Recommendation System with Content-Based
Filtering in NFT Marketplace
Edi Surya Negara
1,
*, Sulaiman
1
, Ria Andryani
1
, Prihambodo Hendro Saksono
1
, and Yeni Widyanti
2
1
Data Science Interdisciplinary Research Center, Computer Science Faculty, Universitas Bina Darma, Palembang,
[email protected]d (P.H.S.)
2
Economics and Business Faculty, Universitas Bina Darma, Palembang, Indonesia;
Email: yeniwidyanti@binadarma.ac.id (Y.W.)
*Correspondence: e.s.negara@binadarma.ac.id (E.S.N.)
Abstract—Non-Fungible Token (NFT) is a digital asset that
cannot be exchanged or used, and uses Crypto currency
values according to the type of digital money used, for
example Bitcoin, Ethereum. The NFT Marketplace is a
platform for buying and selling NFT like Tokopedia. This
common problem is often encountered in e-commerce,
especially in the NFT Marketplace, among other buyers often
having difficulty finding products. This makes it difficult for
the NFT Marketplace and sellers to promote products that
match the preferences of potential buyers. A
recommendation system that is very much needed in
overcoming these problems, responding to these problems
the author tries to make a recommendation system using the
Content Based Filtering approach using the cosine similarity.
The results of this study indicate that the Machine Learning
model can provide Top-N recommendations from the
product being sought.
Keywords—crypto currency, non-fungible token, content-
based filtering, Non-Fungible Token (NFT) marketplace
I. INTRODUCTION
The times have made new information and technology
seem unlimited and the flow is so fast like the development
of various blockchain technologies such as
cryptocurrencies and smart contracts [1]. This fairly rapid
development process is also supported by the development
of various algorithms, data processing techniques, and
various highly sophisticated computing technologies [2, 3].
Developments in Cryptocurrencies today are promising,
citing the publication of Commodity Futures Trading
Regulatory Agency (BAPPEBTI), it is stated that the
Crypto tax in Indonesia is planned to be at a rate of 0.05%
and this tax rate is lower than stocks which are subject to
0.1%, this makes business activists in the field of
competition. Digital competition creates a market
specifically for NFT [4]. The increase in NFT transactions
in Indonesia is an opportunity to boost state revenues
which will enforce tax regulations [5].
NFT is a digital asset such as music, in-game items,
paintings, and videos stored in smart contracts. NFT is
unique which means that there are no duplicates in each
NFT ownership. Since the pandemic in Indonesia, the
development of NFT has increased quite rapidly, this can
be seen from the increasing number of local marketplaces
that release transaction services, for example, Tokomal.
This happens because there are many local Indonesian
creators who are competing to make extraordinary works
of art. The beginning of NFT was scrutinized by the
broader community after someone named Ghozali
managed to sell his digital works at a fantastic price,
namely at a price of USD 1 million.
NFT is a representation of digital assets that cannot be
exchanged or exchanged on a par with other NFTs or of
the same type, and the concept of NFT is digital
authenticity that cannot be replicated [6]. Marketing and
business strategies to compete in the NFT The marketplace
includes a recommendation system. Recommendation
systems have been present and are widely used by almost
all business fields where the public or consumers need
information as advice in decision-making [7, 8]. Use a
recommendation system will have a profit, because the
more precise the system in recommending NFT can make
prospective buyers more comfortable in choosing and
increase the possibility to buy more diverse products [9].
Piyadigama et al. also conducted research showing that the
use of a recommendation system for NFT products can
increase the number of sales. So that there is an increase
and progress that encourages the development of the NFT
recommendation system [10]. The recommendation
system is useful for filtering abundant data into important
and useful information for the company.
This research discusses how to create a recommendation
system model with a content-based filtering approach,
content-based filtering approach works by suggesting
similar items based on the user's past activity or being
viewed in the present to the user. The more information the
user provides, the better the recommendation system's
accuracy.
Manuscript received August 22, 2022; revised September 29, 2022;
accepted October 24, 2022; published June 7, 2023.
518
Journal of Advances in Information Technology, Vol. 14, No. 3, 2023
doi: 10.12720/jait.14.3.518-522
II. LITERATURE REVIEW
A. Non-Fungible Token (NFT)
NFT is a practice that makes value in a digital artwork
the only or pure one, the value in the artwork cannot be
duplicated so the value will be very unique [11]. NFT itself
is a digital file in the form of a token and non-exchangeable
which mostly uses the Ethereum blockchain technology to
identify ownership of digital assets which can be music,
videos, pictures, collections, or other digital files such as
equipment or in-game characters [12].
NFT can be valued at very expensive because it has its
unique and historical value, apart from that, because the
artistic taste of the owner. Ownership of an NFT is
evidenced by an immutable and cryptographically secured
record on the blockchain that is meant to be taken as proof
by others in the cryptosp here that someone is the owner of
the underlying asset, which is similar to a digital certificate
of title or stamp of authenticity [3]. The method of
communication on a decentralized and distributed
blockchain, where each block contains a cryptographic
hash to form a network.
NFT stored in the blockchain, every transaction
occurring on the NFT will be recorded in a Smart contract
which will issue a unique code and store it. Smart contracts
themselves can simply be called the history of
transactions [1], smart contracts aim to make the
transaction process easier, more flexible, and more
efficient. With the Smart contract, it is possible to carry out
credible transactions without a third party [14]. Smart
contracts are created using a series of programming codes,
where there are parties who enter into contracts
automatically in the blockchain system.
Lennart Ante researched the Non-Fungible Token (NFT)
Market and Its Relationship with Bitcoin and
Ethereum [15]. This research obtained results where the
price of bitcoin triggers an increase in NFT sales and the
value of cryptocurrency in the market affects growth in the
NFT market but there is no opposite effect between NFT
and bitcoin and cryptocurrency. Borri et al. conducted
research on The Economics of Non-Fungible Tokens [16].
This research obtained results where NFT has low
exposure but significantly predicts returns on the NFT
market.
The main reason the author doesn’t use the collaborative
filtering approach in collaborative filtering requires user
likes or rating data, while in this research the data has an
imbalance where a lot of data do not have likes or ratings
compared to the other way around. If use this approach,
there will be inaccuracies in the model in the prediction
process.
B. Recommendation System
Recommendation system is a software tool and
technique that provide suggestions for items that are most
likely to be of interest to a particular user [17, 18]. The
recommendation system predicts a user’s rating or
preference for a particular item. These recommendations
are made based on past user behavior or other user
behavior [19]. So, the system will recommend something
to the user based on behavioral data or preferences over
time.
The recommendation system does not recommend
specific items but recommends a number of items that may
match the user’s preferences, the output of the
recommendation the system is a Top-N recommendation.
The main purpose of the recommendation system is to
increase product sales as well as sell various items (see Fig.
1).
Figure 1. Recommendation system.
C. Content Based Filtering
Content-Based Filtering uses the availability of content
(often also referred to as features, attributes, or
characteristics) of an item as a basis for providing
recommendations [20, 21]. The main reason for making a
recommendation system using this approach is that there
is little data available and the resulting recommendations
make users gain an understanding of why an item is
considered relevant to them. This method will work by
sorting items by top similarity [22]. While the drawback of
this approach is that the accuracy of the model depends on
the keywords entered [23] (see Fig. 2).
Figure 2. Content-based filtering.
519
Journal of Advances in Information Technology, Vol. 14, No. 3, 2023
III. METHODOLOGY
This research uses experimental research methods,
where the experimental method is carried out intentionally
by researchers by giving certain treatments to research
subjects in order to evoke an event or condition that will
be studied and the consequences [24]. This research is
divided into several stages including data collection, data
preprocessing, proposed model, testing and evaluation.
The research stage can be seen in Fig. 3.
Figure 3. Research stage.
Figure 4. Count cosine similarity using the output of TF-IDF.
Fig. 4 is the process or method carried out in obtaining
recommendations, first converting the data into vector,
generated cosine similarity, and then getting
recommendations.
IV. EXPERIMENTAL AND RESULT
This research uses NFT art collection 2021 data sourced
from Kaggle.com [25]. In this dataset there are 4 data
formats, including; gif, image, video, and CSV. The author
does not use the entire dataset, but only uses tabular data
in CSV format where in the dataset there are 4189 rows
and 15 columns, example datasets are shown in Fig. 5. In
the preprocessing stage, several steps are carried out,
converting the data series into list form, creating a
dictionary to determine key-value pairs from nft_name and
nft creator, and transforming to vector using TF-IDF (see
Fig. 1).
Figure 5. Sample dataset.
Next stage is the process carried out from the EDA
(Exploratory Data Analysis) process.
Fig. 6 shows of likes on NFT is lots of blanks compared
to an NFT that gets likes, with the above visualization
deciding to use a content-based filtering method instead of
a collaborative filtering method. Because the collaborative
filtering method requires rating data from users to get
recommendations (see Fig. 7).
Figure 6. Likes on NFT.
Figure 7. NFT frequency.
The next stage is conversion the dataset using TF-IDF
where TF-IDF is a representation scheme commonly used
in information retrieval and document extraction systems
that are relevant to certain queries. The technique is to find
the representation of the important features of each NFT
name category using the TF-IDF vectorizer() function
from the Sklearn library. The TF-IDF technique in this
research is a process in feature engineering to find text
representations and convert data to vector form which is
carried out before the modeling stage.
TF-IDF is defined by two quantities TF and IDF, where
TF (Term Frequency) works by measuring how often a
word or term appears in a particular text and normalizes
by dividing the number of occurrences by the length of the
document. The TF calculation formula can be seen in
Eq. (1). Whereas IDF (Inverse Document Frequency)
works by measuring the importance of terms across the
corpus, IDF considers terms that are very common
throughout the document and weighs rare terms. The IDF
calculation formula can be seen in Eq. (2), and the last step
is to multiply the TF value with IDF to get the TF-IDF
score Eq. (3).
520
Journal of Advances in Information Technology, Vol. 14, No. 3, 2023
(1)
(2)
(3)
The results obtained in vectorization can be seen in the
Fig. 8 below.
Figure 8. Vectorization results using TF-IDF.
The next stage is modeling using the process of
calculating the degree of similarity between NFTs with
cosine similarity, cosine similarity can find similarities
between one nft_name and another. The results obtained
in the process of calculating the degree of similarity using
TF-IDF can be seen in Fig. 9.
Figure 9. Vectorization results using TF-IDF.
Next step is to create the name recommendations’
function (see Fig. 10).
Figure 10. Name_recommendation function.
The name recommendation function has the meaning of
taking the highest number of k values from the similarity
data (in this case: cosine sim DF data frame). Then, take
the data from the highest to the lowest weight (level of
similarity). This data is entered into the closest variable.
Next, it is necessary to remove the NFT name that is being
searched for so that it does not appear in the list of
recommendations.
The recommendation system made using a content-
based filtering approach produces a very satisfactory
recommendation, where the results of recommendation
show that the name of the NFT and the inputted words
have similarities in 10 output names in the
recommendation system. Testing is done by calling the
name recommendation function shown in Fig. 11.
The main limitation of this research may occur when
using only NFT item names for recommendations. Using
only names for recommendations will not work when the
main NFT collection (For example: Bored Ape Yacht Club)
is used in most cases, as most have the same name as the
number.
Figure 11. Result recommendation system using a content-based
filtering.
V. CONCLUSION
Based on several scenarios starting from the analysis to
the testing stage, the following conclusions can be drawn,
the method proposed this research obtains results that can
recommend NFT names appropriately, and the approach
method used this research is suitable to be applied to the
recent NFT marketplace or new NFT marketplace, when
compared collaborative filtering method that requires
rating data from user. Recommendation system is made
based on the level of similarity. Common similarities can
be measured in the recommendation system, preferences
and tastes. In addition, similarities can also be found
through other data and information such as user
demographics and social status. Suggestions from the
author for further research to apply Exploratory Data
Analysis (EDA) in more detail, as well as the use of larger
datasets that can be obtained from the results of crawling
data from the NFT marketplace.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
AUTHOR CONTRIBUTIONS
Edi Surya Negara was the research lead carried out
research design development and supervised all research
development and lead the writing of manuscripts.
Sulaiman wrote program code to test the research model
521
Journal of Advances in Information Technology, Vol. 14, No. 3, 2023
that had been built. Ria Andryani assisted in the writing
and proofreading of the manuscript. Prihambodo Hendro
Saksono assisted in developing the research design, and
Yeni Widyanti assisted in data collection. All authors had
approved the final version.
ACKNOWLEDGMENT
The authors would like to thank Universitas Bina Darma,
Data Science Interdisciplinary Research Center and DIKTI
AI Centre, Directorate General of Higher Education,
Ministry of Education and Culture of the Republic of
Indonesia for the support and facilities provided.
REFERENCES
[1] E. S. Negara, A. N. Hidayanto, R. Andryani, et al., “Survey of smart
contract framework and its application,” Information, vol. 12, no. 7,
2021.
[2] N. Nurhachita and E. S. Negara, “A comparison between deep
learning, naïve bayes and random forest for the application of data
mining on the admission of new students,” IAES International
Journal of Artificial Intelligence, vol. 10, no. 2, 2021.
[3] E. S. Negara, K. Keni, and R. Andryani, “Data prediction for coffee
harvest using least square method,” in Proc. IOP Conference Series:
Materials Science and Engineering, IOP Publishing, 2020, vol.
1007, no. 1, 012193.
[4] Suryani Suyanto & Associates. (Apr. 2021). Bappebti explained
about the tax imposition plan on crypto. [Online]. Available:
https://www.ssas.co.id/bappebti-jelaskan-soal-rencana-pengenaan-
pajak-pada-kripto/ (in Indonesia)
[5] D. Liana, “Tracing the potential of NFT as a source of tax revenue,”
Center for Budget Studies, Agency for Expertise DPR RI, 2022.
[6] A. Popescu, “Non-Fungible Tokens (NFT)-innovation beyond the
craze,” in Proc. 5th International Conference on Innovation in
Business, Economics and Marketing Research, 2021.
[7] M. Irfan, A. D. Cahyani, et al., “Recommendation system: Online
book with collaborative filtering method,” Jurnal Teknologi
Technoscientia, pp. 76–84, 2014.
[8] F. Ferdiansyah, E. S. Negara, and Y. Widyanti, “Bitcoin-USD
trading using SVM to detect the current day’s trend in the
market,” Journal of Information Systems and Informatics, vol. 1, no.
1, pp. 70–76, 2019.
[9] W. E. Pangesti, R. Suryadithia, M. Faisal, et al., “Collaborative
filtering based recommender systems for marketplace applications,”
International Journal of Educational Research & Social Sciences,
vol. 2, no. 5, pp. 1201–1209, 2021.
[10] D. Piyadigama and G. Poravi, “An analysis of the features
considerable for NFT recommendations,” arXiv preprint,
arXiv:2205.00456, 2022.
[11] A. S. Ginting and S. Sadono, “Alternative exhibition space in the
form of a non-fungible token art market platform,” in Proc. Art &
Design, vol. 9, no. 2, 2022.
[12] R. M. Sukmariningsih, A. Nurudin, and E. Nursanty, “Imposition
of tax laws on cryptocurrencies and NFTs in Indonesia,” Owner:
Riset dan Jurnal Akuntansi, vol. 6, no. 2, pp. 1644–1654, 2022.
[13] K. B. Muthe, K. Sharma, and K. E. N. Sri, “A blockchain based
decentralized computing and NFT infrastructure for game
networks,” in Proc. Second International Conference on
Blockchain Computing and Applications (BCCA), 2020, pp. 73–77.
[14] E. Negara, A. Hidyanto, R. Andryani, and D. Erlansyah, “A survey
blockchain and smart contract technology in government agencies,”
in Proc. IOP Conference Series: Materials Science and
Engineering, vol. 1071, no. 1, 012026, 2021.
[15] L. Ante, “The non-fungible token (NFT) market and its relationship
with bitcoin and Ethereum,” FinTech, vol. 1, no. 3, pp. 216–224,
2022.
[16] N. Borri, Y. Liu, and A. Tsyvinski, “The economics of non-fungible
tokens,” SSRN, 2022, doi: 10.2139/ssrn.4052045
[17] M. Braunhofer, M. Elahi, and F. Ricci, “User personality and the
new user problem in a context-aware point of interest recommender
system,” in Information and Communication Technologies in
Tourism, Springer, 2015, pp. 537–549.
[18] S. Sutjiningtyas, A. A. Dharmawan, et al., “Design and build a shoe
product recommendation system for an online store using the user-
base collaborative filtering method,” Bulletin of Information
Technology (BIT), vol. 3, no. 2, pp. 143–148, 2022.
[19] E. S. Negara, D. Triadi, and R. Andryani, “Topic modelling twitter
data with latent Dirichlet allocation method,” in Proc. International
Conference on Electrical Engineering and Computer Science
(ICECOS), 2019, pp. 386–390.
[20] B. Tessy, S. Azvy, W. Yuwono, and I. Syarif. “Recommendation
system for property search using content-based filtering method,”
in Proc. 2018 International Conference on Information and
Communications Technology (ICOIACT), pp. 25–29, 2018.
[21] A. Wijaya and D. Alfian, “The laptop recommendation system uses
collaborative filtering and content-based filtering,” Jurnal
Computech and Bisnis, vol. 12, no. 1, pp. 11–27, 2018. (in
Indonesian)
[22] S. Suwarno and T. Fernando, “Design a design service marketplace
using the content-based filtering method,” Prosiding Seminar
Nasional Ilmu Sosial dan Teknologi (SNISTEK), no. 4, pp. 324–330,
2022. (in Indonesian)
[23] B. Mobasher, H. Dai, T. Luo, et al., “Improving the effectiveness
of collaborative filtering on anonymous web usage data,” in Proc.
the IJCAI 2001 Workshop on Intelligent Techniques for Web
Personalization (ITWP01), 2001, pp. 53–61.
[24] A. Jaedun, “Experimental research methodology,” Faculty of
Engineering UNY, vol. 12, 2011. (in Indonesia)
[25] A. D. Haan. (2021). NFT-art-dataset. [Online]. Available:
https://www.kaggle.com/datasets/vepnar/nft-art-dataset /
Copyright © 2023 by the authors. This is an open access article
distributed under the Creative Commons Attribution License (CC BY-
NC-ND 4.0), which permits use, distribution and reproduction in any
medium, provided that the article is properly cited, the use is non-
commercial and no modifications or adaptations are made.
522
Journal of Advances in Information Technology, Vol. 14, No. 3, 2023
