Evolutionary Dynamics of Cultural Memes and Application to Massive Movie Data

Home , Chick flick, The Last Picture Show, The Shawshank Redemption

Seungkyu Shin Graduate School of Culture Technology and BK21 Plus Postgraduate Programme for Content Science, Korea Advanced Institute of Science & Technology, Daejeon, Republic of Korea 34141

Juyong Park Graduate School of Culture Technology and BK21 Plus Postgraduate Programme for Content Science, Korea Advanced Institute of Science & Technology, Daejeon, Republic of Korea 34141 and Sainsbury Laboratory, University of Cambridge, Cambridge, United Kingdom CB2 1LR

The profound impact of Darwin’s theory of evolution on biology has led to the acceptance of the theory in many complex systems that lie well beyond its original domain. Culture is one example that also exhibits key Darwinian evolutionary properties: Diﬀerential adoption of cultural variants (variation and selection), new entities imitating older ones (inheritance), and convergence toward the most suitable state (adaptation). In this work we present a framework for capturing the details of the evolutionary dynamics in cultural systems on the “meme”—the cultural analog of the biological gene—level, and analyze large-scale, comprehensive movie–meme association data to construct a timeline of the history of cinema via the evolution of genres and the rise and fall of prominent sub- genres. We also identify the impactful movies that were harbingers to popular memes that we may say correspond to the proverbial “Eve” of the human race, shining light on the process by which certain genres form and grow. Finally, we measure how the impact of movies correlates with the experts’ and the public’s assessment.

I. INTRODUCTION illustrate his theory of biological evolution [1]. To make a deeper analogy with the current under- Charles Darwin’s theory of evolution has been highly standing of biological evolution, we study cultural evolu- influential in understanding the universal properties of tion occurring on the level of the “meme”—the cultural systems that reproduce and change over generations. In analog of the biological gene—which acts as a unit car- his seminal work The Origin of Species, Darwin postu- rying cultural ideas that can be transmitted from one lated that all organisms share a common ancestor, change mind to another [13–16]. While the analogy between the over time through natural selection, and give rise to new evolution in biology and in culture are clear (as Darwin species that underlie the impressive present-day diversity himself has recognized), specific mechanisms at work can of living things on Earth. On the most fundamental level, be very different [17–19], as in the case of meme and gene. Darwinian evolution stipulates that organisms evolve to The most important difference is that in cultural evolu- adapt to the environment by an iterative process com- tion there can be an arbitrary number of “parents” from prising variation, competition, and inheritance [1–3]. Its whom a newly created work can take after since the ac- profound contribution to the understanding of biologi- tion of inheritance takes place in the mind of the creator cal evolution has prompted an active effort to apply the of new works, unlike the precisely two in sexual repro- theory to many domains other than biology, leading to duction of organisms. This is visualized in more detail the coinage of the expression “Universal Darwinism” [4]. in Fig. 1 (a) and (b). In biological evolution (Fig. 1 (a)), Culture is one example domain where evolution with each gene of an offspring always comes from either of identifiable key Darwinian properties can be observed, the two of its parents (or rare cases of mutations or re- evolving through a differential adoption of cultural vari- combination error). In cultural evolution (Fig. 1 (b)) on ants in a manner analogous to the evolution of biologi- the other hand, a new work’s memes can originate from cal species [5–10]: A new cultural product displays de- an arbitrary number of “parents” and undergo frequent tectable variations from the others, competes with others mutations. in the marketplace to be selected by consumers. Success- While the study of cultural evolution is an active arXiv:1903.02197v1 [physics.soc-ph] 6 Mar 2019 ful variants are selected and thrive over others that may field [6, 17, 20–30], recent developments in data and then perish and disappear. The successful variants fur- machine-learning techniques are providing even newer ther inspire the creators of later products to imitate or in- opportunities for understanding cultural evolution by herit their properties and further adapt [11, 12]. Cultural leveraging the rich feature sets available of cultural prod- evolution is thus the idea that beliefs, knowledge, arti- ucts [31, 32]. In addition, the commodified nature of facts, and human creations that constitute culture un- cultural products in the present day means that the con- dergo a deeply analogous process by which species evolve sumer choice as selection pressure is increasing, speeding through selective retention of favorable variants. In fact, up the rate of evolution that could allows us to more eas- in The Origin of Speies, Darwin himself frequently cited ily observe the evolutionary process in a short time scale. cultural changes (primarily linguistic developments) to Cultural production in today’s high-risk, high-return environment of cultural production would also benefit from 2

(a) Biological Evolution (b) Cultural Evolution

A couple of parenthood Possible single or multiple parenthood

Rare Mutation Frequent Mutation or Recombination error (Creativity)

Transmission between Possible transimission between adjacent generations distant generations

“Special “Love “Prison “Survival” Effects” Story” Escape” “Superhero” “Dreams”

Schindler’s List 0.837 0.354 0.388 . . . . . 0.258 0.041 0.154 (1993)

The Shawshank Redemption 0.697 0.471 0.384 . . . . . 0.986 0.14 0.475 (1994)

Titanic 0.658 0.928 0.957 . . . . . 0.086 0.054 0.199 (1997) .

The Matrix 0.435 0.984 0.462 . . . . . 0.336 0.417 0.527 (1999)

The Dark Knight 0.473 0.686 0.249 . . . . . 0.178 0.96 0.165 (2008)

Inception 0.46 0.852 0.313 . . . . . 0.302 0.219 0.995 (2010)

FIG. 1. The analogy and the differences between evolutions in (a) biology on the gene level, and (b) culture on the meme level. Cultural evolution features an arbitrary number (from single to multiple) of parents, frequent mutations, and transmission between distant generations of memes. (c) Cultural products (movies in our paper) can be represented as a meme vector of the relevance score (association strength) with each meme. We use a meme vector of dimension 900, providing a rich set of descriptors for movies. The shaded components indicate the memes with particularly high relevance score (0.8 or greater). a deeper scientific understanding of the cultural evolu- March 2015. Tags are user-attached metadata that de- tionary process. scribe the movies’ themes or related concepts, typically a single word or a short phrase such as “nostalgic,” “ar- tificial intelligence,” and so forth. The 1, 100 unique tags feature numerical relevance (association) scores between II. DATA AND MOVIE MEMES 0 and 1 to movies, computed via machine-learning algo- rithms on user-contributed reviews and ratings [34]. Tags We analyze the data from MovieLens, a movie rec- thus contain bits of information about the movies, and ommendation service that also provides a stable bench- each movie is composed of different combinations of such mark dataset to researchers [33]. The dataset contains tags with varying relevance scores. Therefore we can con- 20 million ratings and 465, 000 tags on 27, 000 movies sider each tag as a meme constituting the movies. The provided by 138, 000 users between January 1995 and 3

(a) Action (1788) : “Action / Fight Scenes / Fast paced” (b) Adventure (1475) : “Adventure / Special Effects / Franchise” Drama (5568) : “Drama / Relationships / Intimate” Romance (2094) : “Love story / Relationships / Love” Documentary (470) : “Intimate / Narrated / Political” Comedy (4029) : “Comedy / Humor / Goofy”

Animation (465) : “Animal Movie / Computer Animation / Talking Animals” Fantasy (750) : “Fantasy / Special Effects / Magic” Sci-Fi (684) : “Sci Fi / Special Effects / Future” Thriller (1348) : “Suspense / Tense / Murder” Mystery (777) : “Mystery / Murder / Suspense” Horror (1029) : “Horror / Splatter / Supernatural” Crime (1829) : “Crime / Murder / Corruption”

Musical Sci-Fi Fantasy Adventure Romance

Comedy

Animation Action

Drama

Horror Documentary Crime Thriller Mystery

1920 1940 1960 1980 2000 2020 Year

FIG. 2. The correspondence between the network communities of movies and genres. (a) Network of movies based on the meme vector similarity between movies (PCC larger than 0.8). The communities detected in the network (colored) correspond well to established genres. (b) Top three relevant memes within each genre. (c) Phylogeny of movie history showing how the movie genres have grown over time. The movie production years are given in the x-axis. Cross-linkage between genres is shown to increase over time, leading to hybrid genres such as romantic comedy (romance and comedy). parallel between gene and meme are clear from Fig. 1; as production companies), “Oscar”, “best of 2005”, etc. In an individual organism can be viewed as a set of distinct addition, we consolidated tags that are synonyms, e.g. genomes, an individual movie can be viewed as a distinct “ﬁght scenes” and “ﬁghting”, using hierarchical cluster- set of memes. In this work we consider the 10, 380 movies ing methods on their similarity in relevance scores to the that have relevance scores across all the memes. We also movies. This grouping of tags is based on their actual re- eliminated the tags not related to the content of movies, lationship to the movies, rather than semantically. This such as simple facts (names of directors, actors, writers, process is reminiscent of identifying the so-called linked 4

History of Popular Memes by Era

Film Noir Cult Classic Nudity Wartime Tense Goofy All Movies 1902 1942 1972 1976 1988 1992 1945 Ninja Kung fu Buddy Movie Wartime 007 Tense Cult Classic Natural Disaster CGI Action 1924 1955 1962 1969 1985 1991 1995 2006 1988

Family 70mm Cult Classic Ninja Gunfight Island Gunfight Kung fu CGI 3D Adventure 1902 1952 1961 1965 1972 1977 1990 2004 2009 1991 1993

Depression Film Noir Tense Nostalgic Life Philosophy Drama 1915 1944 1972 1982 2004 2007

Gunfight GLBT Western Tear Jerker Chick Flick Sex Based on a Play Runaway Bleak Nostalgic Passionate Gay Relationships Romance 1919 1954 1966 1973 1983 1989 1994 2003 2007 2013 1967 1990 1995 1968 Drinking Based on a Play Ninja Relationships Screwball Comedy Broadway Goofy Gay Crude Humor Comedy 1917 1936 1948 1954 1988 1995 2006 2011 1991 2008

Lions Island Chocolate Tear Jerker Fairy Tale Mummy Allegory Meditative Cult Classic Halloween Big Budget CGI Fantasy 1902 1942 1947 1954 1959 1963 1969 1977 1991 2000 2008 2013 1994

Cult Classic Monster Astronauts Splatter Action CGI Gothic Monster Mad Scientist Low Budget 70mm Cyborgs Big Budget Sci Fi 1927 1941 1953 1961 1966 1972 1981 1986 1992 2000 2007 1973 1989

Espionage Franchise Book High School Cold War Slasher Film Noir Tense Camp Erotic Survival Thriller 1922 1947 1963 1983 1992 2009 2013 1964 1986 1965 1987

Wartime Obsession Murder Lawyer Torture Mystery 1920 1941 1975 1982 1991 2007

Supernatural Supernatural Gothic Gothic Gruesome Big Budget Gore Life & Death Mummy Sci Fi Visceral Franchise Torture Afterlife Horror 1920 1940 1953 1958 1972 1977 1989 1995 1999 2006 2012 1941 2013

Film Noir Cult Classic Buddy Movie Mystery Light Visceral Nostalgic Action Nudity Torture Action Crime 1922 1940 1945 1950 1970 1974 1982 1988 1992 2005 2013 1990

1900 1920 1940 1960 1980 2000 2020 Year

FIG. 3. History of popular memes by era. The top row shows the most popular memes over all the movies, while the rest show those within each genre. 5 m

Upsurge of popularity after the movie M Popularity of Meme

Before the movie M’s release After the movie M’s release

n-3 n-2 n-1 n n+1 n+2 n+3 Time (year of movie M’s release)

FIG. 4. Defining the impact of a movie on the rise in the popularity of a meme. Such a movie features two characteristics, it precedes the meme’s rise (top), and it has be highly relevant to the meme (bottom). genes in cells, pairs or sets of genes located on the same of the detected communities shows that they correspond chromosome inherited together in organisms. Our final well to widely-accepted movie genres such as Thriller, data for analysis consists of 900 unique memes. The sys- Comedy, Romance, and so forth. They are presented tem of movies and tags can be represented as a weighted along with the top-three highest-scoring memes in each bipartite network, mathematically a matrix of the “meme genre in Fig. 2 (b). These memes intuitively support vectors” as follows: the community-genre equivalence, for instance, “Action”, “Fight Scenes”, and “Fast Paced” being the top three M ≡ {mij} (1) memes for the genre that is clearly Action. Fig. 2 (c) where i denotes a movie, j denotes a tag meme and each shows the movies and the networks along the movie pro- element is between 0 and 1. The matrix is also visualized duction years on the x-axis. The two staple genres in in Fig. 1 (c). terms of their size and longevity are Drama and Com- edy, having been produced steadily since the very early years. It also shows that the genres increasingly cross A. Clusters of Movies: Genres over to one another, especially starting from the 1980s. In particular, Comedy shows a high level of interaction with Romance, clearly related to the birth of the so-called We can now define the similarity between movies, for Romantic Comedy. Drama shows an increase in inter- instance via the Pearson Correlation Coefficient of their action with Horror and Action as well. We may label meme vectors. We can then construct a network of Fig. 2 (c) the Phylogenetic tree [36] of movies, although movies by connecting movie pairs whose similarity ex- in culture where cross-fertilization is free to occur and ceeds a certain threshold (0.8 in Fig. 2). This process thus definition of a “species” may be less clear-cut. is analogous to identifying one’s kins via similarity in genetic makeup. The network and the communities of movies are shown in Fig. 2 (a) [35]. A manual inspection 6

B. Popular Memes and Impactful Movies (a) Distribution of relevance score before & after preprocessing

With the movies’ meme scores and production year 0.14 0.030 Density after preprocessing data, we can follow the rise and fall of memes within 0.12 genres, in particular the ones that enjoy a sustained sta- 0.025 0.10 tus and earn a position in history. To lessen the effect 0.020 of the yearly fluctuations in the number of movies pro- 0.08 0.015 duced, we use a three-year time window. Within each 0.06 time window we measure the average meme score within 0.010 each genre, and focus on the popular ones defined as 0.04 having been among the top three high-scoring ones for 0.02 0.005 three or longer consecutive time windows. The results Density before preprocessing 0.00 0.000 are shown for all movies and for each genre are shown in 0.0 0.2 0.4 0.6 0.8 1.0 Fig. 3. (Self-explanatory ones such as “black and white” and “silent” from the early 1900s, and “1950s” have been (b) Distribution of average score change by a year omitted from the figure.) before & after preprocessing 0.30 0.030 Based on the popularity of memes throughout time, we Density after preprocessing can ask more detailed questions regarding many interest- 0.25 0.025 ing aspects of the dynamics. For instance, in evolution the question of when a successful species appeared (the 0.20 0.020 proverbial “Eve” for the humans) is an intriguing and 0.15 enduring one. In our context, it would be asking what 0.015 are the movies that acted as the harbingers of a future 0.10 0.010 prominent meme. In the simplest terms, we employ the following criteria to find it: The movie itself should be 0.05 0.005 highly relevant to the meme, and the movies produced Density before preprocessing 0.00 0.000 after it should have a significantly higher relevance than -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 the ones produced before it. This is illustrated in Fig. 4 which shows movie M itself with a high relevance to (c) Distribution of Impact values meme m that has a significantly higher score after M 0.05 than before it (we again consider a three-year time window). On a technical level, to combine the two factors we 0.04 first need to understand the behavior of the meme score m. Fig. 5 (a) shows the distribution of m (red) being 0.03 right-skewed, calling for regularization to make it better behaved, which in this case can be achieved by taking Density 0.02 the logarithm log(m + ), shown in blue, with = 0 since 0.01 m are all nonzero in our data. On the other hand, the difference between mean meme scores after and before 0.00 (all in a three-year window) a movie is already close to - 10.0 - 5.0 0.0 5.0 10.0 a normal distribution centered around near zero, which Standardized Impact Values ( z-score ) means we can use it as-is, though to make the product of the two terms positive, we transform the latter to take FIG. 5. (a) The distribution of the meme score m before on the value in the range (0, 1). See Fig. 5 (b). We then (red) and after (blue) regularization. (b) The distribution define the impact of a movie of meme score m to be of average meme score before and after any production year. The original distribution is nearly regular (red), necessitating I(m) ≡ log(m) × (mafter − mbefore). (2) no further processing than rescaling and translation to make it positive (blue). (c) Distribution of the impact of the movies, Since we are interested in the movies with the highest the product of the two variables shown in (a) and (b). impact, so we focus on those whose z-scores of impact µ ≡ (I −I)/σI exceed some value which set to be 5 in our work. Fig. 6 shows the history of the meme “Film Noir” that saw particular success between the 1920s and 1940s, sidered a classic in the Film Noir genre in history. Alfred and its most impactful movies. The meme’s average rele- Hitchcock has the largest number of impactful movies in vance score is shown in Fig. 6 (a), exhibiting three major the genre, shown in Fig. 6 (c) along with other classics. peaks with the largest one in the 1940s. Fig. 6 (b) shows In Fig. 7, we show the six most prevalent memes by era the impactful movies with µ > 5, mainly produced in the (see also Fig. 3) with their years of the highest peak av- early 1940s. According to our measure the most impact- erage relevance and the most impactful movies around ful movie is Billy Wilder’s Double Indemnity, indeed con- them. We also find that impactful movies in cinema- 7

History of the Meme “Film Noir” (a) 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 1920 1940 1960 1980 2000 2020

1944 (b) Double Indemnity (9.69) Murder, My Sweet (9.62) 1946 The Woman in the Window (9.61) The Killers (7.57) Laura (9.40) The Strange Love of Martha Ivers (7.57) To Have and Have Not (7.97) The Big Sleep (7.57) Gaslight (6.75) Gilda (7.30) 1920 Lifeboat (6.27) Notorious (6.74) The Cabinet of Dr. Caligari (6.04) Arsenic and Old Lace (5.89) The Spiral Staircase (6.62)

1945 1941 1943 Detour (9.43) The Maltese Falcon (5.47) Shadow of a Doubt (7.52) Scarlet Street (9.41) Ossessione (6.80) Mildred Pierce (8.94) Spellbound (8.42) The Lost Weekend (6.86) Brief Encounter (6.22) And Then There Were None (6.03) The Body Snatcher (5.80) I Know Where I'm Going! (5.28) (c) Director Movie Director Movie

Alfred Hitchcock 1943 Shadow of a Doubt (7.52) Billy Wilder 1944 Double Indemnity (9.69) 1944 Lifeboat (6.27) 1945 The Lost Weekend (6.86) 1945 Mildred Pierce (8.94) 1946 Notorious (6.74) Howard Hawks 1944 To Have and Have Not (7.97) 1946 The Big Sleep (7.57)

Fritz Lang 1944 The Woman in the Window (9.61) 1946 1945 Scarlet Street (9.41) Robert Siodmak The Killers (7.57) 1946 The Spiral Staircase (6.62)

FIG. 6. History of the meme “Film Noir.” (a) The rise and fall of the yearly average relevance score of the meme “Film Noir”. We ﬁnd three major peaks in the early years, with the highest one in the 1940s. (b) The impactful movies (z > 5) in the meme “Film Noir”. The values in parentheses are the z-scores. The most impactful movie turns out to be Double Indemnity from 1944 by Billy Wilder, in fact considered one of the most inﬂuential in cinema studies. (c) Two directors, Alfred Hitchcock and Robert Siodmak, with the most movies in the list of impactful movies. 8

Impactful Movies on Popular memes by Era

Wartime Film Noir 0.3 Tense 0.2 Cult Classic Goofy Nudity 0.1

0.0 Normalized - 0.1 Meme Popularity 1940 1950 1960 1970 1980 1990 2000 Film Noir Cult Classic Nudity Wartime Tense Goofy

1942 1945 1972 1976 1988 1992

1940 1941 1942 The Great Dictator (8.19) Sergeant York (10.19) Mrs. Miniver (9.59) Wartime Foreign Correspondent (7.17) How Green Was My Valley (6.01) To Be or Not to Be (9.57) His Girl Friday (4.82) Suspicion (5.12) Casablanca (9.51) 1944 1945 1946 Double Indemnity (9.69) Detour (9.43) The Killers (7.57) Film Noir Murder, My Sweet (9.62) Scarlet Street (9.41) The Strange Love of Martha Ivers (7.57) The Woman in the Window (9.61) Mildred Pierce (8.94) The Big Sleep (7.57) 1943 1971 1972 Shadow of a Doubt (4.19) Duel (3.65) Aguirre, the Wrath of God (3.63) Tense The Ox-Bow Incident (3.94) Dirty Harry (3.62) Deliverance (3.57) Sahara (3.79) Play Misty for Me (3.61) The Other (3.50) 1970 1971 1972 Beyond the Valley of the Dolls (6.33) The Abominable Dr. Phibes (6.41) The Thing with Two Heads (5.32) Cult Classic The Honeymoon Killers (5.31) Harold and Maude (6.38) Pink Flamingos (5.25) Where's Poppa? (5.29) Vanishing Point (6.34) Tales from the Crypt (5.25) 1971 1972 1973 And Now for Something The Hot Rock (4.83) Live and Let Die (4.06) Completely Different (4.95) Goofy The Chinese Connection (4.79) The Golden Voyage of Sinbad (4.01) The Big Boss (4.81) Horror Express (4.30) Sleeper (3.80) The Million Dollar Duck (4.76) 1971 1972 1980 The Last Picture Show (5.33) The Big Bird Cage (6.13) Melvin and Howard (3.46) Nudity Straw Dogs (5.20) Boxcar Bertha (5.92) Atlantic City (3.37) Summer of '42 (4.48) Last Tango in Paris (5.55) Wholly Moses! (3.18)

FIG. 7. The six most prevalent memes by era with their years of the highest peak average relevance and the most impactful movies. The impactful movies in cinema-specific memes such as “Wartime” and “Film Noir” show higher z-score than those of general adjectives such as “Tense” and “Goofy.” The top one percent impactful movies of “Wartime” and “Film Noir” have higher z-scores than 3.27 and 3.09, while those of “Tense” and “Goofy” have higher z-scores than 2.57 and 2.54. specific memes such as “Wartime” and “Film Noir” tend C. Impact and Public Evaluation to have in general higher z-score than those made of general adjectives such as “Tense” and “Goofy.” The top one The impact defined in our paper is based on the preva- percent impactful movies of “Wartime” and “Film Noir” lence of memes, indirectly constructed using computa- have higher z-scores than 3.27 and 3.09, while those of tional means on various types of user input. We then “Tense” and “Goofy” have higher z-scores than 2.57 and ask whether the impactful movies have also been re- 2.54. ceived well by the audience, which could be a measure of fitness in the Darwinian sense: If a movie is not well received, the chance of passing its memes onto future movies would certainly diminish. We use the Internet Movie Database(IMDb) and Rotten Tomatoes data for this purpose. IMDb provides two types of data, IMDb 9

ately high vote counts. Fig. 8 shows the relationship 7 Density 10 0.7 between these variables. IMDb Vote and IMDb Rating 106 are only modestly correlated with Pearson correlation co- 0.6 efficient of 0.189 ± 9.085 × 10−7, whereas IMDb Rating 5 10 0.5 and Rotten Tomatoes are more strongly correlated with 4 0.782 ± 4.24 × 10−7. More telling would be whether im- 10 0.4 3 pactful movies differ from others with respect to these 10 0.3 indicators. To see if this is the case, we divide the movies 2 10 0.2 into two groups: Highly impactful (µ > 5) with regards log(IMDb Vote) 1 to at least one meme (total 2, 567 out of 10, 380 movies) 10 0.1 and the others (7, 813 movies). We then find that the 100 0.0 impactful movies outscore the others consistently among 2.0 4.0 6.0 8.0 10.0 ratings, averaging 6.92 (IMDb Rating) and 7.07 (Rotten Tomatoes), respectively, versus 6.56 and 5.96 (p = .000). IMDb Rating The IMDb Vote, however, turns out to be nondiscrimi- Density natory for the two groups (4.21 and 4.18, p = .054). This 10.0 tells us that the sheer interest in a movie is likely unre- 0.6 lated to the eventual ratings given by the viewership, and 8.0 0.5 has more to do with hype or marketing.

6.0 0.4

0.3 III. CONCLUSION 4.0 0.2 Here we have proposed studying the evolutionary dy- 2.0 0.1 namics of cultural systems at the meme level. The meme,

Rotten Tomates Rating the biological analog of the gene, is a human-generated 0.0 0.1 tag that had variable relevance to different cultural prod- 0.0 2.0 4.0 6.0 8.0 10.0 ucts. We were able to construct the network of movies IMDb Rating based on meme-level similarity, discovering the community structure that corresponded well with common genre designations. We also found that cross-genre connections FIG. 8. Impact and the evaluations by the expert and the increased over time, indicating rising complexity in meme public. (a) The relationship between IMDb Rating and IMDb Vote. The Pearson coefficient is 0.189[errors?], showing weak compositions. We observed clear evidence of the memes correlation. The blue lines represent average values for im- rising and falling in popularity, and identified the impact- pactful movies and the red line for non-impactful movies. ful movies that herald the success of a meme. Further- IMDb Vote shows almost no correlation. (b) The relation- more, we examined the viewership’s reception of those ship between IMDb rating and Rotten Tomatoes rating. The movies and found positive correlations, meaning that for Pearson coefficient is 0.782, showing higher correlation. The a movie to impact the future and give rise to a new suc- impactful movie group shows higher average rating than non- cessful meme it must appeal to the consumer base as well, impactful movie group for both IMDb and Rotten Tomatoes not merely be different from its predecessors. rating. The general nature of our framework and findings sug- gest it can be applied in many diverse cultural areas where similar data are available, which is becoming increasingly the case thanks to massive consumer partici- Vote (the number of votes cast) and IMDb Rating (the pation and advances in data collection. In an era where average scores from the votes). Rotten Tomatoes, on cultural products are produced systematically and pre- the other hand, is an aggregation service of reviews by sented to an ever-widening pool of consumers, the works critics. Thus IMDb represents the general public’s re- of this kind to understand how culture evolves due to ception of movies, whereas Rotten Tomatoes presents market pressure has the potential to lead to valuable in- the experts’. We use the log of the IMDb Vote since sights on how culture, technology, and the society inter- it is unbounded, and a few movies feature disproportion- act.

[1] Charles Darwin, On the origin of species by means of [2] Ronald Aylmer Fisher, The genetical theory of natural natural selection, or preservation of favoured races in the selection (Clarendon Press, Oxford, 1930). struggle for life (John Murray, London, 1859). [3] Robert M Schwartz and Margaret O Dayhoﬀ, “Origins 10

of prokaryotes, eukaryotes, mitochondria, and chloro- [20] B. S. Hewlett and L. L. Cavalli-Sforza, “Cultural trans- plasts,” Science 199, 395–403 (1978). mission among aka pygmies,” Am. Anthro. 88, 922–934 [4] R. Dawkins, The Selfish Gene (Oxford university press, (1986). Oxford, 1976). [21] R. Aunger, “The life history of culture learning in a face- [5] Ruth Benedict, Patterns of culture (Houghton Mifflin to-face society,” Ethos 28, 445–481 (2000). Harcourt, Boston, 2005). [22] A. Mesoudi and M. J. O’Brien, “The cultural transmis- [6] L. L. Cavalli-Sforza and M. W. Feldman, Cultural trans- sion of great basin projectile-point technology i: an ex- mission and evolution: a quantitative approach, 16 perimental simulation,” Am. Antiq. 73, 3–28 (2008). (Princeton University Press, Princeton, 1981). [23] R. McElreath, M. Lubell, P. J. Richerson, T. M. War- [7] Luca L Cavalli-Sforza, Marcus W Feldman, Kuang-Ho ing, W. Baum, E. Edsten, C Efferson, and B. Paciotti, Chen, and Sanford M Dornbusch, “Theory and observa- “Applying evolutionary models to the laboratory study tion in cultural transmission,” Science 218, 19–27 (1982). of social learning,” Evol. and Hum. Behav. 26, 483–508 [8] Jared M Diamond, Guns, germs and steel (W. W. Nor- (2005). ton, New York, 1997). [24] R. M. MacCallum, M. Mauch, A. Burt, and A. M. Leroi, [9] W. H. Durham, “Cultural evolution: Overview,” Int. E. “Evolution of music by public choice,” Proc. Nat. Acad. Soc. Behav. Sci. 5, 3049–3052 (2001). Sci. 109, 12081–12086 (2012). [10] A. Mesoudi, A. Whiten, and K. N. Laland, “Perspec- [25] M. Pagel, “Human language as a culturally transmitted tive: Is human cultural evolution darwinian? evidence replicator,” Nat. Rev. Genet. 10, 405 (2009). reviewed from the perspective of the origin of species,” [26] R. D. Gray and Q. D. Atkinson, “Language-tree di- Evolution 58, 1–11 (2004). vergence times support the anatolian theory of indo- [11] Carmela Rosalba Guglielmino, Carla Viganotti, Barry european origin,” Nature 426, 435 (2003). Hewlett, and Luigi Luca Cavalli-Sforza, “Cultural vari- [27] George Basalla, The evolution of technology (Cambridge ation in africa: Role of mechanisms of transmission University Press, Cambridge, 1988). and adaptation,” Proc. Natl. Acad. Sci. 92, 7585–7589 [28] Robert A Hinde and Les A Barden, “The evolution of (1995). the teddy bear,” Anim.l Behav. 33, 1371–1373 (1985). [12] Christopher Robert Hallpike, The principles of social [29] Henry C Plotkin and Francis J Odling-Smee, “A evolution (Oxford University Press, Oxford, 1986). multiple-level model of evolution and its implications for [13] R. C. Lewontin, “The units of selection,” Annual Rev. sociobiology,” Behav. Brain. Sci. 4, 225–235 (1981). Ecol. Syst. 1, 1–18 (1970). [30] Everett M Rogers, Diffusion of innovations (The Free [14] Stephen Shennan, Genes, memes, and human history: Press, New York, 1962). Darwinian archaeology and cultural evolution (Thames [31] Robert Aunger, The electric meme: A new theory of how & Hudson, London, 2002). we think (Simon and Schuster, New York, 2002). [15] David Hull, “Taking memetics seriously: Memetics will [32] S. Shin and J. Park, “On-chart success dynamics of pop- be what we make it,” Darwinizing culture , 43–67 (2000). ular songs,” Adv. Complex. Syst. , 1850008 (2017). [16] Adam Kuper, “If memes are the answer, what is the ques- [33] “Movielens dataset in grouplens website,” https:// tion,” Darwinizing culture , 175–188 (2000). grouplens.org/datasets/movielens/, accessed: 2019- [17] Robert Boyd and Peter J Richerson, Culture and the evo- 03-01. lutionary process (University of Chicago press, Chicago, [34] J. Vig, S. Sen, and J. Riedl, “The tag genome: Encod- 1988). ing community knowledge to support novel interaction,” [18] Robert Boyd and Peter J Richerson, “Why culture is ACM. T. Intel. Syst. 2, 13 (2012). common, but cultural evolution is rare,” in Proc. Brit. [35] R. Lambiotte, J. C. Delvenne, and M. Barahona, “Lapla- Acad., Vol. 88 (Oxford University Press Inc., 1996) pp. cian dynamics and multiscale modular structure in net- 77–94. works,” arXiv preprint arXiv:0812.1770 (2008). [19] Julian Haynes Steward, Theory of culture change: The [36] W Ford Doolittle, “Phylogenetic classification and the methodology of multilinear evolution (University of Illi- universal tree,” Science 284, 2124–2128 (1999). nois Press, Illinois, 1972).