Analysis of the Polarization of Sentiment in Chinese Film Reviews

and Related Factors

Haoyuan Qin

Faculty of Engineering and Information Technology, The University of Melbourne, Victoria, 3010, Australia

Keywords: Sentiment Polarization, Film Reviews, Natural Language Processing, Box Office Analysis.

Abstract: At a time when film production is increasingly targeting specific audiences, the phenomenon that some

Chinese film reviews have more emotional outputs between different audiences rather than discussions about

the content of the film has attracted widespread attention, but there is still a lack of specific qualitative and

analytical indicators for emotional polarization in reviews. This paper analyzes the public data of Douban

film reviews of 28 film samples, divides the emotional tendencies of their film reviews into three categories

to analyze whether they have emotional polarization, and studies the correlation between polarization and

other factors through quantitative analysis. This paper divides the types of film review polarization into three

categories and analyzes that on the Chinese film review platform, the film actor factor is more likely to cause

emotional polarization in film reviews than other factors and needs to be paid more attention to when taking

advantage of consumers' emotional consumption behavior. Based on this, this paper makes the following

suggestions: In the future, the film review emotional polarization index can be used as a new indicator for

analyzing the controversy and content effectiveness of film content, and more film review data can be

analyzed.

1 INTRODUCTION

In the context of the mature film industry today, the

audience's evaluation of movies has formed a huge

amount of film review data on the Internet. With the

diversification of entertainment forms, the audience's

aesthetic standards for film and television works have

also changed. The audience's evaluation and the

discussion in the comment area have become

important indicators for some users to choose movies.

The light-hearted and funny "popcorn movies" are

still hot-selling, but as an important indicator for

evaluating movies, the movie's comment area

sometimes gives negative opinions. Although some

comedy movies have achieved success at the box

office, there are controversies about cultural criticism

in the film review area, especially whether their jokes

are vulgar or disrespectful, such as "making fun of the

poor". Even in the comment area, some comments are

not directed at the content of the movie but simply

output emotions. On the other hand, although the

quality of movie content varies, many works have

been labeled as "vulgar" because of their overly

deliberate, dramatic, and low-quality movie plots, and

the film review emotions are polarized or one-sidedly

positive/negative comments, but they also have a

strong appeal because of their fast pace and easy to

fascinate people. Therefore, if it can find out the

factors related to the "polarization" and "one-

sidedness" of film reviews, and the factors that make

the reviews as a whole tend to output emotions rather

than content, that is, the emotional polarization of

film reviews, it may have a positive impact on the

review of film production and the early publicity and

promotion. In addition, since the emotional tendency

is generally more extreme when the reviews are

emotional outputs lacking content, studying the

emotional polarization of film reviews may also

provide more quantitative indicators for the analysis

of the controversial degree of film content.

At present, with the rapid development of

machine learning and natural language processing,

many practical models have been put into use. As an

important branch of natural language processing,

sentiment analysis aims to understand and judge the

emotional color contained in the text through

computer algorithms and models, whether it is

positive, negative, or neutral. Nowadays, many

powerful and practical models have emerged, and

some of them have even successfully entered the field

of commercial applications. By using these natural

language models, combined with methods such as the

Qin, H.

Analysis of the Polarization of Sentiment in Chinese Film Reviews and Related Factors.

DOI: 10.5220/0014151400004942

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

In Proceedings of the 2nd International Conference on Applied Psychology and Marketing Management (APMM 2025), pages 703-709

ISBN: 978-989-758-791-7

703

sentiment dictionary of the film industry, the

emotional tendency of film reviews can be obtained

more accurately, and some emotional tendency

patterns with statistical validity can be obtained by

analyzing a large number of comments in the film

review area. Specifically, when a film review is input,

the model can quickly analyze the vocabulary,

sentence structure, and context, thereby determining

the emotional attitude expressed by the author when

writing the review, whether it is praise, criticism, or

neutral evaluation of the film (Ullah et al., 2023). In

addition, due to the openness of review data, a large

number of public user reviews can be easily obtained

from domestic film review platforms. The openness

and accessibility of this data provide conditions for an

in-depth understanding of user feedback and

emotional dynamics in the film market and also

provide a strong reference for the development and

optimization of the film industry.

In recent years, with the development of natural

language processing technology, the research on

combining film reviews with sentiment analysis

mainly includes using machine learning methods to

try to extract effective content from the words and

sentences in emotionally extreme film reviews and

improve them. Lee, Sang Hoon, and others use

sentiment dictionaries to assist in analyzing the

sentiment of film reviews and effectively improve the

accuracy of models in the field of film review analysis

(Lee et al., 2016). Because there is a correlation

between the emotionality of film reviews and the

effectiveness of the content, sentiment analysis can be

used to classify film reviews to find more high-

quality film reviews that are worth referring to.

Sudhanshu Kumar and others combine sentiment

analysis with traditional recommendation systems to

filter and classify emotional reviews through

sentiment analysis to recommend reviews and content

that are closer to the user's emotional tendencies

(Kumar et al., 2020; Soubraylu & Rajalakshmi,

2021). In terms of specific analysis methods, they

include using hybrid models, pre-screening and

classification based on text length, and building

extreme word recognition dictionaries to perform

more accurate sentiment analysis, etc., to improve the

accuracy of analysis (Peng & Cheng, 2023). At

present, there is relatively mature analysis software in

the commercial market that can monitor the

generation and decline of hot topics and even judge

social media group polarization, which act as

"accelerators" to jointly promote the generation of

group polarization (Rabiee et al., 2024; Utmhikari,

2017). However, there is still a lack of research from

a specific perspective on the correlation between

audience emotional polarization and film-related

factors.

The core issue of this study is what factors of the

film are mainly related to the emergence of the

emotional polarization phenomenon in the audience's

evaluation in the comment area. Unlike existing film

review sentiment analysis research, this study focuses

on the audience emotional polarization phenomenon

itself. This phenomenon is of great value because if it

can be determined that the type of film or specific

factors are correlated with the audience's emotional

polarization or even causal relationship, it will have a

positive effect on the filmmaker's marketing strategy

formulation, helping them to accurately locate

potential users and improve the success rate of

marketing activities. Specifically, this study aims to

define and classify the emergence of audience

emotional polarization and analyze whether there is a

correlation between the type of film, box office,

ratings, and audience emotional polarization to

provide possible new indicators for the analysis of

film reviews.

2 METHOD

The data for this study comes from the public film

review data uploaded to Kaggle in 2017, which is

sourced from the Douban platform (Yuan et al.,

2020). The data mainly includes the Chinese name,

English name, review content, rating, number of

likes, and other information about the movie. The

earliest movie in the data is Iron Man 1 in 2008, and

the rest of the movies are from 2012 and later. This

data collection method avoids the copyright and

technical difficulties that may be faced by self-

collected data and can quickly obtain a large amount

of real and public film review data.

2.1 Sentiment Tendency Analysis

This study aims to explore whether the type of movie

affects the audience's emotional polarization. The

analysis model selected is the Baidu voice sentiment

tendency analysis model, which has been put into

commercial use and is relatively stable. From the 28

movies in the database, 2000 reviews of each movie

were randomly selected for analysis. Before data

analysis, it first cleaned the comment confidence

(confidence>0.5), then performed sentiment analysis

on each comment, and finally combined statistical

methods to analyze the sentiment distribution and its

possible polarization phenomenon, and obtained the

correlation analysis between the audience's sentiment

APMM 2025 - International Conference on Applied Psychology and Marketing Management

704

polarization phenomenon and the box office, ratings,

and movie types. Specifically, after separating the

extremes, if most of the extreme values are positive,

it is considered to be a positive extreme; otherwise, it

is a negative extreme. However, when the positive

and negative extreme values are relatively balanced,

they need to be discussed in categories. For example,

comments with a positive sentiment bias greater than

0.99 are called positive extreme comments, and vice

versa. After assuming that comments with a positive

sentiment bias greater than 0.99 or less than 0.01 are

extreme values, the number of extreme comments of

each movie is counted and sorted, and it can be

concluded that among the 28 movies, "Nine-story

Demon Tower" and "Why the Flute is Silent"

accounted for more than 66% of negative emotions,

showing a typical negative emotional polarization

phenomenon.

By identifying the opinion keywords and

emotional words in the comments, it is judged

whether there is a conflict of opinion between the

comments. If there are a large number of comments

that refute and argue with each other, and the

sentiment tendencies of these comments are opposite

(one is positive, the other is negative), it means that

there may be emotional polarization in the comment

area. For example, in the comment area about a

certain movie, some users praised the actors for their

excellent performance (positive), while other users

strongly criticized the actors' acting skills (negative),

and the two sides argued fiercely, which may be a

manifestation of emotional polarization. At the same

time, by counting the number and proportion of

controversial comments, the phenomenon of

comment emotional polarization can also be

identified. For example, use Baidu's sentiment

tendency analysis interface to calculate the proportion

of extreme values (prob>0.99/non-middle 50%) and

use distribution statistics indicators to count the

proportion of different sentiment tendencies. If the

proportion of controversial comments gradually

increases, it means that the conflict of opinions in the

comment area is intensifying, which may cause

emotional polarization. Even if the emotional

differences between groups are calculated in real-

time, if the difference in emotional tendencies

between different groups increases, it means that the

emotional differentiation of commentators is

intensifying. Such a possible phenomenon of

emotional polarization can also be predicted in

advance and may have an impact on real-time

marketing strategies (Schwertman et al., 2004).

2.2 Correlation Analysis

Correlation is a statistical indicator used to measure

the strength and direction of the linear relationship

between two or more variables. When analyzing

movie parameters, correlation can help analyze and

understand the degree of correlation between

different parameters (such as movie ratings, box

office, number of moviegoers, etc.). Commonly used

correlation indicators include the Pearson correlation

coefficient, which measures the degree of linear

correlation between two variables, and its value range

is between -1 and 1. A value of 1 indicates a complete

positive correlation, a value of -1 indicates a complete

negative correlation and a value of 0 indicates that

there is no linear correlation. This article uses the

Pearson correlation coefficient to perform correlation

analysis after one-hot encoding nonlinear parameters

such as movie types and actors.

2.3 Word Frequency Analysis of Film

Reviews

In the process of film review sentiment polarization

analysis, the frequency statistics of text data is an

important means to help identify the most

representative keywords in the comments and reveal

the core topics discussed by the audience under

different movie types or emotional tendencies.

To ensure the accuracy of the analysis, this study

first preprocessed the review text, including removing

special characters and punctuation marks and using

the stop word list of Harbin Institute of Technology

to remove stop words from the review text to reduce

the interference of invalid information. After the text

is cleaned, this study uses word frequency statistics,

TF-IDF (word frequency-inverse document

frequency), and other methods to analyze the

distribution of high-frequency words in each movie

review and combines word cloud, table, and other

visualization methods to display the distribution of

keywords, to intuitively present the key content

discussed in the comment area. For example, for

movies with high box office but serious emotional

polarization of film reviews, it is possible to observe

whether high-frequency keywords are concentrated

on specific topics (such as actors, plots, special

effects, etc.) and further analyze whether these

keywords are related to the emotional tendency of

film reviews.

Analysis of the Polarization of Sentiment in Chinese Film Reviews and Related Factors

705

3 RESULT

3.1 Movie Genres and Emotional

Tendencies

As shown in Fig. 1, in the study of movie genres and

audience emotional tendencies, it was found that the

average emotional tendencies of audiences for

musicals, animations, and science fiction films

showed a significant positive trend. In contrast, youth

and romance films showed a different trend, with

audience emotions tending to be more negative, and

in the evaluation of such films, the phenomenon of

emotional polarization was particularly prominent,

that is, the audience's evaluation was more likely to

be divided into two completely different poles, and

the difference and opposition between positive and

negative evaluations were more significant.

Figure. 1. Average emotional tendency of movies of

different genres.

3.2 Analysis of Three Types of

Sentiment Polarization

Since the quartiles can clearly and directly display the

five important features of the data: the extremes, the

upper and lower hinges (quartiles), and the median

therefore using the quartiles to display the sentiment

tendency score of each movie can directly determine

the proportion of extreme values through the width of

the upper and lower hinges (Guo, 2020). Based on the

in-depth exploration and analysis of the sentiment

analysis quartiles of 28 movies, it can be seen that the

distribution area occupied by some movies in the

quartiles is relatively wide, that is, the box area is

relatively large. This feature means that in the

audience's evaluation system of these movies, the

distribution of positive and negative emotions tends

to be balanced, and its sentiment tendency presents a

diversified pattern, reflecting a large emotional

discreteness. On the contrary, the distribution area of

some other movies in the quartiles is highly

concentrated, and the range is relatively narrow. The

evaluation of these movies focuses more on the

positive emotion dimension, and the proportion of

negative emotions is relatively small. From the

perspective of the theory of emotional polarization,

this phenomenon may imply that such movies are less

likely to cause negative reviews, and the audience

group maintains a relatively consistent positive

emotional attitude towards the movie as a whole,

which in turn causes its distribution in the quadrant to

show a relatively concentrated trend, which is in

sharp contrast to those movies with dispersed

emotional tendencies.

To analyze the three different types of emotional

polarization tendencies more clearly, this study chose

to observe the emotional analysis quadrants of three

specific movies in detail (Fig. 2). It can be seen that

the box of Chronicles of the Ghostly Tribe is

generally lower and the median is almost zero, which

means that the film review sentiment of this movie

tends to be negative, and the most frequent words in

its film reviews are discussions about the original

work, actors, and directors. The box of The Mermaid

is relatively large, and the median is relatively

centered. An in-depth analysis of the main reasons

behind it may be because the film focuses on the

construction and portrayal of the relationship between

actors at the core theme level of narrative and

discussion. The word frequency analysis of the film

reviews shows that the director's name, the leading

actor's name, environmental protection, plot, and

other representative keywords appear. The existence

of these elements may, to a certain extent, reflect the

wide distribution of audience evaluations in the

emotional dimension, which increases the degree of

dispersion of evaluations and makes the emotional

tendency more diversified and decentralized. The box

of Zootopia presents a smaller scale. This

phenomenon intuitively reflects that its evaluation

distribution has a highly concentrated characteristic

and mainly tends to be in the positive evaluation

category. Its degree of emotional polarization is more

concentrated than that of the other two films. The

audience's evaluation of the film is highly consistent,

and the proportion of negative evaluations is

extremely low. The words with a high evaluation rate

in their comments include animals, plots, and stories,

that is, discussions on the content of the movie.

20%

40%

60%

80%

100%

Youth

Drama

Romance

Adventure

Action

Fantasy

Animation

Comedy

Science…

Crime

Musical

Horror

Disaster

Number of negative reviews

Number of positive reviews

APMM 2025 - International Conference on Applied Psychology and Marketing Management

706

Figure. 2. Quartile chart of sentiment analysis of film reviews for the three films "Chronicles of the Ghostly Tribe", "The

Mermaid", and "Zootopia".

This study uses the proportion of positive

emotions in extreme emotional reviews greater than

66% and less than 33% as the classification of

whether positive polarization and negative

polarization occur. When the proportion of positive

and negative extreme emotional reviews is similar, it

is necessary to analyze specific movies. Generally,

the proportion of noun entities in film reviews of

films with emotional polarization will show obvious

repetition and reduction. Among the 28 movies

analyzed in this study, 9 movies have positive

emotional polarization, 2 movies have negative

emotional polarization, and the remaining 17 movies

have similar proportions of positive and negative

extreme emotional tendency reviews.

At the same time, combined with the movie

ratings, 29.09, 49.65, and 85.98 correspond to the

positive emotional tendency ratios of 28.9, 49.6, and

86.4 for the three movies "Nine-story Demon Tower",

"The Mermaid", and "Zootopia", respectively. It can

also be judged that the emotional tendency of film

reviews and movie ratings have a high correlation.

3.3 Correlation Between Audience

Needs and Emotional Polarity

The data analysis shows that the emotional

polarization of the audience is weakly positively

correlated with the box office. The viewing needs of

Chinese audiences are mainly divided into seven

categories: story, role, director/team, special effects,

art, actor, and action (Li et al., 2022). Among all the

movie data, the two movies with the most obvious

positive and negative polarization are Zootopia and

My Sunshine. After removing stop words using the

stop word list of Harbin Institute of Technology, the

word cloud diagram is generated using WorldCloud.

In My Sunshine, which has the most extreme negative

emotions, the most frequent categories are actor and

story, including discussions about the actor's acting

skills and comparisons with the plots of the TV series

of the same name. In Zootopia, which has the most

extreme positive emotions, the most frequent

categories are character and story. Through a simple

comparison of the two, the story of the movie is a

more important factor affecting the audience's

emotional tendency than other factors.

Whether it is a movie with positive or negative

emotional polarization, the story factor has a large

proportion in the discussion. Therefore, it can be

considered that there is a strong correlation between

the plot of a movie and the polarization of film

reviews. When promoting a movie, special attention

should be paid to avoid any points that make the

audience feel abrupt or even disgusted. At the same

time, the keyword review rate in negatively polarized

film reviews is more concentrated than that in

positively polarized film reviews, which may mean

that the outbreak of negative emotional tendencies is

also highly uncertain. Once there is an opinion leader

at the core of the topic, negative emotions are likely

to become polarized (Rabiee et al., 2024). In addition,

compared with positively polarized film reviews,

negatively polarized film reviews have more

discussions about non-film content, such as actors

and directors.

Analysis of the Polarization of Sentiment in Chinese Film Reviews and Related Factors

707

4 DISCUSSION

4.1 Movie Types and Emotional

Tendencies

When comparing different types of movies, this study

found that love and youth movies are more likely to

have audience emotional polarization than other types

of movies and tend to be negative, and negative

comments are more focused on actors rather than

roles than positive comments; the average emotion of

the audience of musical movies tends to be positive.

Combined with the results of word frequency analysis

of film review data, the difference between positive

and negative comments in the comments of love and

youth movies with strong storylines is more

significant, and the audience generally likes elements

such as music and dance.

In terms of the content of the discussion, strong

story movies are more likely to touch the audience's

emotions and memories and are more likely to lead to

polarization of audience evaluations. Therefore,

when targeting such strong story movies, both in the

early marketing and publicity, it is necessary to pay

attention to the emotional discussions that such

movies may bring and carry out targeted publicity,

especially the publicity of actors. Appropriate

publicity may bring higher attention to such movies

more easily, but inappropriate publicity may also

make such movies more likely to receive low ratings.

For musicals and dances, the data of this study

shows that the average emotion of the audience of

musicals and dances tends to be positive, but there are

also obvious problems, such as a single discussion

structure and a small number of comments. It may be

necessary to pay more attention to other aspects

during promotion, especially factors such as plot and

actors that are more relevant to the audience's

emotions, to attract the audience's attention.

4.2 Correlation Between Emotional

Polarization and Film Indicators

The box office of movies with obvious audience

emotional differentiation in the comment area is often

relatively high, which may be because emotional

polarization will cause more discussion and attention

and then drive the box office to a certain extent. From

the word frequency analysis, it can be seen that

negative polarization is generally more targeted at

specific topics than positive polarization. Therefore,

the promotion of movies may be more targeted at

some specific topics that are more likely to cause

discussion, such as actor selection. Although it may

be more likely to cause emotional polarization in the

evaluation, the discussion will also increase

accordingly and may further improve the box office

performance of the movie.

5 CONCLUSION

This study explored the characteristics of audience

emotional polarization and its related influencing

factors by analyzing Chinese film reviews of 28

different types of films. The study showed that there

was a certain correlation between the phenomenon of

emotional polarization in film reviews and indicators

such as the type, box office, and ratings of the film.

Specifically, youth and romance films are more likely

to cause emotional polarization, while science fiction

and musical films tend to have positive emotional

evaluations. This finding suggests that the type of

film plays an important role in the phenomenon of

emotional polarization.

In addition, the study also found that the

phenomenon of emotional polarization in film

reviews is closely related to the performance of film

actors. Negatively emotionally polarized film reviews

often focus on discussions about the actors' acting

skills or their past reputations, while positive

emotional polarization tends to focus on the

characters or core plots of the film. Therefore, film

production and publicity teams should focus on actor

selection and plot design during the planning stage to

reduce the risk of negative evaluations and enhance

the audience's overall viewing experience.

Although this study conducted a detailed analysis

of the phenomenon of emotional polarization in film

reviews through natural language processing and

statistical methods, the universality of the research

conclusions still needs further study due to the

limitations of sample size and film types. Future

research can further explore the causes and potential

impacts of the phenomenon of emotional polarization

in film reviews by expanding the sample range,

combining more types of film data, and introducing

more accurate sentiment analysis models.

In short, the study of the phenomenon of

emotional polarization in film reviews not only

provides a new perspective on the emotional response

of the audience for the film industry, but also provides

a useful reference for research in related fields. In the

future, with the continuous development of sentiment

analysis technology, this research direction will have

a wider application value.

APMM 2025 - International Conference on Applied Psychology and Marketing Management

708

REFERENCES

A. Ullah, S. N. Khan, N. M. Nawi, Review on sentiment

analysis for text classification techniques from 2010 to

2021. Multimedia Tools Appl. 82(6), 8137-8193 (2023)

G. Peng, X. Cheng, Research on the formation mechanism

of social media group polarization in hot events. J. Inf.

Resour. Manag. 13(2), 42-52 (2023)

H. Rabiee, B. T. Ladani, E. Sahafizadeh, A social network

model for analysis of public opinion formation process.

IEEE Trans. Comput. Soc. Syst. (2024)

J. Yuan, J. Shi, J. Che, C. Xu, J. Wang, Modeling and

simulation analysis of public opinion polarization in a

dynamic network environment. Concurrency Comput.

Pract. Exper. 32(19), e5771 (2020)

L. Guo, Sentiment analysis of domestic film reviews based

on deep learning, Ph.D. thesis, Guangxi Normal

University (2020)

M. Li, Y. Zhang, S. Wang, Comparative study of Chinese

audiences' demand and satisfaction for Chinese and

foreign films based on Word2vec and TF-MONO

algorithm. Intell. Eng. 8(2), 073-086 (2022)

N. C. Schwertman, M. A. Owens, R. Adnan, A simple more

general boxplot method for identifying outliers.

Comput. Stat. Data Anal. 47(1), 165-174 (2004)

S. H. Lee, J. Cui, J. W. Kim, Sentiment analysis on movie

review through building modified sentiment dictionary

by movie genre. J. Intell. Inf. Syst., Korea Intell. Inf.

Syst. Soc. (2016)

S. Kumar, K. De, P. P. Roy, Movie recommendation system

using sentiment analysis from microblogging data.

IEEE Trans. Comput. Soc. Syst. 7(4), 915-923 (2020)

S. Soubraylu, R. Rajalakshmi, Hybrid convolutional

bidirectional recurrent neural network based sentiment

analysis on movie reviews. Comput. Intell. 37(2), 735-

757 (2021)

Utmhikari, Douban movie short comments dataset. Kaggle

(2017). Retrieved from

https://www.kaggle.com/datasets/utmhikari/doubanmo

vieshortcomments/data.

Analysis of the Polarization of Sentiment in Chinese Film Reviews and Related Factors

709