Personality Traits and Music Genre Preferences: How Music Taste Varies Over Age Groups

Abstract

Personality traits are increasingly being incorporated in systems to provide a personalized experience to the user. Current work focusing on identifying the relationship between personality and behavior, preferences, and needs o‰en do not take into account di‚erences between age groups. With music playing an important role in our lives, di‚erences between age groups may be especially prevalent. In this work we investigate whether di‚erences exist in music listening behavior between age groups. We analyzed a dataset with the music listening histories and personality information of 1415 users. Our results show agreements with prior work that identi€ed personality-based music listening preferences. However, our results show that the agreements we found are in some cases divided over di‚erent age groups, whereas in other cases additional correlations were found within age groups. With our results personality-based systems can provide beŠer music recommendations that is in line with the user's age.

CCS CONCEPTS

•Information systems ! Recommender systems; •Human-centered computing ! User models; User studies;

INTRODUCTION

Personality has shown to be a stable construct over time, and reƒects the coherent paŠerning of one’s a‚ect, cognition, and desires (goals) as it leads to behavior [ 24 ]. Œis stability and coherency of personality has shown to be useful for systems to infer users’ preferences and to provide personalized experiences to users (e.g., [ 8 ]). Hu & Pu [ 19 ] showed that personality-based personalized systems have an advantage over personalized systems not incorporating personality information in terms of increased users’ loyalty towards the system and decreased cognitive e‚ort.

Œe relationships between personality traits and users’ behavior preferences and needs are increasingly being investigated (e.g., health [ 17, 26 ], education [ 2, 21 ], movies [ 3 ], music [ 8–10, 14, 27 ]). Œese studies normally analyze their sample as a whole and do not consider di‚erences based on age groups. ArneŠ [ 1 ] showed that especially those in their adolescence and emerging adulthood phases experience a heightened chance of ”storm and stress” 1 in which they try to €nd their place in society. Hence, di‚erences may occur in behavior, preferences, and needs throughout di‚erent phases in life.

To investigate the relationship between personality and music genre preferences over di‚erent age groups, we used a subset of the myPersonality dataset. Next to users’ personality scores, this subset consist of the listening history of Last.fm (an online music streaming service) 2 users. By analyzing the listening histories of 1068 users in relation to their personality and age, we found important di‚erences across age groups. Our insights may help to inform personalized music systems. For example, personalitybased music recommender systems can improve their cold-start recommendations (e.g., [ 5, 28 ]) by beŠer knowing which music genres to recommend to their users of di‚erent age groups. 2

RELATED WORK

Currently, there are two di‚erent personality related research directions focusing on: 1) personality-based personalization (e.g., health [ 17, 26 ], education [ 2, 21 ], movies [ 3 ], music [ 8–10, 14, 27 ]) and 2) implicit personality acquisition from user-generated content (e.g., Facebook [ 12, 16 ], TwiŠer [ 22 ], Instagram [ 11, 13 ], and fusing information [ 25 ]). For example, in the area of personality-based personalization Ferwerda et al. [ 15 ] looked at di‚erences in how users browse for music (i.e., browsing music by genre, activity, or mood) in an online music streaming service. Chen, Wu, & He [ 3 ] investigated diversity preferences in movie recommendations. In the area of implicit personality acquisition research mainly focuses on user-generated content of users’ social media accounts. ‹ercia et al. [ 22 ] found that how users behave on TwiŠer consist of cues to predict their personality. Similarly, Golbeck, Robles, & Turner [ 16 ] were able to develop a personality predictor based on the characteristics of a user’s Facebook account.

Current personality-based research does not take into account di‚erences between age groups. However, ArneŠ [ 1 ] notes that especially those in their adolescence and emerging adulthood phases may show deviant behavior. With music been shown to play an important role in our lives by providing support for a whole range 1Storm-and-stress is a term €rst coined by Hall [ 18 ] to refer to a period in life in which people experience turmoil and diculties. 2hŠp://www.last.fm/ of daily activities we engage in (e.g., sports, studying, sleeping) [ 23 ], di‚erences (e.g., listening behavior, preferences, and needs) across age groups may especially be prevalent.

In this work we analyze a dataset of an online music streaming service consisting of the total listening history of their users. With this dataset we investigate whether di‚erences in music listening behavior exist. 3

METHOD

In order to investigate the relationship between personality and music genre preferences across age groups in an online music streaming service, we made use of the myPersonality dataset. 3 Œe dataset originates from a popular Facebook application (”myPersonality”) that is able to record psychological and Facebook pro€les of users that used the application to take psychometric (e.g., personality, attitudes, skills) tests. Œe dataset contains over 6 million test results, with over 4 million Facebook pro€les. Users’ personality in the myPersonality application was assessed using the Big Five Inventory to measure the constructs of the €ve factor model: openness, conscientiousness, extraversion, agreeableness, and neuroticism.

We only used the subset of the myPersonality dataset that contains the music listening history of Last.fm users (i.e., play-count of artists that a user listened to) and the day of birth in order to calculate their age. Œe subset consists of users’ complete listening histories (i.e., from the moment they started to use Last.fm) until April 27 (2012). We complemented the dataset by adding the listening events of each user until December 18 (2016) by using the Last.fm API. 4 A total of 1066 Last.fm users with 40 million listening events from 101 countries are represented in the subset.

Œe 1066 Last.fm users were split into three di‚erent age groups according to the primary life stages [ 4 ]: adolescence (age: 12-19), young adulthood (age: 20-39), and middle adulthood (age: 40-65). Having the day of birth of the 1066 Last.fm users as well as their complete listening history (with listening date), we could traceback users’ age when listened to a certain song. Hence, users could fall into multiple age groups, which resulted in a sample size bigger than the original sample. Œe €nal dataset consists of 1479 Last.fm users divided over three age groups (adolescence: n =581, young adulthood: n =850, middle adulthood: n =48).

Œrough the Last.fm API, we crawled additional information about the artists by using the ”Artist.getTopTags” endpoint. Œis endpoint provided us with all the tags that users assigned to an artist, such as instruments (“guitar”), epochs (“80s”), places (“Chicago”), languages (“Swedish”), and personal opinions (“seen live” or “my favorite”). Tags that encode genre or style information were €ltered for each artist. Œe €ltered tags were then indexed by a dictionary of 18 genre names retrieved from Allmusic. 5 For each user in an age group, the artists that were listened to were aggregated by the indexed genre with their play-count. Œe genre play-count for each user was then normalized to represent a range of rϵ[ 0,1 ], this in order to be able to compare users with di‚erences in the total amount of listening events. 3hŠp://mypersonality.org/ 4hŠp://www.last.fm/api 5hŠp://www.allmusic.com For the analysis we €ltered out users with zero play-counts (users who registered, but did not make use of Last.fm) and people listening to less than €ve di‚erent artists. Œis le‰ us with a total of 1068 users (adolescence: n =472, young adulthood: n =563, middle adulthood: n =33). Œe normalized genre play-counts by the di‚erent age groups are shown in Figure 1.

To investigate music listening di‚erences between personality traits, Spearman’s correlation was computed between personality traits and the genre play-count to assess the relationship of personality and genre preferences. Alpha levels were adjusted using the Bonferroni correction to limit the chance on a Type I error. Œe reported signi€cant results adhere to alpha levels of p <.001 (see Table 1). Œe results of music genre preferences per personality trait per age groups are discussed in the following sections. 4.1

Openness

Adolescence (age: 12-19). For the adolescence group, several positive correlations were found with music genres: New Age (r =.142), Blues (r =.130), Country (r =.117), World (r =.114), Folk (r =.230), Jazz (r =.139), Vocal (r =.132), and Alternative (r =.131). Œose scoring high on the openness scale show in general higher listening behavior to these music genres.

Young adulthood (age: 20-39). For the young adulthood group, we found the same kind of positive correlations with music genres: New Age (r =.105), Blues (r =.167), Country (r =.126), World (r =.217), Folk (r =.231), Jazz (r =.106), Vocal (r =.170), and Alternative (r =.116). An additional positive correlation was found with Electronic (r =.106), and a negative correlation with Rock (r =-.104) indicating that those scoring high on openness tend to listen less to Rock music in this age group. Next to Folk music, a stronger correlation was found for World music as well.

Middle adulthood (age: 40-65). When those scoring high on openness reach middle adulthood, their variation of listening to music genres shrinks signi€cantly, but the strength of the correlations increases. We found positive correlations with Blues (r =.358) as well as with Folk (r =.368) music. Indicating that although the variation gets less, the music genre preferences for middle adulthood becomes more prominent. Adolescence (age: 12-19). Œose scoring high on conscientiousness in the adolescence group show mainly negative correlations with the listened music genres: Reggae (r =-.102), Punk (r =-.130), and Alternative (r =-.108). Œe results indicates that for this age group, conscientious users listen less to these music genres. Young adulthood (age: 20-39). Negative correlations were also found between music genres and conscientiousness for the young adulthood group. Although the Punk (r =-.103) and Alternative (r =-.108) music genre is in line with the adolescence group, instead of Reggae, this group listens less to Folk (r =-.114) music.

Middle adulthood (age: 40-65). We found two correlations for

the middle adulthood group: Jazz (r =.510) and Alternative (r =.507). Both correlation coecients show high e‚ects between conscientiousness and the music genres. 4.3

Extraversion

Adolescence (age: 12-19). Œe adolescence group show negative correlations with Classical (r =-.136), World (r =-.102), Punk (r =-.111), and Metal (r =-.148). A positive correlation was found with R&B (r =.106). Œe results indicate that extraverts in their adolescence phase listen less to Classical, World, Punk, and Heavy Metal music. However, the in general listen to more R&B.

Young adulthood (age: 20-39). For those scoring high on ex

traversion and fall in the young adulthood group show negative correlations with Rock (r =-.102), New Age (r =-.184), Classical (r =-.146), and Heavy Metal (r =-.126). A positive correlation was found with Rap (r =.108) music.

Middle adulthood (age: 40-65). Œe middle adulthood group

of the extraverts show a positive correlation with R&B (r =.326) and a negative correlation with Heavy Metal (r =-.339). 4.4

Agreeableness

Adolescence (age: 12-19). Œe adolescence group show only a positive correlation with agreeableness for Folk (r =.101) music. Œis indicates that agreeable users show in this age group show a preference for Folk music.

Young adulthood (age: 20-39). A more varied music preference is shown for the young adulthood group. Positive correlations were found for Country (r =.184), Folk (r =.110), and Pop (r =.194) music. A negative correlation was found for Heavy Metal (r =-.105) music. Agreeable users in their young adulthood phase seem to prefer to listen to Country, Folk, and Pop, but less to Heavy Metal music.

Middle adulthood (age: 40-65). Œe middle adulthood group

show a negative correlation with Heavy Metal (r =-.339) music, which indicates that their preference to listen to Heavy Metal goes down when reaching the age of middle adulthood. 4.5

Neuroticism

Adolescence (age: 12-19). Neurotics in their adolescence phase show positive correlations with Punk (r =.101) as well as with Alternative (r =.129) music, indicating an increase preference for these music genres.

Young adulthood (age: 20-39). Only a positive correlation with Alternative (r =.137) was found in the young adulthood group. Middle adulthood (age: 40-65). For the middle adulthood group, music preferences seem to switch. A positive correlation was found with Heavy Metal (r =.372) and a negative correlation was found with Blues (r =-.552). 5

DISCUSSION

Our results show that there are di‚erences in music listening behavior between personality traits, and that these di‚erence can be further broken down by age groups. Overall, our results show that users in their adolescence and young adulthood phases show most variation in their music listening behavior. Not only does the variation become much less when reaching middle adulthood, the correlation strength increase signi€cantly. Œis indicates that music preferences of the middle adulthood group becomes more established, which is in line with the storm-and-stress argument [ 1 ].

Œe openness trait shows most variation in listening to di‚erent music genres amongst the personality traits. Œis is in line with one of the few works that investigated the relationship between personality traits and music listening behavior [ 23 ]. However, what their €ndings do not show is that there are di‚erences when considering age groups. For example, the addition of a preference for Electronic music in the young adulthood group.

Also the conscientiousness trait shows agreement with prior work [ 20 ]. However, additional unique correlations were able to be identi€ed when taking di‚erent age groups into account. Our results show that the adolescence group shows an additional negative correlation with Reggae, the young adulthood group shows an additional correlation with Folk music, and the middle adulthood group shows an additional positive correlation with Jazz music.

Our results on extraversion show agreements with prior works [ 20, 23 ] as well. However, what the results of prior works do not show is that there is a division based on age. For example, our results show that the positive correlation of R&B and Rap, di‚er across age groups. Œe adolescence and middle adulthood group show positive correlations only with R&B, whereas only a positive correlation with Rap was found with the young adulthood group.

For the agreeableness trait, we found agreements with prior work [ 23 ] especially for the young adulthood group show: positive correlations with Country, Folk, and Pop music. Œese full agreements seem to only hold for the young adulthood group. We found less agreements with the adolescence and the middle adulthood group: only Folk music showed to be positively correlated.

Œe agreements we found with prior work [ 20 ] on neuroticism are divided across age groups. Whereas prior works showed grouped correlations with Punk, Alternative, and Heavy Metal music on neuroticism, our results show that these correlations do not hold for all age groups. We show that Punk and Alternative music is positively correlated with neuroticism for adolescence, but only Alternative music is positively correlated with neuroticism in the young adulthood group. Moreover, we show only a positive correlation with Heavy Metal in the middle adolescence group. 6

CONCLUSION & IMPLICATIONS

In this work we investigated whether there are di‚erences across age groups in the relationship between personality traits and music genre preferences. When not considering di‚erences across age groups, we show that we found agreements with prior works [ 20, 23 ] on the relationship between personality and music genre preferences. Whereas prior works analyzed their sample as a whole, we show with our results that di‚erences exist in music genre preferences depending on age groups. With our results we validate the results of prior works, but show that there are cases where the previously found correlations with music preferences are divided over di‚erent age groups, whereas in other cases other (previously unrevealed) correlations show up within age groups.

Our work contributes to the personality-based work for personalized systems. Œe di‚erences between age groups that we identi€ed in this work may have important implications for the creation of personalized systems. Œe focus of the recommendations may di‚er depending on the age groups a user falls in. For example, the recommendations for adolescent extraverts could be focused on R&B music, whereas recommendations for extraverts in their young adulthood could be more focused on Rap music.

For our future work, we will extend our €ndings by actually trying to provide music recommendations to users and perform a user-centric evaluation on the recommendations. For example, including diversity in recommendations have shown to be an important feature on satisfaction [ 29 ]. In addition, Ferwerda et al. [ 6, 7 ] identi€ed the prerequisites for diversi€cation and found di‚erences in diversity needs among personality traits. Our €ndings could help to inform the diversi€cation in recommendations by incorporating di‚erent needs across age groups. For example, those scoring high on openness in their adolescence or young adulthood phases may be given more diverse genres (correlations were found with eight and ten di‚erent genres respectively), whereas the recommendations for those in their middle adulthood can be narrowed down to Blues and Folk.

In this work, we also did not take into account possible cultural di‚erences. Although having the music listening histories of users from di‚erent countries, we disregarded country information in order to keep a big enough sample. In future work we will address possible cultural di‚erences. 7

ACKNOWLEDGMENTS

Supported by the Austrian Science Fund (FWF): P25655. We would also like to thank Michal Kosinski and David Stillwell of the myPersonality project for sharing the data with us.

References

[1] Je‚rey Jensen ArneŠ. 1999 . Adolescent storm and stress, reconsidered . American psychologist 54 , 5 ( 1999 ), 317 .

[2] Guanliang Chen , Dan Davis, Claudia Hau‚, and Geert-Jan Houben . 2016 . On the impact of personality in massive open online learning . In Proceedings of the 2016 conference on user modeling adaptation and personalization. ACM , 121 - 130 .

[3] Li Chen , Wen Wu , and Liang He . 2013 . How personality inƒuences users' needs for recommendation diversity? . In CHI'13 Extended Abstracts on Human Factors in Computing Systems. ACM , 829 - 834 .

[4] Erik H Erikson . 1993 . Childhood and society . WW Norton & Company.

[5] Ignacio Ferna ´ndez-Tob´ıas , MaŠhias Braunhofer, Mehdi Elahi, Francesco Ricci, and Iva´n Cantador . 2016 . Alleviating the new user problem in collaborative €ltering by exploiting personality information . User Modeling and User-Adapted Interaction 26 , 2 - 3 ( 2016 ), 221 - 255 .

[6] Bruce Ferwerda , Mark Graus, Andreu Vall, Marko Tkalcic, and Markus Schedl . 2016 . Œe inƒuence of users' personality traits on satisfaction and aŠractiveness of diversi€ed recommendation lists . In 4 th Workshop on Emotions and Personality in Personalized Systems (EMPIRE) 2016 . 43 .

[7] Bruce Ferwerda , Mark P Graus , Andreu Vall, Marko Tkalcic, and Markus Schedl . 2017 . How item discovery enabled by diversity leads to increased recommendation list aŠractiveness . In Proceedings of the Symposium on Applied Computing. ACM , 1693 - 1696 .

[8] Bruce Ferwerda and Markus Schedl . 2014 . Enhancing Music Recommender Systems with Personality Information and Emotional States: A Proposal. . In UMAP Workshops.

[9] Bruce Ferwerda and Markus Schedl . 2016 . Personality-Based User Modeling for Music Recommender Systems . In Joint European Conference on Machine Learning and Knowledge Discovery in Databases . Springer, 254 - 257 .

[10] Bruce Ferwerda , Markus Schedl, and Marko Tkalcic . 2015 . Personality & Emotional States: Understanding Users' Music Listening Needs. . In UMAP Workshops.

[11] Bruce Ferwerda , Markus Schedl, and Marko Tkalcic . 2015 . Predicting personality traits with instagram pictures . In Proceedings of the 3rd Workshop on Emotions and Personality in Personalized Systems 2015. ACM , 7 - 10 .

[12] Bruce Ferwerda , Markus Schedl, and Marko Tkalcic . 2016 . Personality traits and the relationship with (non-) disclosure behavior on facebook . In Proceedings of the 25th International Conference Companion on World Wide Web. International World Wide Web Conferences Steering CommiŠee , 565 - 568 .

[13] Bruce Ferwerda , Markus Schedl, and Marko Tkalcic . 2016 . Using instagram picture features to predict users€ personality . In International Conference on Multimedia Modeling . Springer, 850 - 861 .

[14] Bruce Ferwerda , Marko Tkalcic, and Markus Schedl . 2017 . Personality Traits and Music Genres: What Do People Prefer to Listen To? . In Proceedings of the 25th Conference on User Modeling, Adaptation and Personalization. ACM , 285 - 288 .

[15] Bruce Ferwerda , Emily Yang, Markus Schedl , and Marko Tkalcic . 2015 . Personality traits predict music taxonomy preferences . In Proceedings of the 33rd Annual ACM Conference Extended Abstracts on Human Factors in Computing Systems. ACM , 2241 - 2246 .

[16] Jennifer Golbeck , Cristina Robles, and Karen Turner . 2011 . Predicting personality with social media. In CHI'11 extended abstracts on human factors in computing systems . ACM, 253 - 262 .

[17] Sajanee Halko and Julie A Kientz . 2010 . Personality and persuasive technology: an exploratory study on health-promoting mobile applications . In International Conference on Persuasive Technology . Springer, 150 - 161 .

[18] Granville Stanley Hall . 1916 . Adolescence: Its psychology and its relations to physiology, anthropology, sociology, sex, crime, religion and education . Vol. 2 . D. Appleton .

[19] Rong Hu and Pearl Pu . 2009 . Acceptance issues of personality-based recommender systems . In Proceedings of the third ACM conference on Recommender systems. ACM , 221 - 224 .

[20] Alexandra Langmeyer , Angelika Guglho¨ r- Rudan , and Christian Tarnai . 2012 . What do music preferences reveal about personality? Journal of Individual Di‚erences ( 2012 ).

[21] Michael J Lee and Bruce Ferwerda . 2017 . Personalizing online educational tools . In Proceedings of the 2017 ACM Workshop on Š eory-Informed User Modeling for Tailoring and Personalizing Interfaces . ACM, 27 - 30 .

[22] Daniele ‹ercia, Michal Kosinski, David Stillwell, and Jon Crowcro‰. 2011 . Our TwiŠer pro€les, our selves: Predicting personality with TwiŠer . In Proceedings of the International Conference on Social Computing (SocialCom) . IEEE, 180 - 185 .

[23] Peter J Rentfrow and Samuel D Gosling . 2003 . Œe do re mi's of everyday life: the structure and personality correlates of music preferences . Journal of personality and social psychology 84 , 6 ( 2003 ), 1236 .

[24] William Revelle . 2009 . Personality structure and measurement: Œe contributions of Raymond CaŠell . British Journal of Psychology 100 , S1 ( 2009 ), 253 - 257 .

[25] Marcin Skowron , Marko Tkalcˇicˇ , Bruce Ferwerda, and Markus Schedl . 2016 . Fusing social media cues: personality prediction from twiŠer and instagram . In Proceedings of the 25th international conference companion on world wide web. International World Wide Web Conferences Steering CommiŠee , 107 - 108 .

[26] Kirsten A Smith , MaŠ Dennis, and Judith Mastho‚. 2016 . Personalizing reminders to personality for melanoma self-checking . In Proceedings of the 2016 Conference on User Modeling Adaptation and Personalization. ACM , 85 - 93 .

[27] Marko Tkalcˇicˇ , Bruce Ferwerda, David Hauger, and Markus Schedl . 2015 . Personality correlates for digital concert program notes . In International Conference on User Modeling, Adaptation, and Personalization. Springer, 364 - 369 .

[28] Marko Tkalcic , Matevz Kunaver, Andrej Kosˇir, and Jurij Tasic . 2011 . Addressing the new user problem with a personality based user similarity measure . In Proceedings of the 1st International Workshop on Decision Making and Recommendation Acceptance Issues in Recommender Systems. Citeseer , 106 .

[29] Martijn C Willemsen , Bart P Knijnenburg, Mark P Graus, LC Velter-Bremmers, and Kai Fu . 2011 . Using latent features diversi€cation to reduce choice diculty in recommendation lists . RecSys 11 ( 2011 ), 14 - 20 .