Herding Behavior-Driven Comedy Film Rating Convergence: A Study of

the Moderating Effects of Emotional Intensity

Yuyang Wang

Beijing Jiaotong University, No.3, Shangyuan Village, Haidian District, Beijing, China

Keywords: Movie Ratings, Herding Behavior, Sentiment Analysis.

Abstract: The study examines the herding effect in comedy films, particularly how user ratings and comments influence

rating behavior. Using data from the Douban movie platform, sentiment and time-series analyses were

employed to assess rating convergence and the amplification effect of emotional intensity. Key findings

include: low-scored films experience large initial score variance fluctuations that stabilize over time, with

emotional factors initially worsening fluctuations but promoting long-term consistency; medium-rated films'

emotional resonance varies by subject, significantly affecting convergence speed; and high-scored films show

obvious rating convergence, less influenced by emotional intensity. Audience rating behavior is impacted by

time, film quality, emotional intensity, and social interaction. The study offers a new angle for the film

industry's marketing strategies, advising rating platforms to enhance sentiment analysis algorithms for more

accurate emotional tendency capture and improved scoring system reliability.

1 INTRODUCTION

With the widespread adoption of the Internet and

social media, user-generated content (UGC) has

emerged as a crucial medium for individuals to

express their opinions. Among various forms of UGC,

online reviews are particularly prevalent across

industries such as film, dining, and e-commerce

(Chevalier and Mayzlin, 2006). For the film industry,

as a kind of online word-of-mouth, users' online

comments not only affect other audiences' movie-

watching choices, but also may have a profound

impact on film box office and long-term word-of-

mouth (Samrat et al., 2024). Especially on platforms

such as Douban, user ratings and comments have

become important criteria to measure the quality of

movies.

However, user rating behavior can be significantly

influenced by pre-existing average ratings, leading to

convergence phenomena, wherein users tend to align

their ratings more closely with the existing consensus.

The behavioral tendency, termed the herding effect,

reflects how individuals adjust their evaluations in

response to the prevailing ratings. Furthermore, the

sentiment embedded in prior reviews can trigger

emotional resonance among users, thereby

influencing their own rating and evaluation behavior

(Qu et al., 2024). The emotional contagion effect may

further exacerbate the herding effect in online rating

environments. Additionally, different types of movies

have different types of audiences (Urszula, 2019),

which may also lead to different degrees of herding

effect among the audience. For example, the degree

of herding effect of comedy movies may be more

dependent on the emotional tendency of users because

of its relaxed and pleasant characteristics. The

questions are of great significance for understanding

the psychological mechanism behind user evaluation

behavior and the optimization of marketing strategies

in the film industry.

2 LITERATURE REVIEW

The herding effect refers to the tendency of

individuals to conform to group behaviors under

social influence, often disregarding their independent

judgment. The phenomenon has been extensively

studied across disciplines such as social psychology,

economics, and behavioral finance. In social

psychology, herding is closely related to conformity,

where individuals adopt group behaviors to gain

social acceptance or avoid exclusion, even when

those behaviors contradict personal opinions (Bond

Wang, Y.

Herding Behavior-Driven Comedy Film Rating Convergence: A Study of the Moderating Effects of Emotional Intensity.

DOI: 10.5220/0013841700004719

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on E-commerce and Modern Logistics (ICEML 2025), pages 207-213

ISBN: 978-989-758-775-7

207

and Smith, 1996). Salganik et al. (2006), through their

music download experiment, revealed how social

influences and herding can lead to inequality and

unpredictability in cultural products. The study found

that when participants were presented with

information about the choices of others, their choice

behavior was significantly affected, resulting in the

popularity of certain cultural products far higher than

others. The phenomenon is also relatively common in

movie rating, and users' rating behavior may be

guided by the existing average rating, and then show

the phenomenon of convergence. Hao et al. (2019),

taking the film industry as an entry point, studied the

decision-making behaviors of consumers under the

influence of social media and found that the herd

effect of consumers still exists in the film field. Lee

et al. (2015) compared the effects of strangers and

friends on user-generated movie reviews and found

that users would be influenced by the mainstream

ratings of previous movies, and would follow the

mainstream or deliberately express different opinions

instead of their own opinions.

Sentiment analysis is the automatic identification

and extraction of emotional tendencies in text through

computational methods. Common methods include

dictionary-based methods and machine learning

methods. Pang and Lee (2008) conducted a

systematic review of opinion mining and sentiment

classification, highlighting that dictionary-based

methods demonstrate superior accuracy in domains

such as film criticism. Taboada et al. (2011) further

explored the efficacy of various sentiment

dictionaries, emphasizing optimization strategies to

enhance classification precision.

Furthermore, audience evaluation behaviors differ

across movie genres. Urszula (2019) noted that genre-

specific factors influence rating patterns. Hao Yuan et

al. (2020) believe that audience's evaluation of

comedy movies is often influenced by factors such as

plot, actor performance and visual effects, so the

rating of comedy movies is more dependent on users'

emotional tendency (Zhang et al., 2021).

To sum up, existing studies have revealed the

widespread existence of herding effect in fields

including movie ratings. However, for the specific

genre of comedy movies, the herding mechanism of

user ratings still needs further research. In particular,

existing research has focused less on how the intensity

of reviews affects the herding effect of ratings, and

whether the effect differs across films of different

ratings levels. Through empirical analysis, the study

will explore the role of user rating and comment

emotion in the herd effect of comedy movies, and

provide a new perspective for the research in the field.

3 RESEARCH PROBLEM

The aim of the study is to investigate the influence of

user ratings and review sentiment on herding effects

in comedy films. Specifically, the research will

address the following key questions:

1. In comedy movies, is there a convergence of

user ratings, that is, does the degree of

deviation from the average rating decrease

with the change of the average rating?

2. Does the emotional intensity of reviews

amplify the herding effect of ratings?

3. Does the herding effect differ among films

with different ratings?

4 DATA PROCESSING

4.1 Data Source

The dataset utilized in the study was sourced from the

Douban Movie Platform, with data collection

conducted via the Octopus web crawler platform. The

raw dataset included various attributes such as movie

titles, user ratings, short review texts, and release

dates. To ensure data representativeness and validity,

the study adopted a series of rigorous selection and

processing measures. First of all, as one of the largest

film communities in China, Douban Film platform

has a huge user base and rich movie data, and its

ratings and reviews have high representativeness and

reference value. Users participate in evaluation and

scoring spontaneously on the Douban platform, and

the data can better restore the average views of the

general public (Liu, 2018). In addition, Douban's film

ratings and reviews are often cited and referenced by

media reports and even professional film reviews,

further demonstrating the authority and reliability of

its data. (Shi, 2024)

Secondly, the Octopus web crawler platform was

employed due to its user-friendly, code-free operation

and robust data extraction capabilities. It offers built-

in templates, extensive network data capture support,

and advanced deduplication and filtering functions,

ensuring the quality and uniqueness of the extracted

data while minimizing redundancy. The features

enhance the efficiency and accuracy of data

acquisition, thereby strengthening data reliability.In

the process of data collection, in order to control the

consistency of film types, the study focuses on

comedy films (Douban film label includes comedy),

and randomly selects films released in recent years as

ICEML 2025 - International Conference on E-commerce and Modern Logistics

208

research objects. According to the overall film score

of the platform, movies are divided into low grades

with less than 6 points, medium grades with 6-9 points

and high grades with more than 9 points. The low and

middle grades selected five films each, while the high

grades selected all three films due to the relatively

small number of films. 400 reviews and ratings were

extracted for each film according to its popularity,

which ensured that each sample film had enough data

for analysis, and covered reviews of different

popularity to improve the comprehensiveness and

representation of the data. In the data preprocessing

stage, blank comments and unrated invalid comments

were cleared, and 4982 valid comments and their

scores were obtained. Through such data screening

and cleaning, the quality and effectiveness of the data

are further guaranteed, and a reliable data basis is

provided for subsequent research and analysis.

4.2 Data Measurement

In the present study, emotion dictionaries are utilized

to determine the emotional intensity of reviews. As

the general Chinese emotion dictionaries are not

capable of explaining movie reviews, emotion

dictionaries suitable for movie reviews are used in the

emotion analysis (Wang, et al., 2022), which not only

merged and cleaned the well-known Hownet

dictionaries and NTUSD dictionaries. A dedicated

sentiment dictionary has also been constructed

specifically for the field of film, which is able to more

accurately capture the emotional tendencies in film

reviews. Compared to the improved Hownet

dictionary and NTUSD dictionary, the dictionary is

8.1% and 10% more accurate at capturing emotions,

respectively. In the dictionary, the weight of positive

emotion words is set to 1, and the weight of negative

emotion words is -1. At the same time, the dictionary

of degree adverbs is used to match degree adverbs in

reviews and assign corresponding weights according

to their impact on emotional intensity. In the

calculation of emotion value, the weight of emotion

words, degree adverbs and negative words is

comprehensively considered, according to the Wang

（2023）’s formula:

Affective value

=

(

positive affective word weight

− negative affective word weight

)

× degree adverb weight)

(1)

Effectively quantifies the emotional tendency and

intensity of the comments.

In addition, to facilitate the calculation of the

emotion-weighted score, the max-minimum

normalization is used to map the emotive value to the

interval [0,1] to eliminate the impact of dimension:



X−x





−x



(

)

Then, by multiplying the mapped star rating and

the normalized emotion intensity, a composite index

"emotion weighted score" is generated to reflect both

the objective value of the user's rating and the

subjective intensity of the emotion tendency, so as to

analyze whether the emotion of the comments can

promote the herd effect (Yao et al.,2017). For

example, a review with a five-star rating 5 combined

with strong positive emotion 2.0 has a weighted score

of 9.

In time series analysis, the dynamic average score

of a movie is an important index to measure the trend

of movie rating over time. However, the movie

history score data of Douban film platform is not

open. In order to analyse the trend of the score over

time, all data is calculated in "days". The cumulative

average of all star ratings and emotion-weighted

ratings from the date of the film's earliest appearance

(i.e. the film's release date) is calculated separately

from the date of release to the date of release, which

can simulate the dynamic change of movie ratings

over time during the film's release.

According to the cumulative average score sum of

each day, the variance of the corresponding day can

be calculated, which can reflect the convergence

degree of the audience for the movie score and the

change of the convergence degree after emotional

weighting.

However, the time series fluctuation caused by the

number of film reviews with heat attenuation leads to

sharp fluctuations in the variance in the later period.

The study proposes an improved smoothing strategy

based on linear weighted moving average (WMA)

(Box et al.,2015). By introducing the standardized

index mechanism of review participation, the model

can keep the sensitivity of recent data while

effectively reducing the variance oscillation caused

by the decrease of the number of comments in the

later period.

Specifically, first build a standardized metric

based on comment engagement:

Herding Behavior-Driven Comedy Film Rating Convergence: A Study of the Moderating Effects of Emotional Intensity

209

𝑁𝑜𝑟𝑚𝐶𝑜𝑚



𝐶



max

∈

(

𝐶



)

(

)

Among them:

𝐶



represents the number of comments in the i th

time unit (day)

Ω is the collection of time units covered by the

sliding window

The numerator and denominator standardization

constrained the index range to [0,1], reflecting the

relative intensity of comment activity on that day

A standardized metric of comment engagement is

then used as a standardized weight calculation:

Smoothing rating/

emotion weighted rating variance =

∑

（ ×



)





∑

(



)





(4)

In the study, W=5 is selected as the window size

The smoothed score/emotional-weighted score

variance is obtained, which ultimately reduces the

violent vibration caused by late fluctuations, and can

better capture the stability of scoring trends and

changes in emotional consensus.

5 REGRESSION ANALYSIS

To examine the temporal variation in rating

dispersion and emotional intensity, an exponential

regression model is employed to analyze herding

effects in rating convergence and the amplification

effect of sentiment-driven herding. Regression model

𝑦

(

𝑡

)

=𝑎⋅𝑒



+𝑐 takes time t as the independent

variable and score variance or emotion weighted

score variance as the dependent variable, where:

1. a is the magnitude of the initial variance,

2. b represents the rate of decay,

3. c represents the level of variance that is

stable over time.

The regression coefficient a represents the initial

variance of the movie score. The sign and magnitude

of regression coefficient b reflect the direction and

rate of volatility change. If the regression coefficient

is greater than zero, it indicates that the variance has

a decreasing trend over time, that is, the degree of

dispersion of scores or emotions has decreased, which

can be regarded as the embodiment of herding effect.

Furthermore, by comparing the regression

coefficients of smoothing score variance and affective

weighted score variance, we can further judge

whether affective factors aggravate the convergence

phenomenon. Among them, when processing the

regression coefficient of emotion weighted score

variance, the previous emotion weighted score was

calculated by multiplying star rating and normalized

emotion intensity, which may lead to compression of

regression coefficient. Therefore, before processing

emotion weighted score variance, divide each sample

point by the normalized average emotion intensity of

the movie. The effect of the compression of

regression coefficient caused by multiplication on the

measurement of herding effect should be minimized.

Next, the regression results will be analysed

horizontally from three aspects: low score, middle

score and high grade.

5.1 Low-Grade Films

Table 5.1: Regression results of low-grade films.

Movie Title

Rating Variance

Regression

Emotional

Rating Variance

Regression

Life on the

Road

y=1.1807e^(-

0.3603x) +2.2916,

R²=0.7189

y=2.5716e^(-

0.2421x)

+1.8081,

R²=0.7935

Ex-lover 4:

Early Marriage

y=0.2832e^(-

0.0144x) +2.4910,

R²=0.7910

y=0.7366e^(-

0.0149x)

+2.3812,

R²=0.7440

Detective

Chinatown 3

y=0.0609e^(0.0698

x) +1.9713,

R²=0.7156

y=-

0.1444e^(3.1805x

) +1.8864,

R²=0.6962

The

Extraordinary

Journey of the

Mozart from

Outer Space

y=0.6940e^(-

0.5473x) +1.8281,

R²=0.9192

y=0.5514e^(-

0.3082x)

+1.6098,

R²=0.7748

Super Family

y=0.0519e^(2.9342

x) +1.6344,

R²=0.0246

y=0.0686e^(0.00

58x) +1.3023,

R²=0.1855

ICEML 2025 - International Conference on E-commerce and Modern Logistics

210

The regression results of low-grade films show that

their score variance has the characteristics of high

initial fluctuation and rapid decay. Taking Life on the

Road as an example, the initial value of the score

variance a+c=1.1807+2.2916=3.4723, and the decay

rate b=0.3603, indicating that controversial content

such as plot logic holes caused sharp differences in

audience evaluation in the early stage. However, with

the passage of time, the score rapidly converges to the

stable value. The phenomenon may be due to the

"dispute-driven discussion" mechanism of low-

scoring films: Extremely bad reviews attract more

viewers to participate in the review, but as the

discussion progresses, some viewers re-examine the

work, resulting in a rapid decline in the variance. For

example, the initial value of The Extraordinary

Journey of the Mozart from Outer Space is 2.5221,

and the decay rate b=0.5473, which further verifies

the dynamic pattern of "high opening and low

walking" in low-score films.

However, the low-grade films' decay rate of

affective score variance is generally lower than that of

ordinary score. For example, the decay rate of the

emotion score of Life on the Road is lower than that

of the ordinary score, and the asymptotic value c is

lower. The suggests that although affective weighting

failed to accelerate convergence, it stabilized the

score at a more concentrated range by filtering out

extreme emotional noise. The result is consistent with

the theory of emotion regulation in psychology.

Sentiment analysis indirectly inhibits the long-term

proliferation of controversial comments by

identifying negative emotional intensity.

5.2 Mid-Grade Films

Table 2: Regression results of mid-grade films.

Movie Title

Rating Variance

Regression

Emotional

Rating Variance

Regression

Grabbing

the Doll

y=0.1709e^(0.094

4x) +2.9905,

R²=0.5988

y=1.3646e^(0.131

6x) +2.6414,

R²=0.8213

Manjiangho

y=0.8444e^(-

0.0745x) +2.9220,

R²=0.8374

y=1.3284e^(-

0.2077x) +2.5380,

R²=0.9558

Hot and

Spicy

y=0.5118e^(-

0.0520x) +3.2106,

R²=0.9364

y=0.9267e^(-

0.0446x) +3.3420,

R²=0.9285

Article 20

y=0.2223e^(0.402

1x) +3.3506,

R²=0.7082

y=0.6803e^(0.182

2x) +3.0684,

R²=0.6676

Flying High

y=0.2471e^(0.236

2x) +3.4188,

R²=0.3440

y=1.8370e^(0.200

9x) +2.9015,

R²=0.9400

The rating variance of mid-grade films shows

significant decay rate differentiation. Taking "Hot and

Hot" as an example, the decay rate of ordinary score

variance b=0.0520, asymptotic value c=3.2106, while

the decay rate of emotional score variance b=0.0446,

asymptotic value c=3.3420, and R

increased from

0.6864 to 0.9285. The indicates that although the

emotional intensity does not accelerate the decline, by

directing the audience to focus on the core theme of

female growth, the impact of minor disputes on the

rating is reduced, and the explanatory power of the

model is significantly improved.

In contrast, in Article 20, the decay rate of

ordinary score variance b=0.4021, asymptotic value

c=3.3506, decay rate of affective score variance

b=0.1822, asymptotic value c=3.0684, and R

decreased from 0.7082 to 0.6676. The shows that

sentiment analysis fails to reconcile the group

antagonism caused by legal disputes, reflecting the

failure of the simplified assumption of the exponential

model for complex social issues.

It is worth noting that the variance decay rate of

emotion score b=0.2077 is significantly faster than

that of ordinary score (b=0.0745), the asymptotic

value c drops from 2.9220 to 2.5380, and R

jumps

from 0.8374 to 0.9558. The result may be due to the

audience's collective memory resonance of historical

themes: emotion weighting accelerates the

convergence of opinions by capturing positive

emotions such as "family feelings" and "heroism".

The phenomenon is called emotional mobilization

effect in communication studies, that is, specific

emotional labels can strengthen group identification

and suppress differences (Cui Junli et al, 2024).

5.3 High-Grade Films

Table 3: Regression results of high-grade films

Movie

Title

Rating Variance

Regression

Emotional Rating

Variance

Regression

Herding Behavior-Driven Comedy Film Rating Convergence: A Study of the Moderating Effects of Emotional Intensity

211

Coco

y=1.2127e^(-

0.0452x) +1.5340,

R²=0.8901

y=0.7074e^(0.4166

x) +1.3301,

R²=0.0436

Dying

Survive

y=0.5601e^(0.1576

x) +1.6231,

R²=0.7002

y=2.0864e^(0.2195

x) +1.5855,

R²=0.5470

Zootopi

y=1.0759e^(-

0.0727x) +1.7293,

R²=0.7554

y=1.6746e^(-

0.0480x) +1.3786,

R²=0.5968

The score variance of high grade films is

characterized by low initial fluctuation and slow

decay. Taking Zootopia as an example, the initial

value of common score variance is 2.8052, decay rate

b=0.0727, asymptotic value c=1.7293, R

=0.7554;

The initial value of emotion score variance

a+c=1.6746+1.3786= 3.0532, decay rate b=0.0480,

asymptotic value c=1.3786, R

=0.5968. The indicates

that the audience's initial evaluation of the classic

works is highly converging, and the emotional

intensity fails to significantly change its dynamics.

The quantization bias of affective weight reduces the

explanatory power of the model.

However, the variance regression of emotion score in

Coco is not significant at all, which may be due to its

artistic expression of the theme of "life and death

cycle", which exceeds the quantitative range of

traditional emotion dictionaries, leading to the failure

of the model. In addition, the initial variance value of

emotion score 3.6719 of Failing to Survive is

significantly higher than the initial value of common

score 2.1832, reflecting the infiltration of external

discussion of social issues into the score, forming a

hypercinematic evaluation dynamic (Lin Hongtong,

2020).

6 CONCLUSION

The study employs an exponential regression model

to uncover the intricate mechanisms governing the

dynamic rating behavior of comedy films on Douban.

The findings indicate that audience rating

convergence is influenced not only by the passage of

time but also by factors such as film quality

classification, emotional intensity of reviews, and the

depth of audience engagement. In the case of low-

rated films, the presence of controversial content

often results in pronounced fluctuations in initial

ratings. While sentiment analysis helps mitigate the

short-term impact of extreme reviews, it remains

insufficient in reversing the non-linear trend of word-

of-mouth decline over time. Although emotional

resonance can accelerate rating convergence for

certain films, it fails to alleviate group polarization

caused by controversial themes or social issues.

Conversely, for high-rated films, the initial

evaluations tend to be highly consistent, leading to

greater rating stability. However, the quantitative bias

introduced by sentiment-weighted scores limits the

explanatory power of deep emotional interactions in

the cases. The innovative contribution of the study is

as follows: First, it focuses on the specific type of

comedy film, and deeply discusses the herding

mechanism of user rating, which fills the gap of

existing research in the field; Second, it not only

focuses on the convergence of user ratings, but also

comprehensively considers the impact of the

emotional strength of reviews on the herd effect, as

well as the differences of the effect in films with

different ratings, providing a more detailed and

comprehensive perspective for the marketing strategy

of the film industry.

Despite the contributions, the study has several

limitations. First, sentiment analysis algorithms still

have room for improvement. Due to the implicit

nature and ambiguity of Chinese language

expressions, reliance on static sentiment dictionaries

alone may fail to fully capture context-dependent and

implicit emotions in reviews. Future research could

enhance accuracy by incorporating irony detection

and contextual semantic analysis techniques. Second,

as the dataset is sourced exclusively from Douban,

audience feedback from other social media platforms

is not considered, which may limit the

generalizability of the findings. Future studies could

validate and extend the conclusions by integrating

data from multiple online platforms. Based on the

findings and deficiencies, the following coping

strategies are suggested for movie scoring platforms

and producers: Movie scoring platforms should

continue to optimize sentiment analysis algorithms

and introduce irony detection and contextual semantic

analysis technologies to more accurately capture

users' emotional tendencies. For example, natural

language processing technology is used to analyse

comments deeply and identify complex emotional

expressions such as irony and metaphor, so as to

improve the accuracy and reliability of emotion

analysis. At the same time, the platform needs to build

a real-time scoring monitoring mechanism to

dynamically monitor movies of different scoring

grades and timely warn of abnormal fluctuations.

Through data analysis and machine learning

ICEML 2025 - International Conference on E-commerce and Modern Logistics

212

algorithms, the system can automatically identify

abnormal patterns in scores to provide decision

support for platform managers.

Producers need to develop differentiated

strategies for different grades based on film ratings

and critical data. Low-grade films should respond

quickly to the controversial evaluation and adjust the

marketing strategy or film content in time to recover

the reputation; Mid-range films can strengthen

emotional labels, guide audiences to form a

consensus, and improve the stability of word-of-

mouth; High grade films should make use of in-depth

reviews and fan activities to maintain and consolidate

their good reputation. In addition, producers can work

with scoring platforms to obtain sentiment analysis

data from audience reviews to gain insight into

audience emotional needs and preferences, so that

targeted optimization can be made during film

production. For example, according to the analysis of

the audience's emotional tendency towards a comedy

film, the plot setting, actor performance and visual

effects of the film are adjusted to better meet the

audience's expectations. By implementing the

strategies, film rating platforms and producers can

better navigate the challenges posed by the herding

effect, enhance the reliability of film rating systems,

refine industry marketing strategies, and ultimately

deliver higher-quality content and services to

audiences.

REFERENCES

Bond, R., Smith, P., 1996. Culture and conformity: A meta-

analysis of studies using Asch's (1952b, 1956) line

judgment task. Psychological Bulletin, 119(1), 111-137.

Box, G., Jenkins, G., Reinsel, G., Ljung, G., 2015. Time

Series Analysis: Forecasting and Control (5th ed.).

Wiley.

Chevalier, J., Mayzlin, D., 2006. The effect of word of

mouth on sales: Online book reviews. Journal of

Marketing Research, 43(3), 345-354.

Cui, J., Jiang, H., 2024. Emotional Mobilization and

Resistance Practices in Hot Events on Douyin

Platform—A Case Study of the Phenomenon of Public

Copywriting Relay Communication. Progress in

psychology, 14 (9), 508-515.

https://doi.org/10.12677/ap.2024.149675

Gupta, S., et al., 2024. How consumers evaluate movies on

online platforms? Investigating the role of consumer

engagement and external engagement. Journal of

Business Research, 176, 114613.

https://doi.org/10.1016/j.jbusres.2024.114613

Hao, Y., Zou, P., Li, Y., et al., 2009. Empirical study on the

impact of online review sentiment on sales revenue

based on movie panel data. Management Review,

21(10), 120-128.

Hao, X., Chen, X., 2019. The Herding Effect of

Consumption Decision of Experience Products and

Mechanism Explanation -An Empirical Interpretation

of Consumer Behavior in Film Industry[J]. Chinese

Journal of Management Science, 2019, 27(11): 176-

188. 10.16381/j.cnki.issn1003-207x.2019.11.018

Lee, Y. J., Hosanagar, K., Tan, Y., 2015. Do I Follow My

Friends or the Crowd? Information Cascades in Online

Movie Ratings. Management Science, 61(9), 2241-

2258. https://doi.org/10.1287/mnsc.2014.2082

Lin, H., 2020. Film Narrative Techniques and Methods:

Analysis of Classic Films. China Film Press.

Liu, Y., 2018. Analysis of the Current Situation and Public

Opinion of Film Rating Communities at Home and

Abroad[J]. Audio-visual, 2018(1): 34-35.

Liu, W., Ma, Z, Wei, X., 2019. A meta-analysis of the

impact of crowding on consumer emotions and

shopping reactions. Acta Psychologica Sinica, 52(1),

16-28.

Pang, B., Lee, L., 2008. Opinion mining and sentiment

analysis. Foundations and Trends in Information

Retrieval, 2(1-2), 1-135.

Qu, W., et al., 2024. A study on the influencing factors of

movie consumer satisfaction based on text mining. Sino-

American Arts and Sciences, 13(4), 115.

https://doi.org/10.12677/sa.2024.134115.

Salganik, M., Dodds, P., Watts, D., 2006. Experimental

study of inequality and unpredictability in an artificial

cultural market. Science, 311(5762), 854-856.

Shi T., 2024. Analysis of LDA Theme Model for Douban

Movie Reviews Based on Text Mining —Taking the

Movie “Let the Bullets Fly” as an Example. Journalism

and Communications Journalism and communication

science, 2024, 12(1), 23-28.

https://doi.org/10.12677/JC.2024.121004

Świerczyńska-Kaczor, U., 2019. Exploring the relationship

between viewer experience and movie genre – A study

based on text mining of online movie reviews. Problemy

Zarządzania – Management Issues, 17(5), 85-9.

https://doi.org/10.7172/1644-9584.85.9

Taboada, M., Brooke, J., Tofiloski, M., Ide, N., Szatrowski,

T., 2011. Lexicon-based methods for sentiment

analysis. Computational Linguistics, 37(2), 267-307.

Wang, T., Zhang, Z., 2022. Research on the construction

method of sentiment dictionary for movie reviews.

Computer and Digital Engineering, 50(4), 843-848.

Wang, w., Wu, j., 2023. Sentiment Analysis of microblog

text based on Sentiment Dictionary and Semantic Rule

Set [J]. Computer Science and Applications, 2023,

13(4): 754-763. DOI: 10.12677/csa.2023.134074

Yao, Y., 2017. Empirical study on sentiment analysis of

online reviews based on user rating preferences.

(Dissertation). Shanghai: Tongji University.

Zhang, Z., Qiang, C., Duan, S., 2021. A review of

personalized movie recommendation algorithms.

Computer Knowledge and Technology, 17(22), 1-10

Herding Behavior-Driven Comedy Film Rating Convergence: A Study of the Moderating Effects of Emotional Intensity

213