Predicting the Popularity of Movies with Social Media

Liangye He

School of Data Science and Engineering, Guangdong Polytechnic Normal University, Heyuan, Guangdong, China

Keywords: Movie Popularity, Correlation Analysis, Descriptive Statistics, Machine Learning.

Abstract: The film industry occupies an important position in the field of culture and entertainment, and the issue of

how to avoid the risk of investing in films has received widespread attention, but the existing prediction

methods still have certain shortcomings in terms of accuracy and practicality. This paper thoroughly analyses

the various dimensions that influence movie popularity, including movie genre, release time, heat, and other

factors, and explores the relationship between these factors and movie popularity. This paper analyses and

concludes that factors such as the combination of actors and directors, the running time of the film, and the

release time have a significant impact on the popularity of the film and thus establish a prediction model for

the popularity of the film. Based on this, this paper makes the following suggestions: if film producers want

to increase the popularity of their films, they should pay attention to the release time of their films and the

combination of directors and actors. At the same time, industry insiders can use the prediction model in this

paper to make decisions about film projects.

1 INTRODUCTION

With the rapid development of information

technology, modern society has fully entered the

information age. Therefore, in the context of this era,

information has become a key element in promoting

social progress and development in various fields,

and the Internet, as the main carrier of information

dissemination, brings together a huge amount of data

resources and provides sufficient information support

for all walks of life. Social media is an important

product of the Internet era and has become the most

widely used tool in people's daily lives, and the

amount of data contained in it is self-evident.

According to Digital 2024, published by We Are

Social and Meltwater, the number of active social

media user identities has surpassed 5 billion,

representing 62.3% of the world's population (Social,

2024). Last year, the global total increased by 266

million, with an annual growth rate of 5.6 percent.

This data not only demonstrates the widespread

popularity of social media but also shows that social

media has great data potential as a major platform for

information dissemination, social interaction, and

cultural exchange.

Film investors want to reduce the risk of

investment, so if they can use scientific methods to

predict the popularity of the film and adopt a

reasonable marketing strategy, film production

companies and distribution companies can avoid the

risk and reduce losses (Li, 2021). For the audience,

what they want more is to be able to watch a good

film, and rating prediction can help them decide

whether to watch the film or not to enhance the

viewing experience. Douban, a famous film website

in China, was mined by scholars for rating prediction

a few years ago (Geng & Guo, 2021; Zou, 2019).

IMDb and TMDB are also popular film websites that

contain a lot of information about the film and

audience comments, and because of their large

number of users, the scores given on the site can be a

good reference option.

There have been many scholarly studies on film

scoring. Among the more recent findings, Liu et al.

were inspired by brain mechanisms to independently

encode static features as control signals and modulate

time-varying contextual features to predict individual

decisions in complex contexts (Liu et al.). Xu et al.

constructed a movie information network from the

relationship between users, movies, and attributes and

used the metapath2vec algorithm to map various

attributes to the corresponding feature space in the

form of network nodes to obtain feature vectors for

rating prediction (Xu et al., 2024). Liu et al. used an

attention network to distinguish the importance of

interactive features and then used DNN to process

higher-order feature combinations to build a

predictive teacher model while using knowledge

662

He, L.

Predicting the Popularity of Movies with Social Media.

DOI: 10.5220/0014150300004942

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 662-667

ISBN: 978-989-758-791-7

distillation to build a streamlined model of the student

model to ensure the accuracy of the teacher model

(Liu et al., 2024). Previously, Michael T. Lash and

Kang Zhao used social network analysis and text

mining techniques to delve deeper into the film's cast

and release season, ultimately speculating on the

film's profitability (Lash & Zhao, 2016). N.

Darapaneni et al. chose XGBoost as the final film

success prediction algorithm after working with film

outline plots and testing it with several models

(Darapaneni et al., 2020). Differently, Rijul Dhir and

Anand Raj chose released data from the film as their

data set, rather than listening to what critics and

others had to say about whether the film would be a

success or not, but the final model's predictive

accuracy wasn't very high, and they're also looking

ahead to future work where they want to include

analysis of user comments from social media (Dhir &

Rah, 2018). Rather than simply choosing the optimal

model, Vedika Gupta et al. integrated and compared

different models before using them to predict the

success of the film (Gupta et al., 2023).

At the same time, with the help of social media

analysis, the accuracy of the prediction model will

further improve, as a result of which provides more

scientific decision-making support for film investors.

Social media is an important platform for modern

information dissemination, contains a large amount of

user-generated content, including film reviews, topic

discussions, etc., and this data reflects the audience's

real-time feedback on the film, as well as their

emotional inclination, which is of high research value

(Castillo et al., 2021).

Based on this, this paper will examine the

correlation between various factors that influence a

film's popularity and its rating and will develop a

prediction model.

2 METHOD

2.1 Research Design

The purpose of this paper is to explore the potential

value and laws of film-related data. Through the

comprehensive analysis of multi-dimensional data

such as film release time and audience rating, this

study reveals the internal relationship between film

popularity and these factors and their influence

degree, that is, the weight, to establish a film rating

prediction model. Meanwhile, social media analysis

can be used to capture the expectation value of the

audience, i.e., the social media buzz, at this point

before the film's release, thus predicting the film's

rating and popularity.

Due to the complexity of the factors that influence

a film's popularity, it is difficult for a simple

prediction model to produce accurate results (Geng &

Guo, 2021; He & Yuan, 2021). First of all, this paper

obtains the film data set through Kaggle, the reason

for choosing Kaggle is because it is a more

authoritative and data-rich website, where there is no

lack of scholars in various fields to select the data

inside for academic research, and then remove null

values as well as outliers in the data set. To explore

the factors that affect the popularity of films, this

paper will carry out statistical analysis and correlation

analysis on the popularity of films, which is difficult

to express in numerical terms. Therefore, the

audience rating of films is selected as an indicator to

measure the popularity of films, which is used as a

dependent variable for correlation analysis, and the

influencing factors are obtained by combining

statistical analysis. The first step of this paper is to

conduct a descriptive statistical analysis based on the

data results to explore the correlation between

nonnumerical items, namely, actors, directors, actors,

and directors' specific combinations and film scores.

The second step is to conduct a correlation analysis

between numerical items and film scores and use the

analysis results to establish a prediction model of film

popularity (Abidi et al., 2020). Because the ultimate

goal is to predict the popularity of films, the

popularity of films is defined by five levels, namely

terrible, poor, satisfactory, good, and excellent. The

research design is shown in Fig.1.

Figure. 1. Research design.

2.2 Data Collection

To better predict the popularity of movies, a perfect

data set is needed to obtain more accurate results.

There are 880,990 original datasets in the data set

used in this paper, including 27 dimensions, such as

Predicting the Popularity of Movies with Social Media

663

film title, actor, director, film type, score, etc. To

ensure consistency, too short and too long movies are

removed from the data set, and only movies with a

movie duration of 80 to 200 minutes are retained. At

the same time, the reference value of film rating

obtained when there are few rating users is not

significant, so to ensure objectivity, films with rating

users less than 100 should be removed. In addition, to

avoid randomness, factors affecting the fairness of the

research results, such as screenwriters, film travel

companies, original titles, etc., are removed. Because

Python is selected for prediction in this paper, the

prediction of floating-point values may affect the

accuracy, so the score is multiplied by 10 manually.

Moreover, to realize the prediction, the profit should

not appear in the data set because this is the data that

can be obtained only after the film is released.

Through the above steps, the data set with 12,997

pieces of data is finally obtained, and 11 dimensions

are reserved for analysis, including vote_average,

vote_count, release_date, revenue, budget, runtime,

popularity, genres, cast, director, imdb_rating,

imdb_votes。

At the same time, in order to focus on the films

with high and low scores, two different data sets are

used, top_1000 (score greater than 7.5) and

lowest_100 (score less than 4.0), which are mainly

used for the statistical analysis of non numerical

items, so the dimensions only include date, score,

genre, director, crew.

3 RESULT

3.1 Specific Combinations and Film

Score

The combination of specific directors and actors may

form a brand effect, which can attract their loyal fans

to a certain extent, thus affecting the film score. In the

most popular top1000 data set, based on its data

characteristics for statistical analysis, this study will

arrange and combine the directors and leading actors,

and evaluate them by combination and score.

As shown in Table 1, when these directors and

actors cooperate, the average film score is high, so it

can be considered that when the above combination

appears, the film score will be high. At the same time,

the replacement of the combination will also have an

impact on the film score, just like the combination of

Akira Kurosawa, the same director, and cooperating

with different actors, Toshir ô Mifune and Tatsuya

Nakadai. Then, do the same work through the lowest

data set, and the rules are similar. Therefore, if a

director and an actor appear in a film at the same time,

it may cause brand effect and get a higher score. On

the contrary, the "chemical reaction" will lead to a

lower film score. Therefore, it can be considered that

the specific combination of director and actor has an

impact on the film score to a certain extent.

Table 1. The top five director and actor groups and the

average score of their films.

Director Actor

Cooperative

times

Average

film score

Akira

Kurosawa

Toshirô

Mifune

7 8.24

Richard

Linklate

Ethan

Hawke

5 7.98

Akira

Kurosawa

Tatsuya

Nakadai

5 8.10

Peter

Jackson

Ian

McKellen

5 7.80

John Ford

John

Wayne

4 7.80

3.2 Main Factors

In the first section of this chapter, actors and directors

are studied. Therefore, in this section, the relevance

between them and film scoring will not be explored,

but other characteristics will be studied. Through the

exploration of correlation, it is found that several

factors are significantly correlated with the film

score, namely the release time and the playing time,

and the corresponding p-value is less than 0.05,

indicating that they have a significant correlation with

the film score. In addition, there is only a weak

correlation between film popularity and film score.

Table 2. Release month and IMDb score.

The first is to explore the correlation between the

release time and the score. To examine the correlation

between the two, their Spearman's Rho and Kendall's

tau were calculated to be 0.091 and 0.064

respectively, so it can be assumed that the release

month has a significant correlation with the film

score. As shown in Table 2, the average film score

showed an upward trend from January to March and

a downward trend from June to August, while the

average film score was higher in September,

November and December.

APMM 2025 - International Conference on Applied Psychology and Marketing Management

664

What's more, the data shows that when the playing

time is too short or too long, the film score will

decline. Therefore, most of the films now control the

film time between 85 and 110 minutes to reduce the

impact of the playing time on the film score. In the

descriptive statistical analysis of the data set, the

average playing time of the film is 107 minutes. This

study chose to use the regression model to find the

correlation between the playback duration and the

film score. At the same time, to avoid the influence of

extreme circumstances, the IQR was used to remove

the abnormal value of the playback duration. The

filtered results are shown in Fig. 2.

Figure. 2. IMDb_rating by runtime (filtered).

The filtered results are more intuitive, and the

slope of the regression line of the regression model is

0.268, indicating that there is a positive correlation

between the playing time and the film score.

This paper uses the Pearson correlation formula as

the calculation basis to calculate the correlation

between the numerical column in all movies and the

movie score, and the correlation matrix is shown in

Fig. 3.

Figure. 3. Correlation matrix.

Figure 3 shows that the correlation between

vote_average and imdb_rating and popularity is not

great. Therefore, even if a film is very popular on

social media, it may only indicate that the marketing

strategy before the film's release was relatively

successful, that the film's theme is attractive, or that it

happens to have a popular actor or director.

3.3 Prediction Model

To avoid the influence of film type on the score, this

study will eliminate this influence factor artificially

so the model can focus on predicting the popularity of

the same type of film. Combined with the above

research, the release time, the length of time, and the

popularity of the film were selected to establish the

model. In this section, classifiers, clustering, and

simple linear regression are used. At first, it is divided

into nearly 100 categories, that is, each score is a

category. At this time, a random forest is used for

prediction (Tan, 2021).

To get more accurate prediction results, this study

divides every 2 points into an interval of 0,1,2,3,4,

corresponding to terrible, poor, satisfactory, good,

and excellent. For the most accurate prediction, the

score is manually multiplied by 10. This allows

research to ignore the inaccuracy caused by floating

point values, i.e., every 20 points is taken as an

interval. Converting continuous values to discrete

values helps to fit the most appropriate values. Taking

80% of all movies as the training set and the

remaining 20% as the test set, four models will be

selected for implementation, namely random forest,

gradient boosting, KNN, and multiple linear

regression model. Especially in multiple linear

regression, the regression value will be rounded off

because the predicted value of linear regression is

usually related to the distribution of the target label,

so the rounded result can match the classification

label in most cases. In particular, when the regression

result is 4.6, it will become five according to the rule,

but there is no 5 in the classification, so it is necessary

to force the value greater than 4 to take four and

continue to follow the rounding between 0 and 4. The

obtained performance indicators are shown in Table

Predicting the Popularity of Movies with Social Media

665

Table 3. Performance index of each machine algorithm.

Random Forest Gradient Boosting KNN Multiple Linear Regression

Accuracy 86.65% 87.00% 60.46% 84.08%

MSE 0.13 0.13 0.45 0.45

MAE 0.13 0.13 0.41 0.41



0.59 0.60 -0.39 -0.39

The gradient boosting shows the best prediction

results, which shows that the gradient climb

algorithm has obvious advantages in dealing with the

problem of film popularity prediction. It can

effectively extract key features from a large number

of data and gradually approach the optimal solution

through the iterative learning process to improve the

accuracy of prediction and the generalization ability

of the model. Therefore, this paper establishes a

prediction model by gradient boosting and inputs the

budget, popularity, release month, playing time, and

user expected score on social media to obtain the

corresponding possible popularity level.

4 DISCUSSION

From the above research, it can be concluded that the

combination of specific directors and actors may form

a brand effect or "chemical reaction", thus affecting

the film score. Therefore, when investors choose

these specific combinations of directors and actors as

investment targets, they can ensure the quality and

popularity of the produced films. Similarly, some

combinations will lead to the reduction of film scores.

Investors can also try to choose some newer

portfolios through public feedback so that they can

learn from mature and better-performing portfolios so

that the new portfolio can better realize innovation for

the development of films.

Film duration also has a partial impact on film

ratings, so it should be limited effectively.

Meanwhile, it should not abandon the integrity of the

content in order to control the length of the film. At

that time, the harm to the film is greater than the

advantage, so investors should pay more attention to

the overall balance between the content and the time.

Choosing a more appropriate time for the release

of the film can enhance the popularity of the film.

Different types of films will be released at different

times. For example, comedy can be released during

the Spring Festival in China, and love films can be

released on Valentine's Day, which may make the

film more popular.

In any case, making a good film is the key to

winning the popularity of the public. Good content

plus the above key factors can make the film more

popular (Zou, 2019). Absolutely, early publicity work

is necessary, otherwise no one knows even though the

film is of high quality, which is one of the reasons for

the low popularity.

5 CONCLUSION

This study firstly explores the influence of actors and

directors' combinations on film ratings through

statistical analysis, then uses regression analysis to

select the main influencing factors and establish the

prediction model of film popularity, and finally

selects the gradient climbing algorithm to complete

the modeling. This paper has been conducted on a

sample of films with a duration ranging from 80 to

200 minutes. It is acknowledged that certain film

genres, such as epic films, may extend beyond this

range. Consequently, the data about these films is

incomplete. It is recommended that subsequent

studies extend the scope of research to include these

films.

The release month and rating of films in different

months have been shown to have a certain impact, but

different regions may show different results. For

example, the release of comedies during the Spring

Festival in China is a significant factor in their

popularity, and different regional cultures have been

found to affect films. In the future, the exploration of

this possibility in a specific region or a region with

the same culture will be a valuable avenue for further

research.

REFERENCES

A. Castillo, J. Benitez, J. Llorens, X. (R.) Luo, Social

media-driven customer engagement and movie

performance: Theory and empirical evidence. Decis.

Support Syst. 145, 113516 (2021)

J. Geng, M. Guo, Douban top 250 movie data mining and

score prediction. Hebei Enterp. (02), 11-13 (2021)

J. Tan, Research on IMDB movie score prediction based on

random forest algorithm. Mod. Comput. (30), 24-31

(2021)

APMM 2025 - International Conference on Applied Psychology and Marketing Management

666

M. T. Lash, K. Zhao, Early predictions of movie success:

The who, what, and when of profitability. J. Manag. Inf.

Syst. 33(3), 874-903 (2016)

N. Darapaneni et al., Movie success prediction using ML,

2020 11th IEEE Annual Ubiquitous Computing,

Electronics & Mobile Communication Conference

(UEMCON), New York, NY, USA, 0869-0874 (2020)

N. Li, Research on investment risk and box office revenue

prediction of China's film industry, Ph.D. thesis,

Dongbei University of Finance and Economics (2021)

Q. He, F. Yuan, Research on the influencing factors of film

consumption and box office prediction in the era of

digital economy: Based on the perspective of machine

learning and model fusion. Price Theory Pract. (09),

163-167+204 (2021)

R. Dhir, A. Raj, Movie success prediction using machine

learning algorithms and their comparison, 2018 First

International Conference on Secure Cyber Computing

and Communication (ICSCCC), Jalandhar, India, 385-

390 (2018)

S. M. R. Abidi, Y. Xu, J. Ni, et al., Popularity prediction of

movies: from statistical modeling to machine learning

techniques. Multimed. Tools Appl. 79, 35583-35617

(2020)

Social, Digital 2024 Global Overview Report (2024).

Retrieved from:

https://wearesocial.com/cn/blog/2024/01/digital-2024-

5-billion-social-media-users/

T. Liu, S. He, Y. Peng, Q. Yuan, D. Zeng, Research on

movie score prediction based on strategy coding. Syst.

Eng. 1-10 (no date)

T. Liu, S. Yu, W. Ni, Movie score prediction based on

attention mechanism and knowledge distillation.

Comput. Appl. Softw. (07), 5-60 (2024)

V. Gupta, N. Jain, H. Garg, et al., Predicting attributes

based movie success through ensemble machine

learning. Multimedia. Tools Appl. 82, 9597-9626 (2023)

X. Xu, M. Zhang, R. Zhao, Y. Zhu, Movie score prediction

based on interaction attribute enhancement. Soft. Guide,

(01), 182-189 (2024)

Y. Zou, An empirical study on the influencing factors of

movie word-of-mouth and the relationship between

word-of-mouth and box office: An analysis based on

Douban score data from 2011 to 2017. West. Leather

(06), 106-107 (2019)

Predicting the Popularity of Movies with Social Media

667