© JUN 2023 | IRE Journals | Volume 6 Issue 12 | ISSN: 2456-8880

Filmview: A Review Paper on Movie Recommendation

Systems
PRIYANSHU MODI1, ATUL KUMAR2, BHASKAR KAPOOR3
1, 2, 3
Maharaja Agrasen Institute of Technology

Abstract- The proliferation of streaming platforms The widespread use of recommendation systems by
has led to a vast array of movie options, making it companies like Facebook, LinkedIn, Pandora, Netflix,
increasingly difficult for users to discover relevant and Amazon is highlighted. The review provides a
Content. To address this challenge, recommendation brief overview of different techniques and methods,
systems have emerged as valuable tools for offering valuable insights for further research in
suggesting movies based on user preferences. We recommendation systems. Jiang et al. [2]’s study
discuss the impact of temporal dynamics and social addresses scalability and practical usage feedback in
influence in improving recommendation accuracy Movie Recommendation Systems. It proposes a high-
and effectiveness. Moreover, we emphasize the efficient recommendation algorithm based on user
importance of incorporating explanations to clustering. The method achieves comparable
enhance user understanding and satisfaction. performance to traditional CF schemes with reduced
Through an examination of evaluation metrics, we time complexity. A Movie Recommendation System,
assess the performance of these systems. Overall, this is constructed for evaluation.
review contributes to the knowledge base, providing
insights into the strengths, limitations, and future The creation of a Movie Recommender System
directions of Movie Recommendation Systems. employing tools like K Means Clustering and K
Nearest Neighbour Algorithms is the subject of
I. INTRODUCTION research by Rishabh et al. [3]. MovieLens is the
dataset utilised, and Python is utilised to run the
With the increasing amount of Content available on system. We offer a number of machine learning ideas,
streaming platforms, it has become more challenging tools, and methods, such as KNN, K-Means
for users to find what they want to watch. To solve this Clustering, Collaborative Filtering, and Content-
problem, Movie Recommendation Systems are a type Based Filtering. The suggested system's
of personalized Recommendation System that uses a implementation details, including its architecture,
combination of machine learning algorithms and user process flow, and pseudocode, are given. With a best
data to provide movie suggestions. RMSE value of 1.081648, the results demonstrate that
the suggested system performs better than the state-of-
In this review paper, we will explore the different the-art methods. Choudhury et al. [4]’s Recommender
types of Movie Recommendation Systems, their systems (RS) address the information overload
strengths and weaknesses, and the factors that affect problem, particularly in movie recommendations.
their performance. We will also examine the various Comparisons are made between the four
evaluation metrics used to measure the efficacy of recommendation models BPNN, SVD, DNN, and
these systems. DNN with Trust. The DNN with trust model
demonstrates the highest accuracy of 83% and a low
II. LITERATURE REVIEW MSE value of 0.74, making it an effective choice for
movie recommendations.
The study by Mahesh et al. [1] presents a review of the
various techniques used in Movie Rec. Systems. It A content-based movie recommendation system using
explores Collaborative filtering, Content-based a variety of factors is suggested by Sahu et al [5].
filtering, Hybrid approaches, and deep learning Based on RS results, film ratings, and voting data, a
methods. Various similarity measures are examined. CNN deep learning model forecasts movie popularity.

IRE 1704718 ICONIC RESEARCH AND ENGINEERING JOURNALS 759

 © JUN 2023 | IRE Journals | Volume 6 Issue 12 | ISSN: 2456-8880

The study surpasses benchmark models in accuracy, Kalyan Kumar et al. [12]’s work aims to develop a
achieving 96.8%. Behera et al.'s paper [6] offers a neural network model for accurate recommendations
collaborative filtering method for movie using the MovieLens dataset. The data manipulation
recommendations that takes temporal effects into process and model implementation are carried out in
account. With improvements of 1.35% and 1.28% on Python. The evaluation metric used is the Hit-Ratio,
the ML-100K and 1M datasets, respectively, analysis with the model achieving an 87% Hit-Ratio. Fiagbe
of the Movielens dataset reveals that it outperforms [13]’s study develops a model using the MovieLens
leading models. Partitional Weighted co-clustering for rating dataset, employing matrix factorization to
movie recommendation is the main topic of Airen et predict movie ratings for unwatched films. The model
al.'s [7] research. The objective is to optimize user and then recommends movies with the highest predicted
movie neighborhoods by adjusting row and column ratings to users. The evaluation showcases the
cluster parameters. Experimental results on a movie system's accuracy, with low values of root mean
database demonstrate the proposed method's ability to squared error (RMSE) obtained.
provide more accurate personalized
recommendations, outperforming existing methods by III. METHODOLOGY
7.91%.
Machine learning algorithms are often the foundation
Gupta et al. [8]’s study incorporates K-NN algorithms of movie recommendation systems, which analyse
and Collaborative filtering, utilizing cosine similarity. user behaviour and data to produce tailored
The approach effectively combines the strengths of recommendations. Collaborative filtering and content-
both methods, mitigating the limitations of Content- based filtering are the two methods utilised for movie
based filtering. The use of cosine similarity provides recommendation systems most frequently.
accuracy which is comparable to Euclidean Distance.
Tahmasebi et al. [9]’s study proposes a deep • Collaborative Filtering:
autoencoder network-based hybrid social Collaborative filtering is a technique that uses the
recommender system with collaborative filtering, similarities and differences in user behavior to make
content-based filtering, and user social impact from recommendations. It can be done through two ways:
Twitter. The findings of the evaluation show that it is
more accurate and efficient than cutting-edge a. User-Based Collaborative Filtering: This method
techniques. Pecune et al. [10]’s study focuses on the pairs people with similar likes and suggests
importance of providing explanations in Movie products they've enjoyed. A distance metric like
Recommendation Systems. A conversational agent Pearson Correlation or Cosine Similarity is used to
was created using a human-centered design determine how similar two users are to one
methodology and it explains its suggestions in a another. The Pearson Correlation formula is as
manner similar to how a human would. The agent's follows:
architecture included the computational model of
explanations, which was then tested in a user 𝛴(𝑥𝑖 −𝑥)(𝑦𝑖 −𝑦)
experiment. The findings show that, regardless of the 𝑟 =
√𝛴(𝑥𝑖 −𝑥)2 𝛴(𝑦𝑖 −𝑦)2
quality of the recommendations, social explanations
improve how well the system and interactions are
received. correlation coefficient, or r.
xi = values of a sample's x-variable
A brand-new graph-based model that takes into x = the average of the x-variable's values
account user similarities, demographic data, and yi = values of a sample's y-variable
location is proposed by Darban et al. [11]. The use of y = the average of the y-variable's values
Autoencoder feature extraction improves the
reliability of recommendations and solves the cold- b. Item-Based Collaborative Filtering: This method
start issue. The efficiency of the suggested approach is suggests products that are comparable to ones that
demonstrated by experimental findings on the dataset. a consumer has previously enjoyed. A distance

IRE 1704718 ICONIC RESEARCH AND ENGINEERING JOURNALS 760

 © JUN 2023 | IRE Journals | Volume 6 Issue 12 | ISSN: 2456-8880

metric like Pearson Correlation or Cosine Matrix Factorization:

Similarity is used to determine how similar two Matrix factorization is a technique used to reduce the
objects are. The formula for Cosine Similarity is: dimensionality of large data sets. It is used to generate
recommendations based on the similarities and
∑𝑛𝑖=1 𝐴𝑖 𝐵𝑖 differences in user behavior. In this approach, the user-
𝑐𝑜𝑠(𝛩) =
√∑𝑛𝑖=1 𝐴2𝑖 √∑𝑛𝑖=1 𝐵𝑖2 item matrix is decomposed into two or more matrices
using singular value decomposition (SVD) or other
matrix factorization techniques. The formula for SVD
Where Ai and Bi are the ith components of the vectors is:
A and B respectively
𝑀 = 𝑈𝛴𝑁
Content-Based Filtering:
Using this method, objects are recommended based on M stands for the initial matrix that needs to be broken
their characteristics. It is used to make down. The eigenvectors of the matrix MMt are
recommendations for products that are comparable to contained in the left singular matrix U. The matrix has
past favourites of a user. A distance metric, such as the singular values and is diagonal. The eigenvectors of
Euclidean distance or the cosine similarity, is used to the matrix MtM make up V, the right singular matrix.
determine how similar two items are. The formula for For applications like movie recommendation systems,
Euclidean Distance is: this breakdown offers insights on the data's structure.

Evaluation and Comparison of Recommendation

𝑛
Models:
𝑑(𝑝, 𝑞) = √∑ (𝑞𝑖 − 𝑝𝑖 )2
𝑖=1 To evaluate and compare various recommendation
models, several metrics and techniques are used,
including accuracy metrics, ranking metrics, diversity
Euclidean n-space's p and q are two points.
metrics, and coverage metrics.
Starting from the space's origin (the starting point), qi
and pi are Euclidean vectors.
Accuracy Metrics:
Accuracy metrics are used to measure the degree to
Hybrid Filtering:
which a recommendation model accurately predicts
Hybrid filtering is a technique that combines both
user preferences. The most commonly used accuracy
Collaborative and Content-based filtering to make
metrics are:
recommendations. It is used to overcome the
limitations of each technique and provide more
a. Mean Absolute Error (MAE): This calculates the
accurate recommendations. Using a weighted average
typical discrepancy between an item's actual rating and
of their recommendations, collaborative and content-
anticipated rating for a certain user.
based filtering are integrated in this method. The
weighted average formula is as follows:
∑𝑛𝑖=1 |𝑦𝑖 − 𝑥𝑖 | ∑𝑛𝑖=1 |𝑒𝑖 |
𝑀𝐴𝐸 = =
𝑛 𝑛
∑𝑛
𝑖=1 𝑤𝑖 𝑋𝑖
𝑊 = ∑𝑛
𝑖=1 𝑤𝑖 b. Root Mean Square Error (RMSE): The difference
between the actual rating and the projected rating is
Weigted Average is W measured by the standard deviation. The formula for
There are n terms to be averaged
RMSD is:
The weights added to the x values are represented by
wi
The data values to be averaged are represeted by Xi ∑𝑛𝑖=1 (𝑥𝑖 − 𝑥̂𝑖 )2
𝑅𝑀𝑆𝐷 = √
𝑁

IRE 1704718 ICONIC RESEARCH AND ENGINEERING JOURNALS 761

 © JUN 2023 | IRE Journals | Volume 6 Issue 12 | ISSN: 2456-8880

Root-mean-square deviation (RMSD) 5. Model Training:

I = the variable i N = number of data points that are The selected model should be trained on the
not missing preprocessed data to learn the underlying patterns and
real observations (xi) time sequence relationships between movies and user preferences.
predicted time series = x̂i
6. Model Evaluation:
To assess the trained model's suitability for producing
c. Precision: This measures the proportion of relevant recommendations, a variety of measures are used,
items recommended to the user. including accuracy, ranking, diversity, coverage, and
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 novelty.
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
7. Model Deployment:
d. Recall: This measures the proportion of relevant Deploying the trained model in a production
items recommended out of the total relevant items. environment is the final step, where it can make
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 recommendations to users based on their preferences.
𝑅𝑒𝑐𝑎𝑙𝑙 =
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑠

IV. MODEL ARCHITECTURE

The steps below are often taken into consideration

while creating a movie recommendation system:
1. Data Collection:
The initial stage is to gather information on user
preferences and movie choices. This information can Fig. The recommendation system basic architecture
be found on a variety of websites, including those that
offer movie reviews, social media sites, and streaming V. DISCUSSION
services with user reviews.
The research presented in this review paper
2. Data Preprocessing: emphasises the crucial part that recommendation
The data, after collection, should be pre-processed to systems play in the film business. Many methods have
remove any duplicates, missing values, or irrelevant been investigated and contrasted, including
information. The data is also transformed into a collaborative filtering, content-based filtering, and
suitable format for the recommendation model. hybrid approaches. It is clear that combining different
strategies, such as applying deep learning algorithms,
3. Feature Extraction: can improve the efficacy and functionality of movie
The next step is to extract features from the recommendation systems.
preprocessed data that can be used to make
recommendations. These features may include movie The review also emphasizes the importance of
genres, actors, directors, ratings, and user preferences. considering dynamic factors in recommendation
systems. Traditional methods often fail to capture the
4. Model Selection: temporal and evolving nature of user-item
There are several alternative recommendation interactions. However, incorporating temporal effects
methods that can be used, including matrix in Collaborative filtering models has shown promising
factorization, collaborative filtering, content-based results in improving recommendation accuracy.
filtering, and hybrid filtering. The kind of data that is
available and the particular requirements of the Furthermore, the integration of social influence and
application determine the best model to use. explanations in recommendation systems has
demonstrated positive outcomes. The use of social

IRE 1704718 ICONIC RESEARCH AND ENGINEERING JOURNALS 762

 © JUN 2023 | IRE Journals | Volume 6 Issue 12 | ISSN: 2456-8880

characteristics and behaviors can provide valuable Vol. 19 No. 3 (2020), p. 18-37 DOI
insights into users' preferences and enhance the quality 10.5565/rev/elcvia.1232
of recommendations. Additionally, incorporating https://ddd.uab.cat/record/232276
explanations in the recommendation process improves [2] J. Zhang, Y. Wang, Z. Yuan and Q. Jin,
users' perception and understanding of the system's "Personalized real-time Movie Recommendation
decisions. System: Practical prototype and evaluation," in
Tsinghua Science and Technology, vol. 25, no. 2,
CONCLUSION pp. 180-191, April 2020, doi:
10.26599/TST.2018.9010118.
In conclusion, Movie Recommendation Systems have [3] R. Ahuja, A. Solanki and A. Nayyar, "Movie
become indispensable tools for users to navigate the Recommender System Using K-Means
vast amount of available Content. This review paper Clustering AND K-Nearest Neighbor," 2019 9th
has provided an overview of various techniques and International Conference on Cloud Computing,
approaches employed in Movie Recommendation Data Science & Engineering (Confluence),
Systems. It is clear that combining various techniques Noida, India, 2019, pp. 263-268, doi:
can produce recommendations that are more precise 10.1109/CONFLUENCE.2019.8776969.
and tailored, such as collaborative filtering, content- [4] Choudhury, S.S., Mohanty, S.N. & Jagadev,
based filtering, and deep learning algorithms. A.K. Multimodal trust based recommender
system with machine learning approaches for
The study also highlights the importance of addressing movie recommendation. Int. j. inf. tecnol. 13,
dynamic factors, such as temporal effects and evolving 475–482 (2021). https://doi.org/10.1007/s41870-
user preferences, in recommendation systems. By 020-00553-2
incorporating these factors, the performance and [5] S. Sahu, R. Kumar, M. S. Pathan, J. Shafi, Y.
relevance of recommendations can be significantly Kumar and M. F. Ijaz, "Movie Popularity and
improved. Target Audience Prediction Using the Content-
Based Recommender System," in IEEE Access,
Moreover, the inclusion of social influence and vol. 10, pp. 42044-42060, 2022, doi:
explanations in recommendation systems has shown 10.1109/ACCESS.2022.3168161.
promising results. The use of social characteristics and [6] Gopal Behera, Neeta Nain, Collaborative
behaviors, along with providing explanations for Filtering with Temporal Features for Movie
recommendations, enhances user satisfaction and Recommendation System, Procedia Computer
perception of the system. Science, Volume 218, 2023, Pages 1366-1373,
ISSN 1877-0509,
Overall, this review paper contributes to the existing https://doi.org/10.1016/j.procs.2023.01.115.
knowledge by presenting an in-depth analysis of [7] Sonu Airen, Jitendra Agrawal, Movie
different techniques and approaches used in Movie Recommender System Using Parameter Tuning
Recommendation Systems. It provides insights into of User and Movie Neighbourhood via Co-
the challenges faced by these systems and suggests Clustering, Procedia Computer Science, Volume
future directions for research. Incorporating 218, 2023, Pages 1176-1183, ISSN 1877-0509,
contextual data and user comments, for example, https://doi.org/10.1016/j.procs.2023.01.096.
could improve movie recommendation systems' [8] M. Gupta, A. Thakkar, Aashish, V. Gupta and D.
effectiveness even further. P. S. Rathore, "Movie Recommender System
Using Collaborative Filtering," 2020
REFERENCES International Conference on Electronics and
Sustainable Communication Systems (ICESC),
[1] Goyani, Mahesh; Chaurasiya, Neha. "A Review Coimbatore, India, 2020, pp. 415-420, doi:
of Movie Recommendation System: Limitations, 10.1109/ICESC48915.2020.9155879.
Survey and Challenges". ELCVIA : Electronic [9] Tahmasebi, H., Ravanmehr, R. &
Letters on Computer Vision and Image Analysis, Mohamadrezaei, R. Social movie recommender

IRE 1704718 ICONIC RESEARCH AND ENGINEERING JOURNALS 763

 © JUN 2023 | IRE Journals | Volume 6 Issue 12 | ISSN: 2456-8880

system based on deep autoencoder network using

Twitter data. Neural Comput & Applic 33, 1607–
1623 (2021). https://doi.org/10.1007/s00521-
020-05085-1
[10] Florian Pecune, Shruti Murali, Vivian Tsai,
Yoichi Matsuyama, and Justine Cassell. 2019. A
Model of Social Explanations for a
Conversational Movie Recommendation
System. In Proceedings of the 7th International
Conference on Human-Agent Interaction (HAI
'19). Association for Computing Machinery,
New York, NY, USA, 135–143.
https://doi.org/10.1145/3349537.3351899
[11] Zahra Zamanzadeh Darban, Mohammad Hadi
Valipour, GHRS: Graph-based Hybrid
recommendation system with application to
movie recommendation, Expert Systems with
Applications, Volume 200, 2022, 116850, ISSN
0957-4174,
https://doi.org/10.1016/j.eswa.2022.116850.
[12] Jena, K.K., Bhoi, S.K., Mallick, C. et al. Neural
model based Collaborative filtering for Movie
Recommendation System. Int. j. inf. tecnol. 14,
2067–2077 (2022).
https://doi.org/10.1007/s41870-022-00858-4
[13] Fiagbe, Roland, "Movie Recommender System
Using Matrix Factorization" (2023). Data
Science and Data Mining. 4.
https://stars.library.ucf.edu/data-science-
mining/4

IRE 1704718 ICONIC RESEARCH AND ENGINEERING JOURNALS 764