Unknown pleasures: techniques of taste in the algorithmic recommendation of unfamiliar art music

ABSTRACT

Research into cultural tastes has commonly sought to analyze and understand preferences in terms of notions of familiarity. Such approaches are inadequate, however, when it comes to examining our engagement with unfamiliar cultural content. This paper responds to this gap by examining how people respond to algorithmic recommendations of culture through a case study of unfamiliar Australian art music. It firstly identifies three different “techniques’ by which audiences engage with and value music: functional, emotional, and intellectual. The analysis then examines how these techniques, together with measures of familiarity and the acoustic “materiality” of the music itself, combine to predict the affective ratings given to music recommendations. The findings show that audiences display a surprising capacity to engage with the unfamiliar. The paper argues for the need to develop more nuanced understandings of the relationship between familiarity and preferences which are capable of accommodating a taste for the unfamiliar.

KEYWORDS:

• Music
• taste
• familiarity
• aesthetics
• algorithms

Introduction

For those who create, present and disseminate music the problem of how to introduce listeners to unfamiliar sounds has been a long-standing challenge. The rise of the culture industries in twentieth century consumerist societies has meant the need to grow audiences and markets has been a concern for those working in fields of cultural production. The development of new audiences has also been argued as crucial to the ongoing health and vibrancy of musical practices (Nowak, 2016). For listeners, the importance of fostering music discovery behaviours forms part of the “emancipatory possibilities offered by music” (Hesmondhalgh, 2022, p. 18). The context in which audiences undertake this discovery has been transformed by the proliferation of digital music services. Not only do they offer audiences access to over 80 million songs, but they support new models of consumption in which users’ listening habits inform algorithmic logics which influence our capacity to encounter this unprecedented wealth of music.

These algorithms, and the software engineers who design them, are increasingly acting as “infomediaries’ that shape how audiences discover and experience cultural content (Chambers, 2023a; Morris, 2015). However, the capacity of these new digital services to broaden access to musical expression is questionable. The world’s largest digital music service, Spotify, reported that 20% of the songs available on its service have never been played (Palermino, 2014). Both researchers (e.g. Anderson et al., 2020) and critics (e.g. Pelly, 2018) have called attention to the lack of musical diversity which can arise from algorithmic recommendation. Developing an understanding of how audiences engage with unfamiliar music, and how they might be encouraged to do so in algorithmic listening environments, is therefore of critical interest.

Dominant approaches in fields such as cultural sociology and music psychology, however, have typically theorised preferences in terms of familiarity. Such theories not only risk circularity – “we like what we already like” (North & Hargreaves, 2008, p. 88) – but also largely fail to account for the possibility of listeners engaging with music that is unfamiliar to them. Through an examination of people’s affective responses to a niche genre of music, this paper responds to this gap by focusing on how people engage with algorithmically recommended unfamiliar music. The niche genre chosen as the case study for this paper is contemporary art music by Australian composers and sound artists. This genre’s suitability stems both from its relative obscurity and its heterogeneity of styles and sub-genres (e.g. ranging from microtonal computer music through to orchestral works) – both of which serve to guarantee a level of unfamiliarity for participants involved in the research. To investigate how people respond to this music the paper considers not just people’s familiarity with the music but also how the role played by different “techniques’ through which people value and appropriate music combines with the acoustic materiality of music to influence affective engagement. As well as extending a sociological understanding of the operation of musical taste, the article argues for the importance of considering factors beyond familiarity to support audiences and artforms in encouraging engagement with unfamiliar music in an age of algorithmic mediation.

Theorising familiarity and taste

A concern for theorising the role of (un)familiarity in our responses to music can be traced to Meyer’s (1956) pioneering work in musical aesthetics. In linking familiarity to an expectation born of experience, Meyer (1956, p. 35) posited that “music in a style with which we are totally unfamiliar is meaningless”. An association between unfamiliarity and a sense of being disoriented is similarly found in sociological theorizations, with Bourdieu observing that the mastery of particular codes is necessary to avoid the beholder being “lost in a chaos of sounds and rhythms” (Bourdieu, 1984, p. 2).

This link between familiarity and musical meaning has underpinned a range of theorizations which have sought to explain musical and cultural tastes. In cognitive psychology, the investigation of taste has frequently been framed in terms of “preferences’, with Juslin (2013) defining the object of this domain of research as “low intensity, long-term, affective evaluations for items’. While affect has been measured with instruments ranging from liking scales (Taylor & Dean, 2019) to facial electromyography (Gerger et al., 2014), the underlying explanatory frameworks of such investigations frequently draw on psychological mechanisms relating to familiarity. From Berlyne’s (1971) influential work on the role of the limbic system in producing reward effects, through to concepts such as mere exposure (e.g. Witvliet & Vrana, 2007), a variety of formulations claim to account for how preferences are guided by, or otherwise related to, familiarity.

For cognitive approaches, the idea that there might be patterns which explain some of the variation in affective responses across different individuals is typically of less concern. Research approaches stemming from critical sociology, conversely, do seek to address the external origins of taste and provide an alternate lens through which the link between familiarity and taste can be theorised. Bourdieu’s (1984) work in Distinction has been a particularly influential example, in which he identifies homologies between the social space of actors and their corresponding cultural interests – locating the sense for what is preferred in socially inculcated embodied practices. This suggests that the appeal of the familiar – and, just as significantly, our rejection of the unfamiliar – can be traced to the ways in which our social environment becomes sedimented in our everyday lives. Bourdieu’s concern with the cultural dispositions of those with varying levels of cultural or social capital has influenced research which seeks to operationalise these forms of capital and examine their relationship with tastes (e.g. Michael, 2017).

Throughout the formulations above, however, there is an element of circularity which limits their utility in considering listeners’ engagement with specifically unfamiliar music. Bourdieu’s framework, for example, emphasises the formation of dispositions which serve to reproduce the social, and so assumes a field of already existing and culturally acknowledged items. In Lembo’s (2017) analysis of people’s later-in-life acquired taste for honky-tonk music, she argues that Bourdieu’s concern with explaining stability (and dispositions already formed) is ill-suited to investigating the acquisition of tastes, as it would demand significant “moments of disjuncture” to trigger re-habituation and the development of new competencies. Such a theorisation is therefore too broad and heavy-handed to offer a lens through which to analyze people’s less intensive engagement with unfamiliar music.

An alternative approach to investigating this apparent paradox is found in research which focuses less on the “what” of cultural tastes and instead draws attention to the “techniques’ by which taste is exercised. Bourdieu himself does draw a distinction between his concepts of opus operatum, which constitutes the “structured products’ which are the objects of our tastes in particular fields, and the modus operandi which represents the governing structuring principles of our tastes and which can be transferred from one field to another. The contrast between the two is highlighted in the critique which Bourdieu (1984, p. 573) makes of social psychology for its “atomistic approach” which seeks partial laws accounting for the products of practice, as opposed to general laws concerning processes of production.

An emphasis on these techniques is found in the “pragmatic turn” of Hennion’s sociology of attachment (Hennion, 2005; Hennion, 2012). As with Lembo, Hennion takes issue with the static conceptualisation of taste which dominates critical sociology, and instead argues for an emphasis on the ways in which audiences form attachments to the objects of their interest. By foregrounding this notion of “techniques of taste” research can turn its attention to the different ways in which audiences appropriate and derive meaning from culture. Importantly, this approach also affords new forms of agency to the objects of taste themselves. Rather than being reduced to arbitrary conduits for social forces, it permits a “return to the work” in analyzing the encounter between social actors and cultural objects (Varriale, 2016). In the context of the present study, it allows a consideration of whether particular musical characteristics within the diversity of Australian art music provide particular “musical affordances’ (Krueger, 2011) – providing an action consequence of positive affective associations for particular audiences on encountering those characteristics.

The idea of a tasting technique which might be more inclined towards the unfamiliar is reflected in the concept of the “cultural omnivore”, which Peterson (1992) originally associated with a capacity to appreciate the aesthetics of diverse cultural forms. Whereas much research simply reduces omnivorous to the inconsistent measurement of the diversity of people’s musical tastes (Robette & Roueff, 2014), Ollivier (2008) instead considers the capacity for omnivorousness to form a particular technique of taste and mode of appreciation. In examining the different discourses which underlie the attitudes of people whose consumption habits would typically classify them as omnivores she identifies a “humanist” mode of cultural openness which exhibits mobility across cultural forms and which places a high value on discovery as a means to learn and stimulate the mind. Ollivier did not identify this tasting technique as simply some enlightened approach to culture, but instead pointed to the high levels of cultural capital it demands and argued that it reflected a reconfiguration of existing social and artistic hierarchies.

Ollivier’s use of interviews to discern a typology of tasting techniques also has its counterpart in quantitative approaches. Survey research by Daenekindt and Roose (2014), for example, demonstrated that distinction among art museum visitors can be located in “ways of preferring”. The authors employed latent class analysis of preferred artists to identify critical, functional, modernist and postmodern consumption aesthetics as techniques which were clustered to socio-demographic variables. As with Ollivier, their research points to the new ways in which Bourdieu’s conception of cultural fields are currently being reconfigured and introduces new levels of complexity which go beyond a simple dichotomy of highbrow and lowbrow tastes.

As ways of hearing and experiencing music, “techniques’ of taste also reflect the acquisition of particular embodied ways of engaging with music. Crossley (2015), for example, argues that the acquisition of a new listening technique involves learning “a new [bodily] way of deriving pleasure from a particular type of music”. Beyond Bourdieu’s homology between social space and taste, a correspondence emerges between different “music worlds’ and the particular visceral bodily conventions which are necessary to appropriate them. And rather than treating these capacities as an “accomplishment” to be explained by social context, Hennion (2015) argues that the development and deployment of bodily modes of appreciation should be analyzed in their own right. For Glitsos (2019) this extends to the ways in which the bodily experience of music is produced not just by the conventions of particular styles, but by the materialities of our increasingly digital listening environments. In the case of music recommendation algorithms, she speaks positively of the exciting possibilities they represent – enabling creative listeners to “explore new affective potentials’ (Glitsos, 2019, p. 115).

While not all commentators agree with the emancipatory picture of algorithms portrayed by Glitsos, her emphasis does reflect the increasing attention given to the role played by algorithms in shaping our listening environments. In surveying criticisms of streaming’s effects on music culture, Hesmondhalgh (2022) considers the tension between pessimistic voices decrying the mindlessness encouraged by such algorithms and research which suggests listeners actually exhibit surprising levels of reflexivity about their engagement with algorithmic logics. The current paper does not seek to take sides in this debate but instead seeks to provide quantitative scrutiny of what occurs in the “encounter with difference” (Hanrahan, 2018) which algorithms facilitate on digital music platforms.

Research questions

Drawing on the preceding discussion, the analysis of how audiences engage with unfamiliar music in the present study is guided by the following research questions:

• Research Question 1: What are the “techniques of taste” which can be identified in how people value and appreciate music and to what extent do they correlate with sociodemographic characteristics?

• Research Question 2: For audiences encountering algorithmic recommendations of unfamiliar music, how do these techniques of taste combine with familiarity, sociodemographic variables and characteristics of the music itself to predict affective enjoyment?

Materials and methods

Data collection and recruitment

The data for the analysis is taken from an online survey of 350 Australian adults conducted in October 2018. The research questions were addressed by asking participants about: (i) aspects of their general musical behaviours and attitudes toward music; (ii) their evaluations of a personalised set of music recommendations; and (iii) sociodemographic characteristics. A summary of the variables collected for analysis is provided in Table 1 below. Ethics approval for data collection was provided by Western Sydney University’s Human Research Ethics Committee (H12046).

Table 1. Summary of variables.

Level	Variable	Type	Range
Participant	Age	Continuous	18–89
	Gender	Categorical	Male; female^a
	Education level	Ordinal	8-level scale
	Cultural capital [CAMSIS-OZ]	Continuous	0–1
	Social capital [number of occupations]	Continuous	0–1
	Social capital [mean CAMSIS-OZ of occupations]	Continuous	0–1
	Musical interest classification	Categorical	3-level scale
	Agreement with EFA statements (see Table 2)	Ordinal	4-point scale
Music	Style: Modernist ⇔ Minimalist	Continuous	0–1
	Style: Experimental ⇔ Traditional	Continuous	0–1
	Composer	Categorical	750 values
Recommendation Rating	Familiarity	Ordinal	7-point scale
	Liking	Ordinal	7-point scale
	Similarity	Continuous	0–1

^aNon-binary gender options were recorded but excluded from analysis due to low frequency.

To facilitate investigation of the role played by familiarity, a range of different recruitment channels were used with the aim of obtaining participants who reflected a spectrum of familiarity with Australian art music. Participants were recruited through invitations sent to a mailing list of the Australian Music Centre¹ and through links published on the websites of ABC Classic FM² and APRA.³ A classification of familiarity with Australian art music was obtained by asking participants whether they had an existing interest in this particular genre (“high familiarity”), if they otherwise had an interest in classical music (“medium”) or no prior interest in any classical genres (“low”).

Techniques of taste

The analysis of the first research question’s concern with techniques of taste adapts and extends the approach developed by Daenekindt and Roose (2014) for investigating art museum patrons. Drawing on the different approaches to music consumption found in music psychology and sociology of music literature, the survey presented participants with 14 statements (see Table 2) which aimed to capture the various ways in which people relate to, value and incorporate music into their everyday lives. Participants were asked about their level of agreement with each statement using a 4-point scale ranging from “Strongly Disagree” (1) to “Strongly Agree” (4). The data was analyzed using exploratory factor analysis (EFA) to identify latent variables among the 14 statements using polychoric correlation matrices (Holgado–Tello et al., 2010).

Table 2. Statements on modes of musical consumption and exploratory factor analysis loadings.

#	Statement	Factor 1	Factor 2	Factor 3	Factor 4
1	Having a similar taste in music often helps me relate better to friends and colleagues	0.17	0.00	0.03	0.12
2	I am able to enjoy most musical styles or genres	0.28	−0.30	0.33	0.13
3	Music doesn’t have to sound beautiful	−0.04	−0.38	0.74	−0.01
4	I find it difficult to enjoy music unless I feel I understand it	0.02	0.47	−0.06	0.04
5	It is important that music moves me emotionally	0.21	0.64	0.03	0.04
6	Whether music is good or bad is a matter of personal opinion	0.15	0.16	−0.03	0.96
7	I often choose music which will help change my mood	0.54	0.22	−0.02	0.16
8	There is no such thing as bad music	0.01	−0.03	0.08	0.34
9	Listening to certain types of music helps me think	0.59	0.04	0.06	−0.03
10	I often listen to music as an escape from everyday life	0.77	0.20	−0.03	−0.02
11	Generally, it is important that music challenges me in some way	0.09	−0.04	0.49	0.09
12	I can believe music is important without necessarily liking it	0.09	0.02	0.38	0.07
13	Typically, it is important that the music I listen to matches my mood	0.25	0.44	−0.24	−0.05
14	It is important to listen to music which is of our time and place	−0.17	−0.01	0.49	−0.04

Note. Loadings below −0.4 and above 0.4 are highlighted in bold.

Music recommendations

To examine how participants respond to music recommendations as part of the second research question, the researcher developed custom algorithms which presented participants with recommendations from the recorded music collection of the Australian Music Centre (AMC). To support an analysis of the role played by varying degrees of familiarity, two approaches were used to generate the recommendations. The first delivered personalised recommendations in a manner similar to content-based recommender systems by recommending music which shared the highest acoustic similarity with each participant’s existing musical preferences. These preferences were captured by asking respondents to nominate five of their favourite composers or musical artists. These were matched in real-time to a database of over 800,000 iTunes artists to identify those recordings in the AMC collection which had the highest degree of acoustic similarity to music already liked by the participant.

Participants were presented with five of these algorithmically personalised tracks alongside an additional four tracks drawn randomly from the AMC collection. Each track was presented in the form of a 30-second sample and the ordering of tracks was randomised. Participants were not given any indication of the mechanism by which individual recommendations had been generated. After being presented each musical sample they were asked to evaluate the music in terms of both affective liking (using a 7-point scale ranging from 1 (“Do not like at all”) to 7 (“Like a lot”)) and familiarity (using a 7-point scale ranging from 1 (“Not familiar at all”) to 7 (“Very familiar”)).

The acoustic “materiality” of the music being recommended was measured using two variables derived from multivariate acoustic analysis of the AMC recorded music collection. These variables were calculated to capture the maximum amount of variation inherent in the perceptual acoustic space of Australian art music. Each composer featured in the music recommendations was assigned a score on each of the two dimensions – generating a space of relative positions in which the proximity of composers reflects the acoustic similarity of their music. The extraction and analysis of music’s acoustic features in this way is inevitably a reductionist approach, but one which has application in fields ranging from music information retrieval (Corrêa & Rodrigues, 2016) to musicology (Savage & Brown, 2020).

To interpret the position of a composer it is necessary to specify labels which convey the acoustic oppositions which are represented in each of the two dimensions. The approach taken was to plot the positions of a sample of 50 composers and use an inductive approach to identify stylistic trends reflected in the relative positions of the composers. The first dimension was observed to distinguish between the music of Australian composers known for adopting modernist styles at one pole (e.g. Brett Dean), with the opposing pole dominated by composers who work primarily in jazz (e.g. Sandy Evans) or minimalist (e.g. David Chesworth) styles.⁴ This is not to suggest that jazz and minimalism are interchangeable, but that they share acoustic characteristics which distinguish them from other styles in the context of Australian art music. The second continuum was observed as contrasting composers who employ experimental approaches against those working in more traditional idioms. See Chambers (2020) for a full discussion of how these variables were derived and interpreted.

Sociodemographic characteristics

The final set of data collected in the survey related to the sociodemographic characteristics of participants. These variables reflect the research questions’ interest in examining if theories of taste’s basis in socially embodied practices offer a way of understanding variation among participants’ techniques of taste and engagement with unfamiliar music. The approach taken here draws on the measures of social and cultural capital developed by Savage et al. (2013) in their development of a model of social class. Social capital reflects the diversity and nature of the people we interact with in our social networks, which has the potential to influence our cultural tastes. The social nature of our engagement with music, for example, suggests that people with more diverse social networks may, in turn, be habituated to engage with a corresponding diversity of musical styles and have a greater openness to unfamiliar music. Social capital was measured by asking participants to nominate if they socially knew people from 28 occupational categories selected to reflect a spectrum of social stratification as measured using the CAMSIS-OZ scale (Jones & McMillan, 2001). The number of occupations reported by participants, together with the mean status score of all reported occupations, were then calculated as measures of social capital. Cultural capital – as the embodied capacity to engage with culture – was indirectly measured using the CAMSIS-OZ stratification score of the participant’s own occupation alongside their level of education.

Results

The findings below firstly present the results of the exploratory factor analysis (EFA) which discerned three distinct techniques of taste: functional, affective/emotional and intellectual. Participant scores on each of these factors were, in turn, then used as candidate independent variables – along with measures of similarity, familiarity, acoustic materiality and sociodemographics – to predict the affective rankings given to the music recommendations. The first model considers the extent to which liking can be predicted in a model which also includes the level of familiarity (measured on a 7-point scale) participants recorded for each recommendation. A subsequent model considers a more real-world scenario in which a recommendation algorithm would be without such fine-grained familiarity data. This second models examines which factors are most likely to elicit a positive affective response when recommending music to audiences who are unfamiliar with the genre being recommended.

Techniques of consumption

The results of the EFA used to examine how participants value and incorporate music into their everyday lives is shown in Table 2. The number of factors was selected using the parallel analysis criterion method of Humphreys and Montanelli Jr (1975). The resulting four factors were subjected to VARIMAX rotation to aid interpretation. The analysis follows Hair et al. (2019, p. 175) in using a loading cut-off point of ±0.4 for a variable to be included in the interpretation of a factor.

Factors identified in the EFA

The first factor demonstrates substantial loadings from statements which reflect largely functional uses of music: to alter mood, to help think, and to escape everyday life. The identification of a functional technique of taste is of particular interest given the criticism which this mode of listening often attracts. Hesmondhalgh (2022) traces such critiques to an elitism which associates this disposition with a preference for bland and unchallenging music, a passive mode of listening and a lack of musical adventurousness. By identifying this disposition in the EFA, the current study permits empirical scrutiny of the extent to which such assumptions are reflected in participants’ engagement with unfamiliar music.

The second factor places value on the affective and emotional capacities of music: as something which should sound beautiful, stir emotions, and correspond with the listener’s own mood. The importance of understanding music (Statement 4) may appear to reflect a more cerebral response to music, however the statement links it to the capacity to derive enjoyment. The fact that audiences always have access to their emotional response to music – regardless of whether they understand it intellectually – has been shown to have particular importance for audiences seeking to form an attachment to music that is otherwise unfamiliar to them (Chambers, 2023b). The extent to which this embodied response corresponds with an overall positive affect when presented with unfamiliar music – or whether it is more likely to be triggered in response to particular types of music – is investigated in the modelling of participants’ affective ratings below.

Factor 3 represents a more explicitly intellectual frame to the way in which music is valued: emphasising its capacity to challenge the listener, downplaying the importance of beauty and affect, and valuing the music’s cultural significance. Corresponding with Ollivier’s (2008) humanist openness to cultural diversity and Chambers’ (2023b) affective frame of appreciation, it suggests a disposition which will have an increased appreciation for exploring the unfamiliar.

Finally, Factor 4 only includes a single statement above the ±0.4 threshold and on this basis has been excluded from the subsequent modelling. By itself, Statement 6 is less a technique of taste and, instead, is reflective of the extent to which individuals recognise the importance of external referents and hierarchies in how music should be evaluated.

The reliability of the EFA is reinforced by the considerable crossover with the factors identified by Daenekindt and Roose (2014) in their study aimed at discerning “aesthetic dispositions’ as ways of preferring among visitors to art galleries. Their functional and postmodern aesthetics align closely with Factor 1 and Factor 4 respectively. Whereas their emphasis on distinct artistic aesthetics led them to distinguish separate modernist and critical dispositions, these have been subsumed in the present EFA into the general intellectual frame of Factor 3. Similarly, Factor 2 is largely present as a subset of Daenekindt and Roose’s functional aesthetic, whereas musical appreciation warrants considering it as a separate technique of taste.

Locating factors in sociodemographic variables and musical taste

Of the three factors identified above, only Factor 1 (functional) was shown to have a moderate correlation with the demographic characteristics of participants. Small but significant effects were observed for age, social capital and cultural capital – suggesting that an orientation towards adopting a functional disposition toward engaging with music was more commonly found among younger, less culturally elite and less socially connected participants. This finding resonates with Hesmondhalgh’s (2022) argument that the critique of streaming platforms for facilitating functional consumption of music has its origins in elitist anxieties.

There was also only very limited correlation between the factors and the three-level classification of musical interest in Australian art music. The only significant effect was among those with an interest in traditional classical music, who were less likely to employ the intellection frame of Factor 3 in their listening habits. This finding that different techniques of taste do not appear to neatly align with different objects of taste is significant for our understanding of musical preferences. The same techniques can be used to value and appreciate different music and, conversely, the same music can be valued in different ways. At the same time, the presence of only moderate levels of correlation between the factors and particular demographic variables only provides limited support for the notion that different modes of consumption might be the outcome of particular modes of socialisation.

Recommending unfamiliar music

Predicting affective engagement with recommended music

Having identified three broad “techniques of taste” in the factors above, the analysis turns to the degree to which these are significant in predicting positive engagement with the samples of art music recommended to participants. Linear mixed-effects models were fitted to the data, with affective liking scores (on a 7-point scale) as the dependent variable. Candidate independent variables comprised (i) participant assessments of their familiarity with the recording (on a 7-point scale), (ii) participant scores on the three factors identified by EFA, (iii) sociodemographic characteristic, (iv) classification of the participant’s musical interest (Australian art music; traditional classical; non-classical); and (v) two variables reflecting the acoustic materiality of the recording being recommended (modernism vs jazz/minimalism influences; experimental vs traditional idioms). The individual participant and the composer of the recording were modelled as random effects.

The optimal model, specified in Table 3 below, shows that while familiarity is significant in predicting affective liking, the effect is moderate. Each scale-point increase in a participant’s familiarity with a recording increases their corresponding liking by just under 0.2. The impact of familiarity is magnified, however, for those who favour an affective/emotional disposition (Factor 2) toward music. This suggests that a positive emotional response to music is a habitual embodied responses to familiar sounds and patterns. The “musical affordances’ which make this response possible are highly specific and learned over time, which challenges their capacity to derive similar enjoyment from music which is unfamiliar. Furthermore, a high degree of experimentalism in a composer’s music exacerbates the effect of unfamiliarity on producing lower affective liking ratings. It is one thing for music to be unfamiliar, but to be both experimental and unfamiliar is regarded as worse.

Table 3. Linear mixed-effects model predicting liking with perceived familiarity information.

Fixed effects	Coefficient	SE	T-statistic	P-value
Intercept	3.968	0.367	10.825	<.001***
Familiarity	0.195	0.063	3.077	<.01**
Musical Interest: Traditional classical	−0.248	0.209	−1.187	>.1
Musical Interest: Non-classical	−0.472	0.169	−2.794	<.01**
Musical style: Jazz/Minimalism	1.032	0.240	4.308	<.001***
Musical style: Experimentalism	−2.208	0.500	−4.418	<.001***
“Affective/Emotional” Technique (Factor 2)	−2.017	0.507	−3.979	<.001***
Familiarity * Experimentalism	0.366	0.096	3.817	<.001***
Familiarity * Factor 2	0.324	0.087	3.705	<.001***
Random Effects	Variance	SD
Participant	1.036	1.018
Composer	0.070	0.265

Conditional R2 of 66.6%, in which the fixed effects (marginal R2) explain 38.1% of the variance. ** p < 0.01 ; *** p < 0.001

While the model shows that having a prior interest in the genre being recommended results in higher liking ratings, this effect is surprisingly small. There was no difference between those who declared an interest in Australian art music as opposed to those who were only interested in the broader genre of traditional classical music. Participants who identified as having no interest in any classical styles gave lower liking ratings, but the size of this effect is surprisingly small given contemporary art music’s traditional reputation for being difficult and impenetrable. This suggests that the barriers to engaging audiences to explore unfamiliar styles through recommender services may not be as substantial as might be assumed.

Finally, the model also identifies that certain stylistic sub-genres of Australian art music are more likely to be positively received than others. Liking ratings are considerably higher when responding to the music of composers operating in jazz and minimalist idioms as opposed to complexity and modernism. Moreover, high degrees of experimentalism have an even more pronounced negative effect. These findings highlight the importance of considering which specific stylistic trends are most useful as gateways for users exploring a genre.

Recommending an unfamiliar genre

While the previous model helps understand the role of perceived familiarity in the context of affective responses to individual musical samples, the task of making real-world recommendations to users is without such fine-grained data on a user’s familiarity with an as yet unheard piece of music. What recommendation algorithms do typically have access to, however, is the acoustic similarity of candidate recommendations to the music frequently listened to by a user. If a piece of music sounds similar to the music you frequently encounter, it will likely have some degree of familiarity.

The model in Table 4 below considers the challenge of making a recommendation to someone who has minimal previous interest in a particular style of music. To do this it considers only the responses of those participants who nominated their musical interest classification as not encompassing any form of classical or art music. Furthermore, it attempts to better reflect the real-world conditions of recommender systems by replacing the familiarity data used in the previous model with a measure of the similarity of each recording to the music commonly listened to by each participant.

Table 4. Linear mixed-effects model predicting liking without familiarity for non-classical participants.

Fixed effects	Coefficient	SE	T-statistic	P-value
Intercept	4.066	0.648	6.272	<.001***
Musical style: Jazz/Minimalism	1.366	0.549	2.487	<.05*
Musical style: Experimentalism	−2.557	0.642	−3.981	<.001***
“Intellectual” Technique (Factor 3)	1.695	0.685	2.476	<.05*
Acoustic Similarity	0.667	1.376	0.485	>.1
Acoustic Similarity * Cultural capital [CAMSIS-OZ]	−6.235	2.167	−2.877	<.01**
Random Effects	Variance	SD
Participant	1.264	1.124
Composer	0.297	0.545

Conditional R2 of 55.4%, in which the fixed effects (marginal R2) explain 17.7% of the variance. ** p < 0.01 ; *** p < 0.001

The results suggest that the measure of similarity used in this model does not have the same predictive power as previously observed for familiarity in predicting whether participants will have a positive affective response to unfamiliar music. The model only demonstrates a significant effect of similarity as an interaction with high values of cultural capital. This suggests that the capacity for acoustic dissimilarity (i.e. hearing music whose acoustic qualities are unlike those that a listener commonly enjoys) to result in lower liking scores is dependent on having an occupation associated with higher levels of social stratification. For those occupying jobs with lower levels of cultural capital, however, the impact of acoustic similarity is insignificant. These findings should be considered alongside the fact that the sample being considered is those who have no interest in either contemporary Australian or traditional classical music – both domains which have traditionally been associated with cultural elites (Bennett et al., 1999). For elites who do not have this prototypical interest in classical music, they are more likely to hold steadfast to their own musical tastes and be more resistant to exploring beyond them. Instead of “anything but heavy metal” (Bryson, 1996), it suggests a case of “anything but classical” by which this group uses a dislike to form a symbolic boundary.

The other main effects found in this model point, firstly, to the stylistic considerations of the music being recommended. For those unfamiliar with this esoteric sub-genre, it shows that liking ratings increase significantly both for music by composers working in jazz and minimalist idioms (as opposed to complexist modernist styles) and for those composers who are less experimental in their approach. This points to the importance of selecting particular “easier” styles over others when introducing new listeners to a genre. Finally, the positive coefficient for the intellectual mode of musical consumption (Factor 3), reinforces the notion of art music’s approachability to those who regard themselves as disposed to a disinterested appreciation of music.

Discussion

The results above demonstrate the nuanced way in which people’s familiarity with music – in terms of both particular works and the broader codes and schemas of perception they exemplify – interacts with both techniques of taste and the stylistic qualities of the music encountered to influence affective responses in the context of digital music recommendation. The substantial effects attributable to elements of musical style, even after any familiarity with that style is taken into account, draw attention to the importance of Hennion’s “return to the work” in the consideration of preferences. Rather than considering genres, or even sub-genres, as homogenous and largely interchangeable forms, they instead contain continuums of musical characteristics which provide varying levels of “musical affordance” (Krueger, 2011) for affective engagement by different groups. It is possible that these stylistic qualities have some form of symbolic association for the listeners which partly shapes their responses, but this ambiguity points to the need for more detailed research into the processes by which attachments are formed to some but not all unfamiliar music. This echoes earlier qualitative findings, such as the ethnographic work of DeNora (2000), which explores how particular techniques are adopted in relation to particular types of musical material. Such work does not seek to disregard the “social and behavioural entailments’ (DeNora, 2000, p. 141) of particular forms of music, but suggests a more holistic way in which they can be considered.

The findings also inform our understanding of the characteristics of musical omnivores when conceptualised as people having an affective openness to engaging with unfamiliar music. When considering how the three techniques of taste identified in EFA influenced the liking of unfamiliar music recommended to participants, it is interesting to note that Factor 3 (intellectual) did not have a significant effect. While an omnivorous sensibility is often linked to modes of appreciation which draw on critical and aesthetic criteria—such as the critics of mass culture found among Ollivier’s (2008) “humanist openness’ omnivores or those whom Hesmondhalgh (2022) identifies as lamenting the lazy modes of consumption fostered by streaming platforms—participants who were associated with this technique of taste did not respond more positively to unfamiliar recommendations than those who adopted the functionalist technique (Factor 1). Instead, the technique associated with the least appetite for adventurous musical listening was those who valued music for its embodied affective and emotional qualities (Factor 2). As music become more unfamiliar, this group gave lower ratings to the music they encountered. Crossley (2015) points to the learned nature of body techniques through which people develop affective preferences and this finding reinforces the idea that the capacity of bodies to have visceral embodied responses to music is not general and transferrable but is developed as a modality of appreciation for specific musical styles.

More generally, however, the findings point to the need to rethink the relationship between familiarity and preferences, particularly in the context of contemporary consumption environments which increasingly involve the algorithmic mediation of culture. Perceived familiarity with music recommendations was significant in explaining the affective rating given by participants, however the size of this effect was relatively small. Even participants with no interest in contemporary art music were not simply alienated by the music they heard. This firstly raises questions as to the assumptions of both Meyer and Bourdieu regarding the necessary link between mastery of codes and meaning. It also highlights that research which emphasises participant knowledge of different genres overlooks the capacity of participants to engage with and derive pleasure from music whose stylistic codes they are not yet acquainted with. While the processes of developing knowledge of diverse styles is of interest, particularly with regard to the social capital it may demand, it should not be confounded with the capacity for people to employ a mode of cultural openness.

The specific techniques of engaging with music identified by way of exploratory factor analysis provide a novel way for considering the different ways by which people engage with, value and utilise music in their everyday lives. Whereas Daenekindt and Roose (2014) sought to align their analysis of art museum audiences with particular aesthetic dispositions, the consumption of music spans a broader range of cultural forms and most often does so without the institutional framing of the museum. The interpretation of latent factors, therefore, points to the more general ways in which the functional, emotional, and intellectual dispositions guide interactions with music. Furthermore, the results showed that these techniques cut across different musical interests: the same music could be appropriated by different tasting techniques. Unlike Bourdieu's homology between social space and cultural tastes, only a moderate correlation was observed between tasting techniques and sociodemographic variables. The findings offer little evidence to locate the adoption of these techniques in particular modes of socialisation and associated processes of distinction.

Moving to the real-world challenge of algorithmic music recommendations, the study firstly demonstrates the limited utility of acoustic similarity to act as a proxy for familiarity – at least in the case of the analyzed scenario where contemporary Australian art music was recommended to people without a prior interest in this genre. Instead, similarity only played a role among those whose occupations were among the social elite. Rather than displaying omnivorous openness to music dissimilar to their existing tastes, these individuals were more entrenched in their opposition to unfamiliar sounds – reinforcing the notion that dislikes form an important role in social distinction (Bryson, 1996) and disputing any straightforward link between cultural openness and the cultural elite. While participants who were inclined toward adopting the intellectual (Factor 3) technique for consuming music were slightly more likely to respond positively to recommendations, the acoustic characteristics of the music itself had the largest effect. The success with which listeners can be introduced to a new genre is therefore likely to be dependent on the selection of music which serves as effective entry points based on having particular musical qualities.

Rather than characterising the influence of algorithmic logics on streaming platforms as positive or negative, the findings echo the call by Hesmondhalgh (2022) for greater attention to “what people do with streaming”. The contrast between a critically minded adventurous listener and a listener who engages in a functional and supposedly passive mode of listening was not shown to be significant in predicting the affective ratings given to recommendations of unfamiliar. While the present study is limited to short-term affective responses in an artificial listening context, it does point to there being fewer obstacles for algorithms to be developed in directions which approximate the kind of emancipatory ideal posited by Glitsos (2019) in her theorisation of the “personal computer bodymind somatechnic”. While the algorithms which dominate mainstream digital music platforms may not yet reflect this ideal, the findings lend support to giving users greater autonomy over the music which is recommended to them.

Overall, the results emphasise the importance of going beyond notions of similarity and familiarity in theorising patterns of consumption, particularly for those with a vested interest in not only understanding but encouraging engagement with the unfamiliar. The ways in which individuals develop particular techniques, and the ways in which they then function as capacities able to derive affective appreciation from different styles of music, warrant further investigation. Extending this programme of research to explore more diverse genres and in less experimental environments of data collection will aid in building a stronger base of empirical data on which these issues can be understood.

Acknowledgements

I wish to thank Associate Professor Liam Magee at Western Sydney University’s Institute for Culture and Society for his feedback on earlier drafts of this article.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The Australian Music Centre is a national service organisation promoting Australian art music. Its mailing list subscriber base will likely have a high degree of familiarity with this genre.

2 ABC Classic FM is Australia's largest classical radio network, broadcasting predominantly pre 20th century music. Participants recruited through this channel will likely be familiar with the broader genre of classical music, but are unlikely to be well-acquainted with Australian art music.

3 The Australasian Performing Right Association (APRA) is an organisation whose members predominantly have an interest in non-classical genres and are likely to be unfamiliar with Australian art music.

4 The music of composers working in the modernist tradition is characterized by features such as dissonance, complexity and non-traditional tonality. Music by minimalist composers typically includes a focus on the modulation of repetitive rhythmical elements.