How to Cite:

Giridharan, N., Nathan, K. S., & Swetha, M. K. (2022). Movie recommendation system
using machine learning. International Journal of Health Sciences, 6(S8), 498–505.
[Link]

Movie recommendation system using machine

learning

Mr. N. Giridharan
Assistant Professor, Computer Science and Engineering, [Link] College
of Technology, Tiruchengode-637 215, Namakkal District, TamilNadu, India
Corresponding author email: giridharan@[Link]

K. Senthil Nathan
Students, Computer Science and Engineering, [Link] College of
Technology, Tiruchengode-637 215, Namakkal District, TamilNadu, India
Email: beingnathanz@[Link]

M. K. Swetha
Students, Computer Science and Engineering, [Link] College of
Technology, Tiruchengode-637 215, Namakkal District, TamilNadu, India
Email: smk937310@[Link]

Abstract---A recommendation system is a system that provides online

users with recommendations for particular resources, such as books,
movies, and music, based on a collection of data from the dataset.
These recommendation systems are beneficial for companies that
collect information from a large number of customers and aim to
provide the best recommendations for the [Link]-commerce,
recommendation systems help in personalizing the user experience by
providing items that the user is most likely to be interested in based
on similar items that the user has been interested in or items that
similar users have been interested in. Using datasets from kaggle to
help us analyse the accuracy of our [Link] factors, including the
genre of the movie, the actors, and even the director, should be
considered while developing a movie recommendation system. The
algorithms can recommend movies based on one or a combination of
twoor more criteria by using collaborativeand content-based filtering
techniques in machine [Link] cold-start problem, or the first
lack of item reviews, is a new user issue that affects the most of these
techniques. Here provided fresh user demographic data instead of
rating history to produce suggestions in order to avoid the cold-start
problem in this project.

Keywords---MRS, CF, CBF, ML, movie recommendation.

International Journal of Health Sciences ISSN 2550-6978 E-ISSN 2550-696X © 2022.

Manuscript submitted: 27 March 2022, Manuscript revised: 9 May 2022, Accepted for publication: 18 June 2022
498
 499

Introduction

Provided a machine learning framework that evaluates the use of a number of

online suggestion attributes for recommendation generation, including user, item,
and reviews. Experiments are carried out using the Movie Lens, Film Trust
dataset, Kaggle, to assess the performance of the suggested architecture.
Here utilized a K-Means architecture, which enables the substitution of particular
algorithms, to reduce the dependency on the user recommendations in web
mining.

People can find suitable movies with the use of recommendation systems, which
are a collection of techniques and algorithms. To predict future actions based on
historical data, they utilize a range of strategies and techniques, such as
vectorization. In this project, the movie is recommended to different types of
online users, and the movie recommendation system is built using the TMDB
dataset and the Movie Lens open-source dataset. There are literally thousands of
options available online that recommends particular resource to particular users
online. Just a few examples include social networking, internet shopping,
YouTube videos, and more. A user can customize a platform and find content
that they like with the use of recommendation systems.

The two types of machine learning methods applied in recommender systems are
content-based systems and collaborative filtering systems. Modern recommender
systems include content-based algorithms and collaborative filtering.
To predictthe users' reactions to movies they haven't yet seen or that are relevant
to their recent browsing history in to provide them movie suggestions. On the
basis of these ratings and past online activity, movies are then categorized and
recommended to users. To do this, predicted future ratings using historical
information on movies and user reviews.

Objectives

i. The major objectives of this work isto suggest movies to users based on
their watching history and ratings by using datasets like MovieLens and
TMDB.
ii. Using the machine learning algorithms like Collaborative Filtering, Content-
Based Filtering, Singular Value Decomposition, our project proposes to
develop movie recommendation system.

Literature Review

Jiang Zhang, Yufeng Wang, Zhiyuan Yuan, and Qun Jin, With the explosion of big
data, have stated that realistic recommendation systems are now important in a
variety of industries, such as e-commerce, social networks, and a number of web-
based services[1]. Fangtao Li, Sentiment and subject lexicon extraction is
essential for opinion [Link] studies have demonstrated that supervised
learning techniques are more effective for this task[2]. Lei Zhang, Interpreting
people's ideas about a user's attributes has been suggested as a key functions of
opinion mining. For instance, the statement "I appreciate the GPS function of
Motorola Droid" indicates approval for the Motorola phone's "GPS function." The
500

feature is "GPS function." [3].Kang Liu, has made offers based on a word-based
translation model, this research suggests a novel method for extracting opinion
objectives (WTM). First, use WTM to mine the relationships between opinion
targets and opinion terms in a monolingual situation [4]. GuangQiu, has
proposed opinion analysis, also known as sentiment analysis or opinion mining,
has received a lot of interest recently due to its many useful applications and
difficult research challenges [5]. Kanjar De, Partha Pratim Roy, and Sudhanshu
Kumar have proposed for use in e-commerce and digital media, recommendation
systems (RSs) have attracted a lot of attention. Collaborative filtering (CF) and
content-based filtering (CBF) are examples of traditional approaches in RSs.
These systems have some drawbacks, such as the requirement of prior user
history and habits for executing the task of recommendation [6].Hsuan Chi
and Chu-Hsing Lin, have proposed one of the most popular recommendation
systems is for movies, and related technologies are always getting better [7]. Arjun
Mukherjee, has proposed writing comments on news stories, blogs, or reviews has
become a common activity on social media. In this study, examine customer
feedback on reviews [8]. Zhiyuan Liu, In the Web 2.0 age, users frequently freely
annotate online content with tags. There has been a lot of interest in automatic
tag suggestion to simplify the annotation process [9].Qin Gao, has presented a
framework for word alignment that can take into account manual alignments in
part. The key component of the strategy is a brand-new semi-supervised
technique that adds a constrained EM algorithm to the widely used IBM Models
[10]. Shumeet Baluja, Users have a huge opportunity to discover content that
interests them because to YouTube's ever expanding video library. The discovery
of fresh content is unfortunately a challenging task because of the size of the
video library and the complexity of searching videos [11].Partha Pratim
Talukdar, has proposed a solution that offer a graph-based semi-supervised label
propagation approach for collecting open domain labelled classes and their
examples from a mix of unstructured and structured text sources [12]. GuangQiu
has made an offer that the sentiment lexicon is crucial to the majority of
sentiment analysis systems. However, because various words may be used in
different areas, it is difficult, if not impossible, to compile and maintain a
universal sentiment lexicon for all application domains. [13]. Robert C. Moore,
themajority of methods for statistical machine translation are built on bilingual
word alignment. The majority of current word alignment techniques are based on
the generative models [14]. XiaowenDin as the user-generated content, such as
blog postings, forum posts, and consumer reviews of lenses, is one of the key
sources of information on the Internet [15].

Existing System

In the current recommendation system, recommender systems use knowledge

discovery approaches to tackle the challenge of providing real customers with a
specific product of the [Link] recommender systems, the recent
growth in customers and items presents some significant challenges. For millions
of customers and items, that these are generating by high-quality
recommendations and performing several recommendations per [Link]
Singular Value Decompositionbased recommendation systems must go through
relatively expensive matrix factorization phases, they can rapidly deliver a high-
quality movie recommendation system. Here proposed and experimentally test a
 501

method that may be used to gradually create a model to enable highly scalable
recommender [Link] scalability issue in a practical movie recommendation
system must be fixed in order to enable real-time recommendations by the use of
content-based filtering with all of its key aspects.

Proposed Methodology

In the proposed work of movie recommendation system suggest by connecting

users with those who share their interests, collaborative filtering with K-Means
Clustering recommends products. It gathers user input in the form of reviews
submitted by users for a certain item and looks for consistency in rating
behaviors among users to identify groups of people with comparable tastes. A
user profile in this case indicates the preferences a user has provided, either
directly or implicitly. As an illustration, consider Amazon's implementation of the
CF method, which proposes products based on customer reviews and purchase
history. Each user, individually, has a list of things that have been either
explicitly or implicitly rated.

Implementation

Dataset Analysis

Toget a better understanding of the movie dataset that could help in creating a
movie recommender system using Python libraries, data from datasets in Kaggle,
MovieLens, and TMDB were analyzed. Here, similarities such as the most popular
films, the most popular genres, the number of films in each genre, and the
number of films rated in each rating category were identified. These similarities
were then analyzed using machine learning algorithms.
502

Data Pre-processing

Creating a database for movie recommendation system and processing data sets
obtained from the MovieLens dataset and TMDB dataset. The use of the dataset
determines the validity of the results, thus creating the database is an important
step. It is possible to have a sufficient number of highly predicted items for
recommendations to each user by using the datasets made available by several
websites, which include users and items with considerable rating history.

Model Building

Python libraries are used to create the recommender system. In addition to the
Pearson CorrelationSimilarity class, which employs the PearsonCorrelation
Coefficient to assess the similarity between users' evaluations, the User Similarity
class was used in this instance of collaborative filtering. CFg is the most popular
recommendation technique, and computing similarity is an important factor of
this technique. The co-occurrence of comparable events is simple, generally
applicable method of determining [Link] is easy to see that the position of
user and item is interchangeable when their count size gap is not too much, thus
proposinga vectorization model to show its performance. The similarity based on
co-occurrence information suffers from cold start, so proposed a content-based
similarity model based on which is another technology in NLP.

Model Evaluation

The movie recommender system constructed in this paper helps the

understanding of how a recommender system works. To evaluate the accuracy
and relevancy of the results produced by the system,here analyzed both the
approaches differently. Then compare the Item based similarity coefficient results
by mapping the MovieID. Using Python pandas libraries, to get the result. As
obvious from the table, movies which are comparable are given a greater
similarity metric. Next,assess the rating predictions on test data against the
actual ratings as indicated in the training data. With recommender systems, the
user and item cold-start problem are always expressed and so evaluated the
system based on these issues and offer implementation design to resolve them.

Algorithm

Content Based Filtering

When making recommendations, content-based filtering compares features of

related goods, services, or pieces of content as well as user information collected
over time. When making recommendations, content-based filtering compares user
profiles to the keywords and attributes that have been allocated to objects in a
database such as the items in an online market. A user's behavior, such as
purchases, ratings (likes and dislikes), downloads, products searched for on a
website and/or added to a cart, and product link clicks, are used to form the user
profile. Assigning attributes to database objects enables content-based filtering by
providing the algorithm information about each object. These characteristics
mainly depend on the goods, services, or information you're [Link]
 503

metrics are calculated during recommendation using the item's feature vectors
and the user's preferred feature vectors from prior records. The top few are then
recommended. When making suggestions to a single user, content-based filtering
does not require the data of other [Link] method is entirely based on
evaluating user preferences against product features. What's recommended are
the products that have the greatest features in common with user interests.
Because product features are so important in this system, it's necessary to talk
about how users choose their favorite [Link] are two methods that can be
used in this situation (possibly in combination). To start, consumers can be
provided a list of features from which they can select the ones that they best
identify with. Second, the algorithm may remember which goods the user has
previously selected and add those features to the user's data. Product features, on
the other hand, can be identified by the product's developers.

Result

The dataset includes different ratings and tags that users have assigned to
different movies as part of the MovieLens and TMDB datasets for movie
recommendation systems. For inclusion, users were chosen at random. Each
selected user had given at least 20 films a rating. MovieLens bases its
recommendations on data provided by web users, such as movie reviews. The
website makes use of several machine learning-based recommendation filtering
techniques, including content-based filtering, collaborative filtering, and hybrid
filtering. New users are prompted to rate how much they love watching various
movie genres in the system (for example, movies with horror or horror
comedies).Even before the user has given the website a big number of movie
ratings, the algorithm can generate first recommendations based on the
preferences gathered from the [Link] entering thewebsite, the user will be
asked to sign in using mobile number or email id to create account in the website,
and they will get a OTP for verification. User can select the genre of their choice
and the movie will be recommended according to the user’s choice by using the
content-based filtering algorithm in machine learning.

Conclusion

In this research, a collaborative filtering (CF) and content-based filtering (CBF)

algorithm-based movie recommendation system based on rating and previously
seen movie history is proposed. Discussed about what has been done to address
these cold-start problems as well as what needs to be done in the form of various
research possibilities and suggestions that may be used to address problems like
context-awareness, grey sheep, and the cold-start [Link] system's
effectiveness has been evaluated using a large dataset that is used to recommend
movies to users very effectively. This paper will offer recommendations and
provide a framework for a movie recommendation system that takes into account
for user opinion.

Future Work

Additionally, gaining a better result and high accuracy may enable the
recommender system to produce a result that is even more accurate than the
504

previous one. Hence having an imbalanced data issue in collaborative filtering.

Similar user ratings are quite rare. So can group persons, items, or both based on
their properties using clustering algorithms like K-Means. Then employ additional
features in the hybrid technique to obtain more accurate predictions.

References

1. Jiang Zhang, Yufeng Wang, Zhiyuan Yuan, Qun Jin, “Personalized real-time
movie recommendation system: Practical prototype and evaluation” published
in Tsinghua Science and Technology, 2020, p. 180-191.
2. F. Li, S. J. Pan, O. Jin, Q. Yang, and X. Zhu, “Cross-domain coextraction of
sentiment and topic lexicons,” in Proceedings of the 50nd Annual Meeting on
Association for Computational Linguistics, Jeju, Korea, 2020, pp. 410–419.
3. L. Zhang, B. Liu, S. H. Lim, and E. O’Brien-Strain, “Extracting and ranking
product features in opinion documents.” in Proceedings of the 23th
International Conference on Computational Linguistics, Beijing, China, 2020,
pp. 1462–1470.
4. K. Liu, L. Xu, and J. Zhao, “Opinion target extraction using word-based
translation model,” in Proceedings of the 2012 Joint Conference on Empirical
Methods in Natural Language Processing and Computational Natural
Language Learning, Jeju, Korea, July 2019, pp. 1346–1356.
5. G. Qiu, L. Bing, J. Bu, and C. Chen, “Opinion word expansion and target
extraction through double propagation,” Computational Linguistics, vol. 37,
no. 1, pp. 9–27, 2019.
6. Sudhanshu Kumar, Kanjar De, Partha Pratim Roy, “Movie Recommendation
System Using Sentiment Analysis From Microblogging Data” published in
IEEE Transactions on Computational Social Systems, 2020, p. 915-923.
7. Lin CH, Chi H, “A novel movie recommendation system based on collaborative
filtering and neural networks. ” In: International conference on advanced
information networking and applications. Cham: Springer, 2019, p. 895–903.
8. T. Ma and X. Wan, “Opinion target extraction in chinese news comments.” in
Proceedings of the 23th International Conference on Computational
Linguistics, Beijing, China, 2019, pp. 782–790.
9. A. Mukherjee and B. Liu, “Modeling review comments,” in Proceedings of the
50th Annual Meeting of the Association for Computational Linguistics, Jeju,
Korea, July 2019, pp. 320–329.
10. Z. Liu, X. Chen, and M. Sun, “A simple word trigger method for social tag
suggestion,” in Proceedings of the Conference on Empirical Methods in
Natural Language Processing, Edinburgh, United Kingdom, 2019, pp. 1577–
1588.
11. Q. Gao, N. Bach, and S. Vogel, “A semi-supervised word alignment algorithm
with partial manual alignments,” in Proceedings of the Joint Fifth Workshop
on Statistical Machine Translation and MetricsMATR, Uppsala, Sweden, July
2019, pp. 1–10.
12. S. Baluja, R. Seth, D. Sivakumar, Y. Jing, J. Yagnik, S. Kumar, D.
Ravichandran, and M. Aly, “Video suggestion and discovery for youtube:
taking random walks through the view graph,” in Proceedings of the 17th
International Conference on World Wide Web, Beijing, China, 2019, pp. 895–
904.
 505

13. ShobhaRani, P., Sivalakshmi, P., Akarsh, A., Bhanuhithesh, C. H., & Aneesh,
C. (2021). Characterization Of Plant Disease Prediction Using Convolutional
Neural Network. Turkish Journal of Computer and Mathematics
Education, 12(10), 1905-1912.
14. G. Qiu, B. Liu, J. Bu, and C. Che, “Expanding domain sentiment lexicon
through double propagation,” in Proceedings of the 21st International Jont
Conference on Artifical Intelligence, Pasadena, California, USA, 2020, pp.
1199–1204.
15. Balasaranya, K. (2021). Real Time Emotion Recognizer And Classifier For
Facial Expressions Based On Machine Learningapproach. Turkish Journal of
Computer and Mathematics Education (TURCOMAT), 12(10), 1958-1964.
16. Rinartha, K., & Suryasa, W. (2017). Comparative study for better result on
query suggestion of article searching with MySQL pattern matching and
Jaccard similarity. In 2017 5th International Conference on Cyber and IT
Service Management (CITSM) (pp. 1-4). IEEE.
17. Rinartha, K., Suryasa, W., & Kartika, L. G. S. (2018). Comparative Analysis of
String Similarity on Dynamic Query Suggestions. In 2018 Electrical Power,
Electronics, Communications, Controls and Informatics Seminar (EECCIS) (pp.
399-404). IEEE.
18. Chanana, M. (2018). Empirical study: relationship between self efficacy and
academic performance. International Journal of Health & Medical
Sciences, 1(1), 28-34. [Link]