicting the feature measures of Tame Impala’s most popular song, Let it Happen. Since I used a pre-existing, clean dataset, I only modified the dataset to extract the features I needed to focus on to have the data. I converted the data into a data dictionary (format: {song: {song information}}) in order to have the data in a more usable format. The main limitation of my analysis code is efficiency. To improve this, I can create a function in which a user can input in a song name of a table of the ten most similar songs. Also, the similar song results are printed instead of stored in a variable for further use. If the result was put into a dataframe or dictionary, this would allow me to create visuals using Python’s seaborn library.
Conclusion
My analysis demonstrates that music from seemingly opposite genres can still be highly similar in certain features. The main takeaways from my analysis is that music streaming platforms cannot necessarily always recommend other Indie bands for Tame Impala fans. It would be more accurate to recommend songs based on the specific Tame Impala songs that a fan listens to the most. Each song is so different from the other and may fall within various genres. Overall, music platforms can improve their customer experience by using danceability, tempo, and instrumentalness measures of songs in their calculation of music recommendations for their users. Rather than recommending songs which fall within the same genre, using other features to make recommendations will allow their users to be exposed to a greater range of artists within their music interests.
Download the jupyter notebook file to see full analysis: https://drive.google.com/file/d/1nKftXlaQ-IJElCOkEKICGG8hMS2k1q4Z/view?usp=sharing
