Daejeon, July 27–31
This research examines the presence and evolution of materialistic practices in top-charting U.S. song lyrics from 1958 to 2021. Combining close readings with interpretable NLP methods, we identify genre-specific patterns, with materialistic expression disproportionately concentrated in rap and increasing markedly from the 1990s onward. We discuss how these trends illuminate the role of music in reflecting and shaping societal values, highlighting how materialistic themes are interwoven with cultural narratives and consumer behavior, and we outline implications for studying status performance and inequality in mainstream popular culture. These findings are preliminary: ongoing work will extend the corpus and evaluate LLM-based analyses to better capture metaphor, stance, and ambiguity beyond surface lexical cues.
Lyrics have always served as powerful vehicles for social narratives. Especially in Pop Culture, they can be used to reflect and often shape the values, desires, and behaviors of contemporary society. In this paper, the specific value we examine is materialistic culture. While exploring popular music in the United States from the late 1950s to the current chart toppers, we aim to understand materialistic topoi in music, expressed through actions or desires, using computational techniques.
Building on S. Frith's observations in "On the Value of Popular Music," the pop song formula is either shaped by or shaping market forces, aligning with Taylor's analyses in understanding the dynamics of new capitalism within the realm of popular music culture (Frith 1996 ; Taylor 2014).
This proposal reviews prior work on materialistic culture and popular music, then states our research question and outlines our computational approach using topic modeling, TF–IDF lexical cues, and NER. We combine these methods with close readings and present initial results, limitations, and future directions.
Materialistic culture has been widely studied across disciplines and broadly refers to the importance placed on acquiring and owning goods and wealth as key measures of success and happiness.
The roots of materialistic culture can be traced back to the industrial revolution, which brought about mass production and increased accessibility to consumer goods. This era marked the beginning of modern consumerism, where material wealth became increasingly associated with social status and personal success. The post-World War II economic boom further amplified these trends, leading to the ‘age of affluence.’ During this period, the proliferation of advertising and the rise of suburban living reinforced the idea that happiness and fulfilment could be achieved through the acquisition of material goods (Cohen 2004 ; McCracken 1990).
Foundationally, Belk conceptualizes materialism through possessiveness, non-generosity, and envy, highlighting how possessions become central to identity and social status (Belk 1985). Related psychological work likewise emphasizes the symbolic functions of goods in self-definition and social comparison (Dittmar 2007).
Popular music intersects with materialistic culture both historically and thematically. In the postwar consumer boom, music was directly mobilized in advertising and promotional culture, aligning with broader shifts in consumption as an identity strategy (Taylor 2014). Contemporary genres—including pop, rap, and hip-hop—frequently foreground wealth, luxury, and consumption, echoing wider cultural ideals of material success. Hip-hop has been a focal point in this discussion: Rose shows how lyrical displays of money, cars, and fashion can operate simultaneously as status performances and as commentary on structural inequality (Rose 2008). Media-effects research further suggests that repeated exposure to materialistic messages can reinforce materialistic values and consumer orientation, particularly among younger audiences (Richins / Dawson 1992).
Despite their ubiquity, lyrics have historically been underused in consumer research; Askegaard notes that systematic engagement remained limited (Askegaard 2010). Early computational approaches examined psychological and affective change through word categories and lexicons (DeWall et al. 2011), while subsequent work used supervised and lexicon-based methods to classify lyrical themes and emotions (Mahedero et al. 2005 ; Napier / Shamir 2018 ; Choi et al. 2018). Recent quantitative studies extend to broader cultural trends, including linguistic simplification over time (Varnum et al. 2021), and to social critiques such as gender bias measured through embedding-based methods (Betti et al. 2023). However, materialism in lyrics has largely been addressed through qualitative analyses (Castillo-Villar et al. 2020 ; Church 2019), leaving room for large-scale NLP-driven studies that trace how materialistic themes emerge and shift across genres and decades.
How has the expression of materialistic practices in the lyrics of top-charting U.S. songs evolved from 1958 to 2021, and how do these musical portrayals reflect contemporary materialistic culture?
We merge two sources: Billboard Hot 100 weekly charts from 1958 to 2021 (Dave 2021) and a Genius lyrics dataset (Nayak 2022). After deduplication, case-normalization of titles and artist names, separation of featuring artists, and filtering to English-language lyrics via the Genius language label, the final dataset contains 13,015 unique songs, representing 44% of 29,681 unique Billboard entries (Figure 1 and Figure 2). Each entry includes year, artist, title, genre, and lyrics.
Figure 1: Unique songs per year
Figure 2: Genres distribution
Lyrics are challenging for automated interpretation because they often rely on figurative language, shifting personas, and occasional irony. Rather than modeling stance, we focus on detecting the presence of materialistic expression at scale. We implement a transparent three-signals pipeline and then combine signals into a final label.
We fit an LDA topic model, sweeping topic counts from 2 to 15 and selecting 12 topics by maximizing average coherence score across runs. We use the pyLDAvis interface with λ ∈ [0.25, 0.5] to interpret topic semantics, and flag clusters as materialistic when they prominently feature high-loading terms about money, luxury goods, or branded consumption, such as Gucci, Bentley, dollar, money.
We build a lexicon grounded in close reading targeting two categories: acquiring/selling terms (buy, purchase, dollars, wealth, price, expensive) and material goods (car, diamonds, jewelry, gold, luxury, brands, fashion). Desire verbs want and need were excluded after close reading revealed their predominant use in love songs. Songs are flagged when lexicon terms receive strong TF–IDF signals.
We apply a general-purpose NER model restricted to organisation-type entities as a brand proxy, accepting known imprecision: common nouns and capitalized compound words occasionally trigger false positives, as confirmed by manual inspection of predictions.
We combine the three signals into a final label by requiring agreement between at least two methods, yielding 1495 materialistic songs (Table 1).
| TM | TF-IDF | NER | # of songs |
| 0 | 0 | 0 | 9114 |
| 0 | 0 | 1 | 400 |
| 0 | 1 | 0 | 1654 |
| 1 | 0 | 0 | 370 |
| 0 | 1 | 1 | 227 |
| 1 | 0 | 1 | 171 |
| 1 | 1 | 0 | 529 |
| 1 | 1 | 1 | 568 |
Table 1: Distribution of method labels
Topic modeling selects 12 topics by maximizing average coherence across runs. Interpreting the 2D topic space as Not Love–Love on the x-axis and Lonely–Party on the y-axis, manual inspection identifies three materialistic topics concentrated near Not Love and Party (Figure 3, Table 2). This topic-based signal flags 1638 songs as materialistic (12.59% of the corpus).
Figure 3: Topics identified in the space [Not Love, Love] (x-axis) and [Lonely, Party] (y-axis)
| Topic | Semantic title | Sampled words (λ ∈ [0.25, 0.5]) |
| 1 | Messy relationship | girlfriend, want, you, nobody |
| 2 | Broken-hearted night | moon, trouble, never, please |
| 3 | I'm better than you | Gucci, got, Bentley, like, f**k |
| 4 | Love letter | love, heart, sweet, need |
| 5 | On the road | summer, star, road, dream |
| 6 | The American Dream | dollar, money, price, fame |
| 7 | Story of a boy | boy, man, war, jealous |
| 8 | Fancy party | Versace, chain, diamond, funk |
| 9 | Betrayal | goodbye, again, lie, somebody |
| 10 | I want you | baby, body, got, come |
| 11 | Let's dance | shake, party, got, come |
| 12 | The muse | she, sixteen, woman, said |
Table 2: Topic modeling results (12 topics)
NER focuses on organisation entities as a proxy for brand mentions and flags 1366 songs (10.49% of the corpus).
Genre and time patterns. Using the final two-of-three decision rule, materialistic expression is strongly genre-dependent. Rap shows the highest prevalence, with 1227 of 2112 songs flagged (about 58%), while pop contains 110 materialistic songs out of 5570 (Figure 4). Over time, detected materialistic lyrics increase sharply from the 1990s onward, closely tracking the growing presence of rap in the corpus, while other genres remain comparatively low and stable (Figure 5). Within non-rap genres, materialistic rates are low and relatively uniform: country at 2% (26/1389), R&B at 4% (73/1646), and rock at 3% (58/2307), suggesting materialistic expression is not a general feature of chart popularity but is concentrated within rap's specific cultural economy.
Figure 4: Distribution of genres with materialistic count and ratio
Figure 5: Counts of rap songs, materialistic songs, and non-rap materialistic songs over time
This work-in-progress traces materialistic expression in top-charting U.S. lyrics from the late 1950s to 2021 and reveals a pronounced genre effect: materialistic themes are concentrated in rap and rise sharply from the 1990s onward, while pop, country, R&B, and rock remain comparatively low.
Methodologically, the pipeline shows how combining weak but interpretable signals can support scalable thematic analysis while remaining auditable through close readings, despite figurative language and implicit meaning. The study underscores the role of music in reflecting and shaping societal values, highlighting how materialistic themes are interwoven with cultural narratives and consumer behavior.
Substantively, the observed rise in rap aligns with accounts of hip-hop as a site where wealth, luxury goods, and conspicuous consumption function simultaneously as status performances and as responses to structural inequality (Rose 2008). The current pipeline intentionally does not resolve this duality, as its scope is prevalence detection rather than stance classification; distinguishing celebratory from ironic or critical mentions remains a target for LLM-based follow-up work. More broadly, this pattern is consistent with theories of material culture in which possessions operate as symbolic resources for self-making and social positioning (Belk 1985 ; Dittmar 2007). Topic modeling was applied to the full corpus without temporal segmentation; the time dimension enters only through metadata, meaning temporal topic modeling remains a promising extension.
Limitations include coverage bias introduced by lyrics availability: older songs are underrepresented in community-curated platforms such as Genius, where contributor attention skews toward contemporary releases, and the sharp post-2000 increase in available songs may inflate the apparent rise in materialistic expression independently of any true cultural shift. Further limitations include genre label noise and imperfect NER performance without a curated brand knowledge base.
We would like to thank Prof. Jérôme Baudry at the Laboratory for the History of Science and Technology at EPFL for his support and guidance. We are also grateful to Dr. Alina Volynskaya, Joel Swai Praz, and Semion Sidorenko for their valuable feedback, encouragement, and contributions throughout the development of this work. We would also like to thank the EPFL Digital Humanities Institute for supporting our conference application and attendance.