Embedded Files
Are all the top modern RnB Songs similar stylistically?
My research is focusing on whether all modern RnB music is the same, as many people say that the genre has become fairly homogeneous. I am trying to find out through textual analysis whether modern RnB music is stylistically similar, or whether the topics that appear in modern RnB and the textual style differ from song to song.
Since I am interested in the style of RnB music as texts, I am engaging in "Content Analysis" as I am "focused on the texts themselves rather than text's relation to their social and historical contexts" (Ignatow, 2018).
In order to conduct stylistic analysis of my corpus I decided to use stylometry. Stylometry is the measure of style, and "it has also been used to study a wide range of other issues, including the evolution of language, the classification of texts into genres, and the identification of individual writing styles" (Rockwell and Sinclair, 2016, pg.104). The use of this tool would ultimately allow me to assess the stylistic similarities between the songs within my corpus.
Below is a dendrogram, which represents how similar certain texts are to each other by using delta distances mentioned above, and coded to create this visualization in R.
One strength that this visualization provides is that it shows exactly how different R&B music is in terms of the words that songwriters use from song to song. There are few close connections between the lyrics which highlight how different the R&B lyrics seem to be. This dendrogram offers a different perspective in that R&B music continues to be completely different from year to year and artist to artist, refuting the claim that “all modern RnB music is the same”.
For example, "Girl on Fire" by Alicia Keys is shown to be completely different from the rest of the RnB songs, and especially "Blinding Lights" by The Weeknd, as they are on opposite ends of the dendrogram. According to Alicia Keys, "Girl on Fire" is about "new beginnings, new perspectives and fresh starts. It’s about finding your own inner strength and channeling it in a way you’ve never tried before. To be “on fire” is to allow yourself the freedom to take full control of who you are and how you want to live your life" (SongMeanings, 2023).
This seems to be quite a bit different than most RnB music which tends to focus on love and dancing rather than being an "inspiring" genre. This is reflected in The Weekend's explanation of their track "Blinding Lights" as a song about "how you want to see someone at night, and you’re intoxicated, and you’re driving to this person and you’re just blinded by streetlights" (Esquire, 2020).
Overall, it seems as though the style of each song within the corpus are pretty different and tend not to share very many characteristics, as shown by the very few overall big clusters, there are much more small individual clusters.
However, I wouldn't be able to stop here and take the information in this visualization at face value as "the tradition of sceptical questioning... creates space for agile interpretation instead of trying to permanently solve interpretive questions" (Rockwell and Sinclair, 2016, pg. 168).
I was a bit skeptical of my results, as it is possible that I might have made some errors in the code, or R returned results that might not be replicated. However, after using Voyant Tools (https://voyant-tools.org) to check my results, I found that the relative frequencies of the most popular words in my corpus lined up with the songs that appeared closest to each other in my first dendrogram.
"Can't Feel My Face" by The Weeknd and "Girls Need Love" by Summer Walker have the highest raw frequency of the word "Love".
"Hold On We're Going Home" by Drake and "Talk" by Khalid have the highest raw frequency of the word "Just".
"FourFiveSeconds" by Rihanna, Kanye West, and Paul McCartney and "Freaky Friday" by Lil Dicky have similar relative frequencies of the word "Got".
If you look closely, the songs that had the highest relative frequency of certain words in Voyant Tools, also appeared very close to each other within my dendrogram.
"Can't Feel My Face" by The Weeknd and "Girls Need Love" by Summer Walker had the highest relative frequency of the word "Love" and appeared closest to each other in the dendrogram (see the green circle below). Additionally "Hold On We're Going Home" by Drake and "Talk" by Khalid had the highest relative frequency of the word "Just" and appeared very close to each other within the dendrogram (see the pink circle below). Finally, I checked two songs that didn't have the highest relative frequency of the word "Got", "FourFiveSeconds" by Rihanna, Kanye West, and Paul McCartney and "Freaky Friday" by Lil Dicky, but had similar relative frequency of that word, and they appeared closest to each other in the dendrogram (see the blue circle below).
This tells us that the songs are overall pretty different stylistically, and the Voyant Tools "Trends" feature corroborates this claim.
In this case, I would say that R is a better measure of how similar these songs are to each other textually, as it presents a tree diagram of similarity, while Voyant Tools doesn't have such a feature other than showing the relative frequency of certain popular words, which leaves us guessing as to what other aspects of the corpus are similar to each other. Although "Hold On We're Going Home" by Drake and "Talk" by Khalid had the highest relative frequency of the word "Just", they weren't the closest songs to each other in the dendrogram as they were separated by "Diamonds" by Rihanna, which isn't reflected in Voyant Tools.
The difference between "Girl on Fire" and "Blinding Lights" is big stylistically, but I think that makes sense within the historical context of both songs. "Girl on Fire" is from the beginning of the 2010s, while "Blinding Lights" is from the beginning of the 2020s. Gen Z was able to begin listening to music much more freely as they grew into adults during this time, skewing the popularity of certain music because of interest in certain lyrics that were different from the generation that loved RnB music in the beginning of the 2010s.
To look further into the similarity of the texts, I decided to subset the text to use only nouns, which is featured in this dendrogram.
This could be explained by the fact that the nouns represented certain topics that appeared in all of the songs, making the value of difference 0, and presenting the songs as such. This might be considered an unhelpful visualization of stylometry, however, it shows that potentially the topics that are sung about by popular RnB artists tend to be the same. Since it seemed that this might be the case, I decided to look at methods to display the most popular topics that appeared within my corpus.
Furthermore, I would be remiss not to explore more of my corpus and articles about RnB music to find information about the similarity of modern popular RnB songs as "it is important to recognize that stylometric methods are not a substitute for other forms of evidence" (Rockwell and Sinclair, 2016, p. 104).
Do the top RnB songs sing about the same topics?
This is is WordCloud I created in R.
This is is a WordCloud that Voyant Tools created of my corpus.
I decided to continue my analysis by creating some WordClouds to get a quick glimpse into the most popular words and themes that appeared within my corpus. I chose to do this because "you can’t just ask what themes are important in a text... however, you can ask what clusters of words (which might indicate themes) have a high frequency (which might indicate importance)"(Rockwell and Sinclair, 2016, p. 170). These WordClouds show that the topics that are sung about within the corpus. However, they are showing different results. Why is this?
My suspicion is that R is taking into account the actual relative frequency of each topic and returning it in the WordCloud, while Voyant Tools is using the raw frequency of certain words to decide how important certain topics are. To look into this, I looked at the "most frequent words" tool to see if this was true (see screengrab below).
It seems that my suspicion was correct, as all the most frequent words appeared within the topic word cloud tool.
To go a bit further into this, I decided to create a scatterplot of each "topic" from my results in R and my results in Voyant Tools because they "can be used to identify patterns, clusters, and outliers in the data, as well as to explore relationships between different variables. In text analysis, scatterplots can be used to explore the relationship between different features, such as word frequency and document length, or to compare the distribution of different features across multiple texts." (Rockwell and Sinclair, 2016, p. 155)
These scatterplots tell us that the overall frequency of the topics returned by R (on the left) is smaller than the overall frequency of the topics returned by Voyant Tools (on the right), which corroborates my claim that R returns topics that are more common relative to the corpus, while Voyant Tools returns sheer frequency. This is shown by the highest topic returned in R being "Need" appearing just under 25 times, and the highest topic returned in Voyant Tools being "Got" appearing over 25 times.
However, it seems that all these topics were found to appear relatively frequently and there is an overall trend of topics that tend to stay similar in RnB music.
Do these topics that popular RnB artists sing about change from year to year?
In order to find whether these topics change, I decided to use topic modeling. Rockwell and Sinclair define topic modeling as a "popular technique for discovering latent themes or topics in large collections of text."
To do this, the topic models use probabilistic methods to identify most likely topics given patterns of word co-occurence within the text. The most widely used model is the "Latent Dirichlet Allocation (LDA), which assumes that each document is a mixture of topics and that each topic is a probability distribution over words" (Rockwell and Sinclair, 2016, p. 109).
Below is a Topic Model of the corpus created in R. It shows that the relative proportion of the topics it found most often within the corpus varies every year, but tends to stay pretty similar. The topics that this model returned were "gon caus tonight shit kill", "love want know say home", "babi everybodi ayi bring need", "work hurt turn wait tell", and "girl wild got babi come". The topics feature a few of the most frequently appearing words within the corpus such as "girl", "love", and "baby" (lemmatized to "babi" to include all instances of the word "baby" including "babies").
Unfortunately, Voyant Tools doesn't have a feature that is able to display the distribution of topics within each year, so it wasn't possible to compare my results in R in this capacity. However, I was able to compare the topics returned in R and in Voyant Tools, with the topics that Voyant Tools returned shown below.
Voyant Tools also showed the prevalence of a certain topic within each song. A few examples of this are shown in the screengrab to the left.
My analysis in R was able to create topics that appeared within the text and then show how those topics appeared over 10 years, while Voyant Tools was able to show topics that were prevalent within the corpus and show their distribution within each song.
R in this case was more helpful in answering my question of whether topics within modern RnB changed from year to year, while Voyant Tools was able to dive a bit deeper and create other topic ideas and show the actual distribution of these topics within each song, at a closer level.
Overall, I think both tools were able to complement each other, however, I feel that R was better at showing the distribution of topics from year to year.
Ultimately I think it is important to recognize the abstraction that topic modeling provides, as it isn't a complete or unbiased interpretation of the corpus. The topics that appear may "represent underlying themes or concepts that are present in the text, and they can provide a useful starting point for further analysis and interpretation" (Jockers, 2014, p. 51). However, topic modeling does not "provide a definitive interpretation of the text or its themes, but it offers a way to explore the patterns and structures in the data and to generate hypotheses about the underlying meaning" (Jockers, 2014, p. 51).
Therefore, there are limitations within this analysis, and room for further interpretation and investigation.
Page updated
Report abuse