=Paper=
{{Paper
|id=Vol-2989/long_paper10
|storemode=property
|title=The Sentiment of Crypto Art
|pdfUrl=https://ceur-ws.org/Vol-2989/long_paper10.pdf
|volume=Vol-2989
|authors=Massimo Franceschet
|dblpUrl=https://dblp.org/rec/conf/chr/Franceschet21
}}
==The Sentiment of Crypto Art==
https://ceur-ws.org/Vol-2989/long_paper10.pdf
The Sentiment of Crypto Art
Massimo Franceschet
Department of Mathematics, Computer Science, and Physics – University of Udine – Italy
Abstract
Crypto art is a beautiful example of a side effect of a technology: blockchain. While commonly
associated with decentralized finance and cryptocurrencies, blockchain technology allowed to create
scarcity in a world – digital art – where everything can be infinitely duplicated and freely saved with
a right-click of the mouse, forstering the growth, burst, and stabilization of a dizzying market made
of artists, collectors, art galleries and curators. In this preliminary work we assess the sentiments
expressed by crypto artists when they create art and those coveted by crypto collectors when they
collect art. We find that artists communicate positive emotions like joy and trust instead of negative
ones like fear and sadness. However, collectors are agnostic to these emotions. This might be useful
information to integrate into a crypto art discovery system that we are currently building.
Keywords
blockchain, non-fungible tokens, crypto art, text mining, sentiment analysis
1. Introduction
Blockchain technology, while commonly associated with cryptocurrencies, has shown potential
to bring radical structural change to the arts and creative industries [11]. Blockchains are hard
to grasp at first. The basic scientific research from which the technology emerged – a journal
paper by Stuart Haber, a cryptographer, and Scott Stornetta, a physicist [6] – is distinct from
the financial systems it later generated – the advent of Bitcoin and other cryptocurrencies [9].
Haber and Stornetta were trying to deal with epistemological problems of how we trust what
we believe to be true in a digital age. In particular, they started from two questions [11]:
1. If it is so easy to manipulate a digital file on a personal computer, how will we know
what was true about the past?
2. How can we trust what we know of the past without having to trust a central authority to
keep the record?
These questions were solved by Haber and Stornetta [6, 7], with a contribution by Satoshi
Nakamoto 17 years later [9], using a combination of tools borrowed from mathematics, com-
puter science, economics and political science.1 Formally, a blockchain is a distributed ledger,
using cryptography to secure an evolving consensus about a token with economic value [13].
In their original, far-sighted proposal, Haber and Stornetta envisaged the adoption of blockchains
beyond texts, and maybe in the context of art as well [6]:
CHR 2021: Computational Humanities Research Conference, November 17–19, 2021, Amsterdam, The
Netherlands
£ massimo.franceschet@uniud.it (M. Franceschet)
DZ 0000-0001-9490-1464 (M. Franceschet)
© 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Wor
Pr
ks
hop
oceedi CEUR Workshop Proceedings (CEUR-WS.org)
ngs
ht
I
tp:
//
ceur
-
SSN1613-
ws
.or
0073
g
1
Satoshi Nakamoto is the pseudonymous used by the person or group of people that developed Bitcoin,
authored the Bitcoin white paper, and created and deployed Bitcoin’s original reference implementation.
310
Of course digital time-stamping is not limited to text. Any string of bits can
be time-stamped, including digital audio recordings, photographs, and full-motion
videos. […] time-stamping can help to distinguish an original photograph from a
retouched one.
Indeed, the blockchain has core use cases in the arts including provenance, fractional own-
ership and digital scarcity. A notable example of the first use case – provenance – is the sale
in 2018 of the Barney A. Ebsworth collection at Christie’s for 318M$. The auction was held
in partnership with the technology provider Artory using a blockchain solution to record in-
formation about the auction and all future sales of the auctioned artworks. As for fractional
ownership, in 2018 the company Maecenas bought Andy Warhol’s 14 Small Electric Chairs
and divided it up into shares sold as so-called ART tokens. The company raised 1.7M$ for
31.5% of the artwork at a valuation of 5.6M$.
Crypto art is related to the third use case of blockchain in art: digital scarcity. The novel
idea is to make a digital file scarce by associating it with a non-fungible token or NFT [2, 5].
NFTs are cryptographic tokens stored on a blockchain that represent something unique, for
instance, a one-of-a-kind collectible, a weapon in a blockchain game, or a portion of digital
land. NFTs are not interchangeable and cannot be divided, as opposed to fungible tokens
(cryptocurrencies like Bitcoin or Ether), which are interchangeable and can be split in smaller
pieces whose sum is equivalent to the whole. In crypto art, an NFT certifies the scarcity
(number of copies), ownership (current owner) and provenance (historical owners and creator)
of a digital artwork. Transferring the NFT is akin to transferring the certificate of ownership
of the artwork. However, like in traditional art, ownership rights generally do not include
intellectual property rights such as copyright claims and rights for any commercial re-use.
Crypto art draws its origins from conceptual art, sharing the immaterial and distributive
nature of artworks, and the rejection of conventional art markets and institutions [3]. A
niche artistic movement until early 2020, crypto art market went parabolic in late 2020 - also
because of the COVID pandemic and the consequent digitalization of our lives - attracting
the attention of major mass media and major auction houses. Recent notable sales of crypto
artworks include:
1. Everydays: The First 5000 Days, by digital artist Beeple, was the first NFT sold at
Christie’s on March 2021 for the record-breaking amount of 69M$ through the crypto
art gallery MakersPlace;
2. The Fungible, by digital artist Pak, is an NFT collection sold in April 2021 at Sotheby’s
in collaboration with crypto art gallery Nifty Gateway for almost 17M$;
3. Nine CryptoPunks from Larva Labs’ own collection sold in May 2021 at Christie’s for
16.9M$. It is, unsurprisingly, the first time an NFT has been offered alongside work by
Andy Warhol and Jean-Michel Basquiat.
As a significant by-product, crypto art is generating increasing amounts of openly available
structured and unstructured data, and this is probably the main feature that sets it apart
from traditional art. Indeed, all trades in crypto art are immutably recorded on a public
blockchain, and this data is potentially available for analysis. Moreover, artworks metadata
like title, description, tags as well as the digital files representing the artworks themselves are
stored on peer-to-peer networks like IPFS and are available to download. On the contrary, in
traditional art this information is typically secreted or available only for a (significant) fee.
311
Mind that, despite the apparent availability of blockchain data, we experienced that col-
lecting blockchain art data is not a trivial task. First of all, most galleries do not offer a
reliable API, hence we had to rely on general purpose solutions like the API of the Ethereum
blockchain explorer etherscan.io. When data are downloaded, they need to be filtered by smart
contracts of the gallery. Typically, a gallery uses different contracts for different purposes and
this information is generally difficult to find and is not well documented. Moreover, changes
like the update of a contract or the deployment of a new contract needs to be taken into
account. Finally, there exists no standard for operability of crypto art galleries. Each gallery
implements its own business logic written in-home using the smart contract language of the
blockchain where it operates, like Solidity on Ethereum. Therefore, each gallery needs to by
treated as a particular case, and this clearly hinders the scalability of the data fetching engine.
Besides open data, another facet of crypto art that distinguishes it from its traditional
counterpart is velocity. In crypto art events happen at every instant: an artist mints a new piece
or accepts a bid made from a collector, a collector bids or directly buys an artwork, two users
flip artworks, and more. From a data science viewpoint, the crypto art market corresponds to
an open, real-time stream of data, more akin to financial trading than traditional art.
In this paper, we focus on the unstructured, textual metadata that accompanies each art-
work. This includes a title, a description and a list of tags for each tokenized artwork. In
particular, we use sentiment analysis [1] on the textual metadata to mine the sentiment of
each artwork. The goal is twofold:
1. discover what sentiment poles (positive or negative) and emotions (anger, fear, anticipa-
tion, trust, surprise, sadness, joy, or disgust) are expressed by crypto artists when they
create digital art;
2. find out if crypto collectors are influenced by these emotions when buying digital art.
It is worth saying that this work is part of a larger project that aims to devise recommen-
dation systems for crypto art. The goal is to help users find artworks that they might have
overlooked otherwise. Users can be artists, interested in discovering similar artists for the
sake of collaboration, or collectors and art investors willing to find new art to purchase that is
somewhat overlapping with that already present in their collection. The sentiment expressed
by an artist or looked for by a collector is precious information in view of this recommendation
system and this sentiment information can be integrated with different signals, for instance
with the aesthetic features extracted from the images that represent the artworks.
2. Sentiment analysis
To answer the above-mentioned questions, we analysed the sentiment of all artworks of Super-
Rare, a peer-to-peer marketplace for non-fungible tokens built on the Ethereum blockchain.
More plainly, SuperRare is a marketplace to collect and trade unique, single-edition digital
artworks. Each artwork is authentically created by an artist in the network, and tokenized as
a crypto-collectible digital item that you can own and trade. SuperRare is one of the earliest
crypto art galleries (it started in April, 2018) and is among the most important crypto art
marketplaces, by popularity and volume of exchanged artworks. As of today (14 June, 2021),
these are some figures for the gallery:
• number of tokenized artworks: 25,375;
312
anticipation
negative
sadness
surprise
positive
disgust
anger
trust
fear
joy
1
disgust
0.8
sadness
0.6
anger
0.4
fear
0.2
negative
0
surprise
−0.2
anticipation
−0.4
joy
−0.6
positive
−0.8
trust
−1
Figure 1: The (Pearson) correlation plot among emotions. The size, color and transparency of the bubble
indicate the correlation. Notice the top-left positive block and the bottom-right negative one.
• number of sold artworks 15,829 (62%);
• sale volume: 46,595.78 ETH or 58,723,268 USD (change rate at transaction time);
• number of active users: 3579;
• number of users that sold at least one artwork: 1679 (47%);
• number of users that bought at least one artwork: 2757 (77%).
The SuperRare dataset was acquired from the gallery’s API and is set available on Kaggle
[4]. For the analysis we used the tidyverse R packages for data science and the tidytext R
package for text mining [12, 10].
We applied lexicon-based sentiment analysis on the text found in the artworks’ metadata
(title, description and tags). It is worth mentioning that on SuperRare gallery this metadata
is written by the artists that mint the artwork and not, for instance, by the art curators
working for the gallery, if any. We used the NRC Emotion Lexicon by Saif Mohammad and
Peter Turney [8]. The NRC lexicon is a list of English words and their associations with eight
basic emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) and two
sentiment poles (negative and positive).
As a start, we correlated the emotion ratings for each artwork. With little surprise, we find
out – see Figure 1 – that the emotion variables cluster in two main groups:
• a positive group containing joy, trust, surprise and anticipation;
313
• a negative group containing sadness, disgust, angry and fear.
Next, for each artwork and emotion, we computed the emotional score of the artwork as the
relative number of words used in the artwork metadata that match the emotion. For instance,
if an artwork has 5 words expressing joy, 3 words expressing sadness, and 2 words expressing
trust, then the emotional scores for joy, sadness and trust are 5 / (5 + 3 + 2) = 0.5, 3 / (5 + 3
+ 2) = 0.3 and 2 / (5 + 3 + 2) = 0.2, respectively. For each emotion, we sorted the artworks
by the emotional score. Here are the artworks leading the emotion rankings along with a short
description:
• The most positive artwork is C.C. Crypto Capital by skygolpe and alecttn (emotional
score 0.39, token id.2 10345) The authors explore the concept of transition, depicting
the idea of renovation through a vibrant, sharp composition.
• The most negative artwork is Kozachok’s Inferno: 9th Circle of Hell: Treachery by
Kozachok (emotional score 0.27, token id 9472). Kozachok’s Inferno is a customized
representation of Dante Alighieri’s Inferno. The negativity of the artwork is immediately
perceived from the start of the description: In this circle, we can notice people being
caught in this frozen lake, from their waist up to their neck, unable to escape. All senses
are present, including the constant pain of the ice. We see a huge creature, a monstrous
and terrifying hybrid entity, playing golf with the heads of the people trapped in the frozen
lake.
• The trust ranking is lead by Gavin Wood, The King of Cross-Chain by undeadlu (emo-
tional score 0.28, token id 19259). The commissioned artwork is dedicated to Gawin
Wood, one of the most privileged minds of the crypto world, involved in several far-
sighted projects like Ethereum, Polkadot and Kusama.
• Emotions fear and anger are captured by Inferno III: Lake of Fire by deathimself (emo-
tional scores 0.28 and 0.17, token id 15388). This is a scary and irated excerpt from
the author’s description: The heat is intense, your flesh feels as if it’s been melted off
entirely, and the smell of sulfur and burning meat fills the air.
• Sadness dominates in The Struggle by mayaguy (emotional score 0.20, token id 24663).
Again, the emotion shines through the words used by the author: Everyone has a mental
struggle of some sort at some point in their lives. It can be difficult to pick yourself
up, the feeling of a great weight burdening your very soul. We try to fight, pushing back
against these feelings, pushing forward, lifting ourselves up day after day. For many the
burden is too much and they collapse under the pressure, for some the pain becomes less
and they can once again stand tall.
• Joy and surprise both emerge in Spring Bloom by kristyglas (emotional score 0.19 and
0.12, token id 19484). This in an adventure series about discovery, persistence and
personal growth, set in a fantasy world that takes life of its own. Some of the tags used
in the artwork actually express these feelings: adventure, color, fantasy, flowers, magic,
nature, plants, spring.
2
To get the URL of the artwork in SuperRare gallery just concatenate the URL prefix https://superrare.
co/artwork-v2/ with the token id of the artwork. For instance, in this case the URL is https://superrare.co/
artwork-v2/10345.
314
Sentiment poles of SuperRare artworks
0.7
0.6
Relative frequency
sentiment
0.5 negative
positive
0.4
0.3
2019 2020 2021
Date
Sources: SuperRare and NRC Emotion Lexicon
Figure 2: The evolution of sentiment poles (positive and negative) of crypto art over time. Positive
sentiments neatly dominate negative ones all over the history of the gallery.
• Anticipation is well represented in Camp Log #256: Abbot by kreaturekastle. (emotional
score 0.19, token id 19920). This is a crypto comic featuring different outdoor-centric
characters. The hopeful motivation for the artwork is quite telling: After over a year
spent in lock-down in our cramped 1st floor Brooklyn apartment near noisy neighbors,
and little exposure to the outside world (we have to be super careful due to folks in our
household being immuno-compromised), suffice it to say I needed some way to escape
aside from video games and our short little walks around the neighborhood.
• Kozachok’s Inferno: 3rd Circle of Hell: Gluttony deservedly leads the disgust ranking
(emotional score 0.14, token id 8795). A few sentences from the description are more
than enough to convey the feeling: Each person is trapped and tied in unbreakable chains
in their own boat, they are not feeling any hunger or thirst, but they are uncontrollably
voiding their bowels every other day inside their boat. Left exposed in the sun and
surrounded by their own feces and vomit in that same water, insects would descend.
Stinging and biting insects like wasps and other mosquitoes would torture the victim,
but worse, others would crawl inside the cracks made by rodents and other unprotected
orifices of the subject and lay eggs, eating them alive from the inside out.
We are now interested to explore the following queries: what are the emotions and senti-
ment poles that are most expressed by crypto artists? What is the temporal evolution of the
emotional spectrum? To this aim, we divided the history of SuperRare gallery in slices of 30
days and analysed the aggregated sentiment of artworks in each time interval. The outcomes
are depicted in Figures 2 and 3. We observe that:
315
Emotions of SuperRare artworks
0.25
0.20
emotion
anger
Relative frequency
anticipation
0.15 disgust
fear
joy
sadness
0.10
surprise
trust
0.05
2019 2020 2021
Date
Sources: SuperRare and NRC Emotion Lexicon
Figure 3: The evolution of emotions of crypto art over time. After an initial start of emotional confusion,
crypto art assumed a positive, joyful and hopeful personality.
1. after an initial period of emotional instability, characterized by chaotic oscillations of
sentiments and emotions, the emotional spectrum chills down, getting more regular and
even;
2. positive sentiments dominate over negative ones;
3. emotions joy, trust and anticipation lead the emotional spectrum. They are followed, at
a distance, by fear and sadness. Disgust is the least represented emotion.
It is worth noticing that these results are not an artifact of the NRC lexicon. Indeed, in
this lexicon there are more negative than positive words (3324 against 2312). Moreover, the
frequency of words associated to emotions are as follows: fear (1476), anger (1247), trust (1231),
sadness (1191), disgust (1058), anticipation (839), joy (689), surprise (534). In particular, joy
is among the least frequent emotions in the lexicon but it is the most represented emotion in
crypto artworks. We conclude that crypto artists, at least for what concerns the dataset at
hand, create with a clear positive intent and aim to mainly express positive emotions like joy,
trust and anticipation. Fear is the most represented negative feeling, while disgust is the least
expressed one.
We now want to find out if the same holds for collectors: do they collect artworks that
express clear emotions? To find an answer we correlated the emotional scores of artworks with
several market variables indicating artwork success, including: number of bids, bid volume,
number of sales, and sale volume of both primary and secondary art market. We noticed
no correlation, either positive or negative, among sentiment and market success of artworks.
We conclude that while the average artist favors some emotions among others, the average
316
collector is agnostic to the emotion expressed by the art they collect. However, particular top
collectors might have prefer some specific sentiments when they buy. For instance, we noticed
that whaleshark, a top collector on SuperRare, prefers positive works more than the average
collector, while momus, another notable collector, favors negative art.
3. Limitations
The present work is a preliminary study and has some known limitations that we want to
acknowledge in the following:
1. we are assuming that the feelings of the artist are expressed by their work and in partic-
ular that they are expressed by the texts that describe it. It is worth noticing that on
SuperRare gallery the text accompanying the work is typically written directly by the
artist when they mint the work and not by curators of the gallery. As a crypto artist, I
can witness that, at least in my limited experience, these hypothesis are valid. However,
a more careful consideration of these assumptions is needed for an extended version of
this work;
2. we did not consider the visual qualities of the image (or video) representing the artwork
and we assume that there is no major discrepancy between emotions expressed in the
image and in the text of the artwork. We do not see good reasons for describing in a
positive way an artwork that expresses negative emotions, or vice versa. Of course, there
might be exceptions but, given the size of the dataset, we consider them not statistically
significant. Furthermore, analyzing the sentiment of an image is not a trivial task and
we postpone this further analysis for future work;
3. we observed the dominance of positive sentiments expressed by crypto artists and noticed
that this is not an artifact of the lexicon used. However, we did not check this prevalence
in other art-related corpora. Is this positivity peculiar of crypto art? Again, being an
early and active member of this space, I can only witness the optimism and enthusiasm of
the majority of individuals working in this thrilling space, hence the result is in some sense
not surprising to me. However, a comparison with texts associated with contemporary
traditional art is a good hint for the future.
4. Conclusion
We have mined the sentiment of more than 25,000 digital artworks tokenized on the SuperRare
marketplace using textual information contained in artwork’s metadata. Our main conclusions
are summarized as follows:
1. artists express clear emotions though their art: they are more positive than negative,
they mainly convey feelings of joy, trust and anticipation instead of negative emotions
like sadness and disgust;
2. on the other hand, the average collector is not much influenced by the emotion expressed
by the purchased art;
3. however, we noticed that single whale collectors prefer certain sentiments when they buy
art;
317
4. finally, emotions expressed by artists were confused in the early period of crypto art but
they soon settled down: crypto art assumed a positive and joyful personality, full of trust
and anticipation.
References
[1] C. C. Aggarwal and C. X. Zhai. Mining Text Data. Springer Publishing Company, In-
corporated, 2012.
[2] J. Bailey. What Is CryptoArt? https://www.artnome.com/news/2018/1/14/what-is-
cryptoart. 2018.
[3] B. P. Finucane. “Creating with blockchain technology: the “provably rare” possibilities
of crypto art”. MA thesis. The University of British Columbia, 2018.
[4] M. Franceschet and G. Colavizza. Artworks, bids and sales from the Superrare crypto art
gallery. https://kaggle.com/franceschet/superrare. 2021.
[5] M. Franceschet, G. Colavizza, T. Smith, B. Finucane, M. L. Ostachowski, S. Scalet, J.
Perkins, J. Morgan, and S. Hernández. “Crypto art: A decentralized view”. In: Leonardo
54.4 (2021), pp. 402–405.
[6] S. Haber and W. S. Stornetta. “How to time-stamp a digital document”. In: Journal of
Cryptology 3.2 (1991), pp. 99–111.
[7] S. Haber and W. S. Stornetta. Method for secure time-stamping of digital documents.
https://patents.google.com/patent/US5136647A/en. 1992.
[8] S. M. Mohammad. “Sentiment analysis: automatically detecting valence, emotions, and
other affectual states from text”. In: arXiv:2005.11882 [cs] (2021). url: http://arxiv.
org/abs/2005.11882.
[9] S. Nakamoto. Bitcoin: A peer-to-peer electronic cash system. https://bitcoin.org/bitcoin.
pdf. 2008.
[10] J. Silge and D. Robinson. Text Mining with R: A Tidy Approach. 1st. O’Reilly Media,
Inc., 2017.
[11] A. Whitaker. “Art and blockchain”. In: Artivate: A Journal of Entrepreneurship in the
Arts 8.2 (2019), pp. 21–46.
[12] H. Wickham and G. Grolemund. R for data science: import, tidy, transform, visualize,
and model data. O’Reilly, 2016.
[13] D. Yaga, P. Mell, N. Roby, and K. Scarfone. Blockchain technology overview. Tech. rep.
National Institute of Standards and Technology, 2018.
318