Research on the Impact of UGC Based on Fluctuation Mode on

Cryptocurrency Market

Kun Jia

1,2

, Yizhen Zhu

and Yuxin Zhang

1,*

Institute of Artificial Intelligence and Change Management, Shanghai University of International Business and Economics,

Shanghai, China

School of Statistics and Information, Shanghai University of International Business and Economics, Shanghai, China

Keywords: Blockchain, Cryptocurrency, Fluctuation Pattern, Sentiment Analysis, Investment Advice, Financial Public

Opinion.

Abstract: As the investment properties and financial properties of cryptocurrencies are accepted or recognized by major

listed companies and even sovereign countries, they are increasingly sought after by ordinary investors around

the world. The frantic influx of a large number of investors into the cryptocurrency market has also led to

more and more market analysis and predictions for retail investors and new investors, as well as popular

science teaching videos on the YouTube website. However, their sentiment towards cryptocurrency is not

must be "right". To ascertain the attitude of video producers in YouTube videos towards the market, this article

uses sentiment analysis to analyze the videos and compare the similarity with the Bitcoin fluctuation data

before and after the video upload. We found that in most cases, video producers ’attitudes have a high degree

of similarity with the ups and downs of the period before the upload of the video, but the low similarity to the

trend for some time in the future. This shows that the cryptocurrency-related videos on YouTube merely reflect

past ups and downs and cannot be used as investment advice, even if the video uploaders are doing so.

1 INTRODUCTION

In January 2021, the "GameStop" incident in the US

stock market suddenly broke out. It was based on the

retail accounts of the largest US forum website Reddit

and the stock topic forums with millions of fans on

the website. and concentrated buying, which was

strongly shorted by hedge funds. The company’s

GameStop stock, and used the power of the group to

push up the price of the stock, forcing hedge funds

Citron Research, Melvin Capital, and other short

positions to surrender. In the end, the hedge funds

ended disastrously, wrote a historic page of Wall

Street.

Reddit netizens who are proud to carry forward

the MEME picture will naturally not miss the

Dogecoin born because of the Doge emoji. Elon

Musk can be described as the best "cargo carrier" for

Dogecoin. The price of Dogecoin has skyrocketed

many times before, in many cases, it is derived from

his tweets, which can be traced back to April 2019. At

that time, the official Dogecoin account launched a

vote for the Dogecoin CEO on Twitter, and Musk was

elected with high votes. Subsequently, Musk tweeted

that Dogecoin is his favorite digital currency and

changed his Twitter account information to "Former

Dogecoin CEO". On April 15 this year, Musk posted

a picture of "dogs roaring on the moon" on social

platforms, implying that he will bring Dogecoin to the

moon, and therefore Dogecoin has a "To the Moon"

Slogan. And his series of actions have also attracted

investors' pursuit, and the price of Dogecoin has risen

all the way.

Since the "GameStop" incident, people seem to be

more willing to believe in the power of self-media

and social media than large companies or institutions.

In particular, a large number of new cryptocurrencies

continue to emerge. In the 24-hour cryptocurrency

market where prices soar and plummet at any time,

everyone hopes to reproduce the myth of Bitcoin in

themselves and get a share in this frenetic market.

However, we compared the actual fluctuations before

the video was released with the binary data generated

after sentiment analysis on the subtitles of the video

released on YouTube, and found that:

In the English video, the similarity between the

two reached 78% and comparing the actual

fluctuations after the video was released, it was found

Jia, K., Zhu, Y. and Zhang, Y.

Research on the Impact of UGC Based on Fluctuation Mode on Cryptocurrency Market.

DOI: 10.5220/0012042400003620

In Proceedings of the 4th International Conference on Economic Management and Model Engineering (ICEMME 2022), pages 707-712

ISBN: 978-989-758-636-1

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

707

that the similarity was only 51.1%. This is almost the

same as the probability of blindly guessing whether

the rise or fall. It is also for investors. It does not

constitute any investment advice. In Chinese video,

the similarity of the former is as high as 99.2%, and

the similarity of the latter is 64.2%. This may be due

to the small amount of data and the fact that the video

content is mostly positive.

2 DESCRIPTIVE STATISTICS

2.1 YouTube Data Descriptive Statistics

Using data mining techniques to search for videos on

YouTube based on the relevance of keywords such as

"Bitcoin", "Ethereum", "cryptocurrency",

"Dogecoin" in Chinese and English, and obtain 1408

valid data in Chinese search results. There are 245

pieces of subtitles that can be generated; 2131 pieces

of valid data in English search results are obtained, of

which 1,691 pieces of subtitles can be generated. The

following does not give special instructions, which

means that the data is as of June 10

(The data source

will be available at https://github.com/SUIBE-

jk/YouTube-DATA).

Figure 1: Chinese and English retrieval to get the number of videos in each period.

As you can see, with the support of the YouTube

video recommendation system, most of the data we

get has been uploaded recently. Of course, this is also

related to the surprising increase in cryptocurrency in

the fourth quarter of last year and the first quarter of

this year.

Make word cloud diagrams for the titles of all the

search results in both Chinese and English. It can be

seen that keywords such as "analysis" and "forecast"

which guide trading appear more frequently; while

"big rise" and "big drop" appear Frequent words that

are eye-catching; in the cryptocurrency market,

Musk, who has "one call to a hundred responses",

appears very frequently in video titles, usually in the

same field as Dogecoin.

ICEMME 2022 - The International Conference on Economic Management and Model Engineering

708

Figure 2: Comparison of Chinese and English word cloud diagrams.

2.2 Cryptocurrency Data Descriptive

Statistics

The data on the rise and fall of the cryptocurrency

used in this article comes from investing.com.

Because of the predecessors' understanding of

cryptocurrency and the research on the risk of

cryptocurrency, it has been very mature and

complete. However, domestic research on

cryptocurrency mainly focuses on Bitcoin, the

original cryptocurrency, and at the same time, there is

not enough attention to the practical application value

of later new cryptocurrencies. At present, there is no

research on whether there is a phenomenon of self-

media misleading investors with general emotional

bias in YouTube videos. Therefore, this article takes

the overall emotional bias in YouTube videos as the

research object and compares the fluctuation patterns

obtained through Bitcoin's daily fluctuations.

Table 1: Data example.

Date Close** O

en* Hi

h Low Volume Ch

2021/6/29 36,342.30 34,477.30 36,410.30 34,247.60 102.51

5.41%

2021/6/28 34,475.90 34,682.20 35,231.20 33,944.90 112.00

-0.58%

2021/6/27 34,678.50 32,247.10 34,685.50 32,041.70 148.80

7.55%

2021/6/26 32,243.40 31,592.10 32,643.00 30,206.90 156.67

2.06%

3 FLUCTUATION PATTERN

By observing the morphology of the time series, it is

found that the morphology of the time series is

composed of three states: rising, maintaining, and

falling. In this article, to dig out the morphological

similarity between the sequences in the massive time-

series data set, we propose a fluctuation pattern. The

concept of FP) saves the fluctuation trend of the

original sequence through the extended Boolean time

series, uses {1, 0,-1} as the value range of the

Boolean time series data to save the original time

series data change trend, and only pays attention to

the original time series. The morphological

characteristics of the ordinal data, regardless of its

numerical value, are represented by 1 as increasing, 0

as maintaining, and -1 as descending. In the

subsequent similarity calculation, the generated

Boolean time-series data is used for similarity

measurement.

Research on the Impact of UGC Based on Fluctuation Mode on Cryptocurrency Market

709

For 𝑋𝒹



，

𝒹



，

⋯

，

𝒹



 , there are

Boolean time-series data 𝐵𝑏



，

𝑏



，

⋯

，

𝑏





and:

𝑏





1 𝑑



 𝑑



0 𝑑



 𝑑



1 𝑑



 𝑑



(1)

B records all the fluctuations of X, so B is called

the fluctuation pattern of X.

4 EMOTION ANALYSIS

Text sentiment analysis can be roughly divided into

two categories: one is based on the sentiment

dictionary to get the sentiment score of the text. The

second is based on machine learning. First, use the

manually labeled text (whether the comment has been

marked as a positive comment or a negative

comment), and put in various algorithm models (such

as Naive Bayes, SVM, etc.) for training, and finally

realize the new comment Classification is essentially

text classification. In this case, the data content is the

subtitle content of the video crawled on the YouTube

platform, that is, the video producer's product

comment on the encrypted currency. The useful

variable in the data is the "comment". We use this data

to obtain a sentiment of the up master on the rise and

fall of cryptocurrencies. The framework diagram of

the sentiment analysis method used in this article is

as follows:

Figure 3: Sentiment analysis method framework image.

4.1 Chinese Text

Based on the sentiment analysis of Chinese text, we

realized it by using snownlp. Through this case, we

learned about snownlp and directly called

SnowNLP(txt).sentiments to calculate the sentiment

score, which filled the gap that the installed library

could not be called, and successfully used the

function to realize the code. The idea of reuse, this

case directly calls the pre-trained algorithm model of

Snownlp, without actually training and tuning the

algorithm model, and the accuracy may need to be

improved. Snownlp can mainly perform Chinese

word segmentation (the algorithm is Character-Based

Generative Model), part-of-speech tagging (the

principle is TnT, 3-gram Hidden Markov), sentiment

analysis, text classification based on the principle of

naive Bayes, pinyin conversion, traditional to

simplified, Extract text keywords and abstracts,

segment sentences, and text similarities based on

TextRank. It is predicted from the model that the up

subject of each video corresponds to the emotional

tendency of the cryptocurrency problem. The

emotional tendency is divided into positive and

negative, which is a typical two-category problem.

4.2 English Text

TextBlob is a Python library for processing text data.

It provides a simple API for common natural

language processing (NLP) tasks, such as part-of-

speech tagging, noun phrase extraction, sentiment

analysis, classification, translation, etc. In this case,

we use the textblob library to do sentiment analysis

on English text. The results are obtained through

word extraction, sentence sentiment value

ICEMME 2022 - The International Conference on Economic Management and Model Engineering

710

calculation, and syntactic analysis. TextBlob has been

looking for words and phrases that can be assigned

polarity and subjectivity and averaged them together.

This is the principle when processing long texts, that

is, simple average.

Table 2: English sentiment analysis example.

title-ID post_title post_content sentiment

sentiment_i

ndex

0 0 NO. 1 *** Negative -1

1 2 NO. 2 *** Positive 1

2 3 NO. 3 *** Negative -1

3 4 NO. 4 *** Positive 1

4 5 NO. 5 *** Positive 1

5 6 NO. 6 *** Negative -1

5 RESULTS & DISCUSSION

We compared the actual fluctuations of a certain

period before the video was released and the time

series generated after emotional classification of the

related video subtitles released on YouTube at this

time using the longest common subsequence (LCS)

and found that two The similarity of the video

producers is greater, that is, the emotional tendency

of the video producer is largely based on the previous

rise and fall of the market, and the actual rise and fall

of the video after the release of the video is compared

with this time series data, and it is found that the

similarity is relatively high. Small, that is, the

emotional tendency of the video publisher and cannot

play a role in predicting the rise and fall of the market.

Table 3: Comparison of results.

Chinese data English data

Before upload

LCS

99.2% 78.0%

After upload

LCS

64.2% 51.1%

6 CONCLUSIONS

This research takes the content of YouTube video

subtitles as text data as an example, uses word cloud

graphs and topic model feature analysis to analyze the

text data of video subtitles content, and uses a python-

based emotional dictionary method to calculate the

emotional score of each comment one by one. The

research analyzes the sentimental tendency of up

owners towards cryptocurrencies and provides

suggestions for investors based on the research results

to help meet the expectations and needs of investors,

thereby reducing the risk of investment. In the

process of text mining, this study failed to identify

and eliminate false comments, and the constructed

dictionary was not complete, and there was missing

vocabulary. Future research will pay more attention

to the quality and authenticity of the data itself and

improve the construction of the dictionary. However,

this case is based on real-world real data, using data

mining methods to analyze the emotional tendency of

video uploaders to cryptocurrency, and provides a

data science research paradigm for investors to make

corresponding investment behaviors.

ACKNOWLEDGMENTS

This work was supported by Shanghai University of

International Business and Economics Postgraduate

Research Innovation Cultivation Program.

REFERENCES

Fang Y, Sugano K, Oku K, Kawagoe K. (2015) Applying a

Multi-dimensional Time-Series Similarity Method to

Typhoon-track Prediction. In: Proceedings of the 2015

IEEE 11th International Conference on e-Science (E-

SCIENCE’ 15). Washington. 259–262.

Fu M, Wang D. (2020) Service quality evaluation of fresh

agricultural products cold chain logistics based on

principal component and neural network. In: IOP

Conference Series: Earth and Environmental Science.

Hulun Buir. 585: 012103.

Li Y, Lu L. (2019) Research on B2C Reverse Logistics

Service Quality Evaluation System. In: Proceedings of

the 2019 5th International Conference on E-Business

and Applications, Bangkok. 10-15.

Research on the Impact of UGC Based on Fluctuation Mode on Cryptocurrency Market

711

Liu B, Zhang Y. (2011) Research on evaluation of third-

party logistics service quality based on dynamic fuzzy

sets. In: MSIE 2011. Harbin. 833-837.

Qian C, Wang Y, Hu G, Guo L. (2015) A novel method

based on data visual autoencoding for time series

similarity matching. In: Proceedings of the 27th

Chinese in Control and Decision Conference. Qingdao.

2551-2555.

Wu J, Lu K. (2019) Chinese weibo sentiment analysis based

on multiple sentiment lexicons and rule sets. Computer

Applications and Software, 36(09): 93-99.

Zhang J, Liu X. (2017) Evaluation of Integrated Logistics

Service Based on SERVQUAL Model. In: International

Conference on Computer Systems, Electronics and

Control (ICCSEC). Dalian. 100-104.

ICEMME 2022 - The International Conference on Economic Management and Model Engineering

712