Research on the Correlation between Sentiment of Danmaku and the

Gift Value across Different Types of Live Streamers:

Dictionary-based Sentiment Analysis and Regression Test

Yuming Hu

Maastricht University, School of Business and Economics

Maastricht, Limburg, 6224 JT 5 a01, Netherlands

Keywords: Live Streaming, Live Streamers, Gift-Sending, Danmaku, Sentiment.

Abstract: Live streaming becomes more and more important as an entertaining method these days, and viewers can

interact with live streamers and others by sending gifts and sending danmaku. This paper had collected

1174040 pieces of danmaku data from 182 live streamers and the corresponding gift data from Huya TV from

Dec.7 to Dec. 23. The author investigated the relation between different types of sentiment of danmaku and

the gift value for entertaining live streamers and game live streamers collectively. To analyse sentiment of

danmaku, the sentiment dictionary made by Dalian University of Technology was adopted, and the regression

model was established to test the correlation. The results show that compliment is positively associated with

the gift value for entertaining live streamers while love and blame are positively correlated with the gift value

for game live streamers, but hatred is negatively associated with the gift value for game live streamers.

1 INTRODUCTION

Live streaming is a data transmission method used to

deliver a video file over the internet in real time

(Cloudfare 2021), and it allows anyone to share his or

her personal experience with the public at live events

(Lu, Xia, Heo, Wigdor 2018). According to Xinhua

net (Liu 2021), the number of live streaming users in

China in 2020 reached 0.617 billion. The market size

of worldwide video streaming industry was about 28

billion USD in 2020 (Live Streaming Market, 2021).

What is more, for the sake of real time, viewers could

interact with live streamers by sending danmaku, a

type of texts scrolling over the screen that every

viewer including live streamer could see, and by

sending valued gifts (Zhou, Zhou, Ding, Wang 2019).

With respect to the importance of gifts, several

figures can illustrate that (Platform Data of all live

streaming platforms in China, Toubang.tv 2021).

Ranked by net gift value in China, during September

2021, the top one live streamer received gifts worth

51 million CNY, the 100

live streamer received 3.5

million CNY; during October 2021, the top one live

https://orcid.org/0000-0001-8501-066X

streamer received 10 million CNY, and the 100th live

streamer received 2 million CNY.

This paper aims to dive deep into the relation

between sentiment of danmaku and gifts-giving

behaviours. Because sending danmaku and sending

gifts are major ways to interact with live streamers

and other viewers, and live streamers’ characteristics

will affect gift-giving intention (Li, Peng, 2021). This

paper argues that the pattern of emotion is distinct

among different types of live streamers and explores

their relationship on gift-giving behaviours.

Specifically, the research question is what types of

emotion can contribute to the gift value in

entertaining live streamer and game live streamers

respectively. In doing so, firstly, this paper will

introduce a different measurement of both sentiment

of danmaku and gift-giving behaviors compared to

the past literature. Secondly, this paper will propose

a carefully-designed data selection process. This

paper has several contributions to both past literature

and live streamers. First, it argues that the number of

gifts is not a good measurement of gift-giving

behaviours and instead, it adopts the gift value as a

proxy. Second, it analyses the correlation between

356

Hu, Y.

Research on the Correlation between Sentiment of Danmaku and the Gift Value across Different Types of Live Streamers Dictionary-based Sentiment Analysis and Regression Test.

DOI: 10.5220/0011178400003440

In Proceedings of the International Conference on Big Data Economy and Digital Management (BDEDM 2022), pages 356-360

ISBN: 978-989-758-593-7

different types of sentiment of danmaku and the gift

value while the past literature (Zhou, Ding, Wang,

2019) only took excitement-related emotion into

account. Third, it contributes to live streamers by

revealing that different kinds of live streamers should

be able to guide different sentiment pattern of

danmaku.

2 LITERATURE REVIEW

Previous literature researched gift-sending behaviors

through several lenses. Based on attachment theory

that a strong emotional bond between individuals and

a particular object can affect people’s gift-giving

intention (Ren, Harper, Drenner, Terveen, Kiesler,

Riedl, Kraut, 2012, Wan, Lu, Wang, Zhao, 2017), Li

and Peng (Li, Peng, 2021) researched the correlation

between live streamer characteristics and gift-giving

intention and the correlation between live scene

characteristics and gift-giving intention; their results

show that trustworthiness and attractiveness of live

streamers are positively associated with gift-giving

intention while the correlation between live scene

characteristics and gift-giving intention is not

significant. Li et al. (Li, Lu, Ma, Wang, 2021) studied

the impact of viewers’ identity on gift-giving

behaviours. Specifically, they explored class identity

and relational identity; viewers’ class identity was

measured by noble membership, a feature embedded

in live streaming platforms to distinguish viewers by

purchasing, and viewers’ relational identity was

measured by whether viewers wore fan badges;

results reveal that viewers’ class identity is positively

associated with the number of paid gifts and

negatively associated with the number of free gifts,

but relational identity is positively related with both

the number of paid gifts and free gifts. However, it is

remarkable that in most of live streaming platforms

in China, a fan badge is only worth 1 CNY, but

viewers can decide whose badge they take. In light of

social interaction theory that considers social

interaction an indispensable desire for human beings

(Baumeister, Leary, 1995), and that social interaction

can affect people’s behaviours (Castilla 2005, Rogers

2010), Zhou et al. (Zhou, Zhou, Ding, Wang, 2019)

investigated the association between danmaku and

the number of gifts, and found that the number of

words, the number of excitement-related words, the

similarity of danmaku, and debate level of danmaku

are positively associated with the gift value.

Nevertheless, it has several limitations. First, the

number of gifts is not a good measure for gift-giving

behaviours as differences among the gift value could

be extremely large. For example, the most expensive

gift in Huya TV (https://www.huya.com) is worth

5000 CNY while the cheapest gift is worth only 0.1

CNY. Second, it only counted the number of

excitement-related words, but words could have

different types of emotions such as happiness,

appraising, and peacefulness.

3 METHODOLOGY

3.1 Data Selection

This paper takes two factors into account in the

process of data selecting. Firstly, Zhu et al. (Zhu,

Yang, Dai, 2017) argue that most gifts in live

streaming platforms are channeled into a few live

streamers. It implies that if people want to research

gift-giving behaviors, they should focus on relatively

large live streamers as they are the most important.

Secondly, there is one thing that past literature

ignored but is important. That is the difference

regarding gift-giving behaviours among different

types of live streamers can be extremely large;

according to Toubang (Live streamers’ list ranked by

gifts value, https://www.toubang.tv), the average gift

value per live streamer in Beauties and star-show

section in China between Nov. 25 and Dec. 25 is 16

thousand CNY while the average gift value in League

of legends, Honour of kings, and Game for peace is

no more than 1 thousand CNY for each. In view of

these points, this paper crawled data of Huya TV

(http://huya.com) from Dec. 7 to December. 23. It is

one of the biggest live streaming platforms in China

and filters datafiles (every datafile is a collection of a

single live streamer) less than 100 kb. The 100 kb

criterion is determined arbitrarily, and the descriptive

information of the dataset will be given later.

The dataset contains the following features: the

live streamer’s ID, the type of channel, the gift name,

the gift number, and danmaku. As shown in Table 1,

182 examples were collected, and there are four types

of live streamers, Beauties and star-show, League of

legends, Honour of kings, and Game for peace.

Specifically, 60 observations are for Beauties and star

show, 44 observations are for League of legends, 42

examples are for Honour of kings, and 36 examples

are for Game for peace. Concerning the gift value and

danmaku number, examples from Beauties and star-

show had obtained 819817 CNY and 226220 pieces

of danmaku; examples from League of legends had

obtained 152947 CNY and 273592 pieces of

danmaku;

examples from Honour of kings had

Research on the Correlation between Sentiment of Danmaku and the Gift Value across Different Types of Live Streamers Dictionary-based

Sentiment Analysis and Regression Test

357

Table 1: Descriptive information of gift and danmaku.

Type of channels

live streamers’ numbe

Gift value

(

CNY

)

Pieces of danmaku

Beauties and star-show 60 819817 226220

League of legends 44 152947 273592

Honor of kings 42 235999 508051

Game for peace 36 121533 166177

Total 182 1330296 1174040

received 235999 CNY and 508051 pieces of

danmaku, and cases from Game for peace had gained

121533 CNY and 166177 pieces of danmaku.

3.2 Research Design

In order to analyse different types of sentiment for

Chinese danamku, this paper introduced a sentiment

dictionary made by Dalian University of

Technology(http://ir.dlut.edu.cn/info/1013/1142.htm

). It includes 27466 sentiment words of 6 major

sentiment types, happiness, praising, anger, sadness,

fear, hatred, and amazement, and these major

sentiment types can be extended into 21 sub-types.

For example, praising can be extended into respect,

compliment, trust, love, and wish. In the stage of data

cleaning, this paper adopted Jieba

(https://github.com/fxsjy/jieba) to tokenize Chinese

words. Moreover, this paper compared the proportion

of emotional words between Beauties and star show

and other types of live streamers, and ranked the most

common 4 types of sentiment. The proportion of a

specific type of sentiment is defined as the proportion

of the number of this type of sentiment to the number

of all emotional words. Table 2 shows that

compliment and blame are the most common

sentiment regardless of types of live streamers. In

three types of game live streamers, happiness is the

third most frequent sentiment, but in live streamers of

Beauties and star-show, love is the third most

frequent sentiment. What is more, this paper

categorized League of legends, Honour of kings, and

Game for peace into the game cluster and grouped

Beauties and star show into the entertaining cluster

Table 2: Comparison of different types of sentiment.

e of channels 1st 2n

4th

Beauties and sta

-show PH, 52% NN, 15% PB, 11% PA, 5%

Lea

ue of le

ends PH, 29% NN, 19% PA, 8% ND, 7%

Honor of kin

s PH, 31% NN, 18% PA, 9% ND, 8%

Game for peace PH, 48% NN, 12% PA, 10% PB, 8%

Note: PH indicates compliment-related words, NN indicates the number of blame-related words, PB indicates love-related

words, PA indicates happiness-related words, ND indicates hatred-related words

In light of these results, this paper constructed a

research model whose dependent variable is the

logarithm of the gift value to the base e and whose

independent variables include the number of

compliment-related words, the number of blame-

related words, the number of love-related words, the

number of happiness-related words, and the number

of hatred-related words.



𝐺𝑖𝑓𝑡𝑠 =𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡+𝛽



∗𝑃𝐻+𝛽



∗𝑃𝐴+

+𝛽



∗𝑃𝐵+𝛽



∗𝑁𝑁+𝛽



∗

𝑁𝐷 (1)

4 RESULT ANALYSIS

Table 3 shows the regression results of the impact of

different types of sentiment on the gift value. Panel 1

in Table 3 is regression results for entertaining live

streamers. The number of compliment-related words

is the only significant variable as its p-value is 4%

smaller than 5%. It indicates that compliment is the

only significant emotion out of five emotions for

entertaining live streamers because its p-value is

smaller than 5%. Its coefficient is 0.0008, which

means compliment is positively associated with the

gift value for entertaining live streamers, and every

additional compliment-related word will increase

0.08% of the gift value. Panel 2 in Table 3 is

regression results for game live streamers.

Significant variables include the number of love-

related words, the number of blame-related words,

and the number of hatred-related words as the p-value

is less than 5% for each. The coefficients of the

number of love-related words and the number of

BDEDM 2022 - The International Conference on Big Data Economy and Digital Management

358

blame-related words are 0.0045 and 0.0019,

respectively, which indicates that love and blame are

both positively correlated with the gift value received

by game live streamers. The coefficient of the number

of hatred-related words is -0.0038, which indicates

that hatred is negatively associated with the gift value

for game live streamers, and an additional hatred-

related word will decrease 0.38% of the gift value.

Table 3: Regression results for entertaining live streamers and game live streamers.

Panel 1

Entertaining live

treamers Panel 2

Game live

treamers

Coefficients

-value Coefficients

-value

Interce

t 7,6579 0,00 6,8857 0,00

PH 0,0008 0,04 -0,0003 0,45

PA 0,0028 0,24 0,0009 0,47

PB -0,0003 0,73 0,0045

0,03

NN 0,0004 0,39 0,0019 0,03

ND 0,0019 0,29 -0,0038 0,00

Note: PH indicates the number of compliment-related words, NN indicates the number of blame-related words, PB indicates

the number of love-related words, PA indicates the number of happiness-related words, ND indicates the number of hatred-

related words

According to the results, it is clear that the

importance of different types of sentiment is distinct

between entertaining live streamers and game live

streamers. What is the reason behind that? Li and

Peng (Li, Peng, 2021) argues that different

characteristics of live streamers have an impact on

gift-sending intention, but they did not state that

different characteristics of live streamers could lead

to different desired sentiment pattern regarding the

gift value In general, entertaining live streamers are

more attractive compared to game live streamers, but

game live streamers are more trustful compared to

entertaining live streamers because entertaining live

streamers exhibit themselves to viewers while game

live streamers exhibit the game content to viewers.

Though do not know the actual mechanism behind

that, this paper believes that different characteristics

of live streamers is the cause of different sentiment

patterns.

Regarding the positive coefficient of the number

of blame-related words for game live streamers, the

result is quite striking as intuitively the author

expected that to be negative. In real life, many

viewers send blame-related danmaku especially after

live streamers made some mistakes. To some extent,

viewers throw blame on game live streamers’

shoulders to show they really engage with them, and

hence it is positively associated with gifts value.

5 CONCLUSIONS

This paper examined the relationship between

different types of sentiment of danmaku and the gift

value for entertaining live streamers and game live

streamers. To conclude, different types of live

streamers require different sentiment patterns to

increase their gift value. Concerning entertaining live

streamers, compliment is positively associated with

the gift value. However, for game live streamers, love

and blame are positively related with the gift value,

but hatred is negatively related with the gift value.

This research has two important implications for

live streamers. First, live streamers should realize

their different characteristics compared to others.

Entertaining live streamers should try to exhibit their

attractiveness to viewers so as to make them express

compliment. Game live streamers should try their

best to earn viewers’ trust and make viewers love

them; meanwhile, game live streamers need to pay

special attention to hatred-related danmaku as it can

harm their gift value. Second, game live streamers

should know how to get along with blame-related

danmaku as the results show that blame is actually

positively associated with the gift value.

This research also has one implication for further

research. The dataset only includes two types of live

streamers and 182 examples. Though this can give

people a basic understanding of the research question,

a larger dataset might be better.

ACKNOWLEDGEMENT

Here, I would like to thank Qing Yin and Tianjiao

Yang for providing valuable ideas regarding the

structure and format of this paper.

Research on the Correlation between Sentiment of Danmaku and the Gift Value across Different Types of Live Streamers Dictionary-based

Sentiment Analysis and Regression Test

359

REFERENCES

Baumeister, R. F., & Leary, M. R. (1995). The need to

belong: desire for interpersonal attachments as a

fundamental human motivation. Psychological bulletin,

117(3), 497.

Castilla, E. J. (2005). Social networks and employee

performance in a call center. American journal of

sociology, 110(5), 1243-1283.

Cloudfare. 2021. What is live streaming? | How live

streaming works. [online] Available at:

<https://www.cloudflare.com/learning/video/what-is-

live-streaming/> [Accessed 18 December 2021].

Li, R., Lu, Y., Ma, J., & Wang, W. (2021). Examining

gifting behavior on live streaming platforms: An

identity-based motivation model. Information &

Management, 58(6), 103406.

Li, Y., & Peng, Y. (2021). What drives gift-giving intention

in live streaming? The perspectives of emotional

attachment and flow experience. International Journal

of Human–Computer Interaction, 1-13.

Liu, Y., 2021. The live streaming users reached to 0.617

billion in 2020 in China. [online] Xinhuanet. Available

at: <http://www.xinhuanet.com/2021-

05/19/c_1127464095.htm> [Accessed 18 December

2021].

Live streamers’ list ranked by gifts value. Toubang.tv.

[online]. Available at:

<http://www.toubang.tv/data/anchor/gift?dt=1>

[Accessed 31 December 2021].

Live Streaming Market: Global Industry Analysis and

Forecast (2021-2027) Component, a., 2021. Live

Streaming Market. [online] MAXIMIZE MARKET

RESEARCH. Available at:

<https://www.maximizemarketresearch.com/market-

report/global-live-streaming-market/23862/>

[Accessed 20 December 2021].

Lu, Zhicong, Haijun Xia, Seongkook Heo, and Daniel

Wigdor. "You watch, you give, and you engage: a study

of live streaming practices in China." In Proceedings of

the 2018 CHI conference on human factors in

computing systems, pp. 1-13. 2018.

Platform Data of all live streaming platforms in China.

Toubang.tv. [Online]. Available at: <

http://www.toubang.tv/data/platform> [Accessed 25

December 2021].

Ren, Y., Harper, F. M., Drenner, S., Terveen, L., Kiesler,

S., Riedl, J., & Kraut, R. E. (2012). Building member

attachment in online communities: Applying theories

of group identity and interpersonal bonds. Mis

Quarterly, 841-864.

Rogers, E. M. (2010). Diffusion of innovations. Simon and

Schuster.

Wan, J., Lu, Y., Wang, B., & Zhao, L. (2017). How

attachment influences users’ willingness to donate to

content creators in social media: A socio-technical

systems perspective. Information & Management,

54(7), 837-850.

Zhou, J., Zhou, J., Ding, Y., & Wang, H. (2019). The magic

of danmaku: A social interaction perspective of gift

sending on live streaming platforms. Electronic

Commerce Research and Applications, 34, 100815.

Zhu, Z., Yang, Z., & Dai, Y. (2017, July). Understanding

the gift-sending interaction on live-streaming video

websites. In International Conference on social

computing and social media (pp. 274-285). Springer,

Cham.

BDEDM 2022 - The International Conference on Big Data Economy and Digital Management

360