See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/348239082

Movie Recommender System Using Collaborative Filtering

Article · January 2021

CITATIONS READS

6 3,570

6 authors, including:

Meenu Gupta
Chandigarh University
88 PUBLICATIONS   537 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Research Paper View project

All content following this page was uploaded by Meenu Gupta on 05 January 2021.

The user has requested enhancement of the downloaded file.

 Proceedings of the International Conference on Electronics and Sustainable Communication Systems (ICESC 2020)
IEEE Xplore Part Number: CFP20V66-ART; ISBN: 978-1-7281-4108-4

Movie Recommender System Using

Collaborative Filtering
Meenu Gupta Aditya Thakkar Aashish
Department of Computer Science & Department of Computer Science & Department of Computer Science &
Engineering Engineering Chandigarh University, Engineering
Chandigarh University, Punjab Punjab Chandigarh University, Punjab
Gupta.meenu5@gmail.co m daditya571@gmail.com aashish.saini26@gmail.com

Vishal Gupta Dhruv Pratap Singh Rathore

Department of Computer Science & Engineering Department of Computer Science & Engineering
Chandigarh University, Punjab Chandigarh University, Punjab
premvishal9090@gmail.co m dhruv.rathore99@gmail.com

most preferred choice. User/Customer satisfaction is key for

Abstract: Movies are one of the sources of entertainment, but building the tool. It is beneficial for both customers and
the problem is in finding the desired content from the ever- companies, as the more satisfied the customer is, the more
increasing millions of content every year. However,
likely he/she would want to use the system for their ease,
recommendation systems come much handier in these
which would ultimately make revenues for the companies.
situations. The aim of this paper is to improve the accuracy
and performance of a regular filtering technique. Although Recommendation system should always be improved as the
varieties of methods are used to implement a recommendation user choice can differ from other users and if the user is not
system, Content-based filtering is the simplest method. Which happy with the result, he/she might not use it again which is
takes input from the users, rechecks his/her history/past the case with our system.
behavior, and recommends a list of similar movies. In this
paper, to prove the effectiveness, K-NN algorithms and Although there are a lot of algorithms, collaborative
collaborative filtering are used to mainly focus on enhancing filtering is the most popular one used by the companies as it
the accuracy of results as compared to content-based filtering. involves user’s interactions more. Collaborative filtering can
This approach is based on cosine similarity using k-nearest predict better than content-based filtering because it analyses
neighbor with the help of a collaborative filtering technique, the user’s browsing history and compares with other users and
at the same time removing the drawbacks of the content- then suggests results [15]. Whereas, the content-based filtering
based filtering. Although using Euclidean distance is takes the user’s information as an input to find similar movies
preferred, cosine similarity is used as the accuracy of cosine
and recommends them in descending order (using cosine
angle and the equidistance of movies remain almost the same. similarity). There’s another method named context-based
Keywords— Movie recommender system, cosine similarity, K- filtering where it extracts more information from the user like
NN algorithms, content-based filtering, collaborative filtering, mood, release date, genre, etc., to give more efficient results.
nearest neighbors
Our goal in this project was to keep our s ystem very accurate
I. INTRO DUCTIO N compared to other recommendation techniques while making
it as simple as possible. Content-based filtering has some
Recommendation systems are predicting systems that radically
drawbacks and a lack of accuracy and preciseness. So the
recommend items to users or users to the items, and
proposed system is the collaborative filtering recommendation
sometimes users to users too. Tech giants like YouTube,
system using nearest neighbors.
Amazon Prime, Netflix use similar methods to recommend
video content according to their desired interest. As the II. RELATED WORK
internet contains huge loads of data, finding your content is
There are many ways of recommending movies using
very difficult and can be very time consuming, thus the
Content-based, Collaborative (User-item, User-user), context-
recommendation plays an important role in minimizing our
based, hybrid methods, and nowadays deep learning is also
effort. These systems are getting more popular nowadays in
used to solve this problem.
various areas such as in books, videos, music, movies, and
other social network sites where the recommendation is used In [1], C. S. M. Wu, D. Garg, and U. Bhandary proposed a
to filter out the information. It is a tool that is using the user’s recommendation system using collaborative filtering where a
information to improve the suggestion result and give out the user’s rating is used to suggest the list. The authors have used

978-1-7281-4108-4/20/$31.00 ©2020 IEEE 415

Authorized licensed use limited to: Carleton University. Downloaded on August 08,2020 at 11:21:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Electronics and Sustainable Communication Systems (ICESC 2020)
IEEE Xplore Part Number: CFP20V66-ART; ISBN: 978-1-7281-4108-4

the Apache Mahout framework and essentially compared the As in [9], R. Van Meteren, and M. Van Someren created a
performances and efficiency of us er-based & item-based recommendation system by comparing the profile of the user
recommendations. with the content of each document in the sets of the collection.
These sets of terms can be represented as the content of the
In [2],R. E. Nakhli, H. Moradi, and M. A. Sadeghi document. The content-based system uses data of users and
proposed the percentage view approach for recommending interest and browsing history to determine the results . As this
movies to the users, it finds relevant movies for the customer requires a lot of domain knowledge, thus becomes a drawback
and then compares the performance with a random movie compared to collaborative filtering.
recommendation system for showing the accuracy of the
project. B. Collaborative Filtering (CF)

In [3], a content-based recommendation system is proposed Filters out the content according to user similar interest
by H. W. Chen, Y. L. Wu, M. K. Hor, and C. Y. Tang using with other users, it basically recommends the items to users
neural networks. In recent years, these are top topics for the that have similar taste [13]. It is also a popular and famous
researchers to work on when they want to build a movie algorithm in the industries. In the memory-based techniques,
recommendation system. there are two popular filtering algorithms [10]. There is
another technique known as model-based which is not as
Different terminology used in implementation of movie reliable as compared to memory-based techniques [17]. Figure
recommender system is discussed below. 2 and figure 3 discussed about the item based and user based
A. Content-based Filtering collaborative filtering.
This recommendation system requires some data or
information on what the user might like or what his previous
watched history. It is based on previous action or explicit
feedback. Most of the systems in the industry don’t use this
approach as they require data or they are not reliable enough.
For example, if a person watches the education documentary
genre more multiple times than the action genre, the person is
more likely to see the most-watched genre in the descending
order. The figure 1 below explains the process.

Figure 2: Demonstration of User-Based CF

Figure 1: Content-based filtering

Figure 3: Demonstration of Item Based CF

978-1-7281-4108-4/20/$31.00 ©2020 IEEE 416

Authorized licensed use limited to: Carleton University. Downloaded on August 08,2020 at 11:21:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Electronics and Sustainable Communication Systems (ICESC 2020)
IEEE Xplore Part Number: CFP20V66-ART; ISBN: 978-1-7281-4108-4

In [4] the user-based, it is assumed that the user will like the Eq. (1) is used to define the cosine similarity of the
items that are liked by users with whom they have similar proposed model.
tastes.

Consider Table 1 as an example, all the users like item A and

people who like item A also like Item C, Item-based are not-
dynamic in nature and do not change.

Table 1: Item-Based CF

USER/ITEM ITEM A ITEM B ITEM C

USER A ✔ ✔

Figure 4: Demonstration of the K-NN algorithm (value of

USER B ✔ ✔ ✔ k=3) [5].

USER C ✔ Recommended
● Item-based collaborative filtering- assumes users
will like items that are similar to the items that are
liked before by the user.
In the item-based like in [8], it is assumed that the user will
like those items that are similar to the other items liked before.
The hybrid approach-This approach provides very accurate
results using both collaborative and content-based filtering
while removing the drawbacks of the algorithms at the same
time. This integrated system is getting more attention
nowadays as it is better than both the algorithms [7].

III. PRO PO SED RECO MMENDATIO N ENGINE

The proposed recommendation system used the collaborative
filtering technique (item-based approach) which is far more
accurate and more efficient to use, as the item based method
can be done offline and because of its non-dynamic nature
whereas the user-based changes. The proposed approach uses
the KNN algorithm to find the distance between the target
movies with every other movie in the dataset and then it ranks
the top k nearest similar movies using cosine angle similarity.
Different techniques used in this proposed algorithm are
discussed below:
● KNN algorithm- is famous in a recommendation
system for its faster predictive nature and low
calculation time. KNN [16] classifies any unlabeled
class to their respective classes by prediction on a
similarity measure as shown in figure 4.
● Cosine similarity- to calculate the distance between
the target movie and the movies in the dataset, cosine
similarity is used. It measures the similarity between
two documents irrespective of how different they are Figure 5: Proposed Collaborative filtering
in size, and calculates the cosine angle between two
vectors in multi-dimensional space[6], Figure 5 shows the proposed collaborative filtering method.
The objective here is to recommend movies using the item-
based technique. First, the extraction of the dataset to gather
information about the target movie and the user’s rating.

978-1-7281-4108-4/20/$31.00 ©2020 IEEE 417

Authorized licensed use limited to: Carleton University. Downloaded on August 08,2020 at 11:21:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Electronics and Sustainable Communication Systems (ICESC 2020)
IEEE Xplore Part Number: CFP20V66-ART; ISBN: 978-1-7281-4108-4

Second, the collaborative filtering begins with the formatting python in this proposed approach. Every movie is rated
of the rating dataset so that it can be consumed by the KNN in a range of 1 to 5(5 being the highest). Figure 6
model, to remove the huge dataset handling problems . The shows, maximum rating of 3 and 4 for the movie in
dataset is reduced according to the popularity removing the respect to other scores.
noisy error pattern to get the sparse matrix.

Now cosine similarity is used to find the distance between

the target movie and other movies, which gives us the top k
nearest neighbor. And finally displaying the required
recommended list of movies with descending order of distance

In the KNN algorithm, if the value of K=1, then the case is

assigned to its nearest neighbor of that class. A case in KNN is
classified by the most majority vote of its neighbors, where the
case is being allocated to the class most common among its
nearest neighbors measured by a distance function.

IV. IMPLEMENTATIO N
Dataset Description- Movie lens dataset is used containing
28M ratings, over 1M tag application and 60k movies

There is two input database:

Figure 6: Count for each rating score
1. Movies- containing movie-id, genres, title, user-id in
Table 2
2. Rating- containing user-id, rating count, timestamp, import pandas as pd
movie-id in Table 3.
Filtering of the datasets is done on the basis only to popular import NumPy as np
ones by filtering the data frame to popular movies only. from scipy.sparse import csr_matrix
Table 2: Movie.csv from sklearn.neighbors import NearestNeighbors
S. No Movie Id Title
0 1 Toy Story (1995) Figure 7: Library used
1 2 Jumanji (1995)
As the dataset is huge and there can be where most of
2 3 Grumpier Old Men (1995)
the movies are not rated or rated by only one,
3 4 Waiting to Exhale (1995)
considering that, a sparse pivot matrix table is
4 5 Father of the Bride Part II (1995)
developed to transform the data frame into a proper
data frame which can be further implemented by KNN
Table 3: Rating.csv and filing 0’s in missing information fields.[13]
S.NO USERID MOVIEID RATING The KNN analyses the pivot table and uses cosine
0 1 1 4.0 similarity to find the similarity with the target movie
1 1 3 4.0 and shows the result.
2 1 6 4.0
3 1 47 5.0 V. RESULT ANALYSIS
4 1 50 5.0
The distance parameter evaluates the similarity or
distance between the chosen movie with the others,
For data analysis, python libraries are used to analyze and the distance is displayed in descending order were
the movie dataset (as shown in figure 7) and to gain the first recommendation.
insight into the dataset that helped us in building the
module. Every user has rated at least 20 movies. The
use of pandas and NumPy and ScikitLearn, scipy
libraries were for efficient results on Jupyter Notebook

978-1-7281-4108-4/20/$31.00 ©2020 IEEE 418

Authorized licensed use limited to: Carleton University. Downloaded on August 08,2020 at 11:21:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Electronics and Sustainable Communication Systems (ICESC 2020)
IEEE Xplore Part Number: CFP20V66-ART; ISBN: 978-1-7281-4108-4

 Recall is basically describes the relevant prediction from

the list of predictions.

 F-measure (F1) is the scores of the harmonic mean of

precision and recall.

 Mean Absolute Error (MAE) is inversely proportional to

the performance of the system, also known as Mean
Median Regression without considering the directions, it
Figure 8 (a): Recommendation List: Recommendations for measures the average magnitude of errors in a list of the
Crimson Tide (1995) set of predictions [11].

For example, the movie Crimson tide (1995), gives this

result in figure 8 (a).
The results in Table 4 show that our model Collaborative
filtering works better and shows better results than the regular
content-based filtering approach. As the value of the TP rate is
stronger than that of the content-based approach, proving the
accuracy is improved. The MAE value of our proposed system
is 0.248 which is lesser than other recommendation systems as
shown in table 5.

Table 4: Performance of recommendation algorithms

Approach TP Precision F1
rate

Content-based 0.591 0.501 0.528

Collaborati ve 0.761 0.782 0.772

based(Item-item)
Figure 8 (b): Recommendation list graph

Figure 8 (b), shows the recommendation is based on

similarity distance are lesser than the distance, more the
similarity with target movie Table 5: Comparison of MAE values
Method Algorithm MAE
To evaluate the performance of the proposed system and to
provide better results, experiments are conducted by Content-based T D-IDF 0.269
comparing any random existing system with our proposed
Collaborative MODEL-BASED 0.265
system based on terms of quality, accuracy, precision, recall,
time computation. Collaborative USER-USER 0.258

The accuracy-numeric value which determines the result of Collaborative IT EM-ITEM 0.248
calculation fulfills to the precise or standard value. The metric
used are discussed below from Eq. (2) to Eq. (5):

 Precision- is the ratio of recommended items that are

relevant to the total number of items on the list [11] [12].

978-1-7281-4108-4/20/$31.00 ©2020 IEEE 419

Authorized licensed use limited to: Carleton University. Downloaded on August 08,2020 at 11:21:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Electronics and Sustainable Communication Systems (ICESC 2020)
IEEE Xplore Part Number: CFP20V66-ART; ISBN: 978-1-7281-4108-4

VI. CO NCLUSIO N 11. P. Pu, L. Chen, and R. Hu, “A user-centric evaluation framework
for recommender systems.” In: Proceedings of the fifth ACM
In this paper, to avoid the use of content-based filtering, the conference on Recommender Systems (RecSys’11), ACM, New
Item-based CF filtering approach is used for obtaining better York, NY, USA, pp. 157-164, 2011.
results. KNN collaborative recommendation system is 12. M. Bekkar, H. K. Djemaa, and T . A. Alitouche, “ Evaluation
proposed using cosine similarity by employing Movielens measures for models assessment over imbalanced data sets,” J Inf
Eng Appl, vol. 3, no. 10, 2013.
dataset containing 28M rating for over 60K movies. The
13. B. Sarwar, G. Karypis, J. Konstan, and J. Riedl, “ Item-based
existing system are compared and found that the proposed collaborative filtering recommendation algorithms,”
system is more reliable and accurate. It is also found that when In Proceedings of the 10th international conference on World
the proposed methodology is applied to different larger Wide Web (pp. 285-295).
datasets, both accuracy, and efficiency increase which proves 14. P. Resnick, N. Iacovou, M. Suchak, P. Bergstrom, and J. Riedl,
that our system is both accurate and as well as efficient. This “GroupLens: an open architecture for collaborative filtering of
netnews,” In Proceedings of the 1994 ACM conference on
item-based filtering is more convenient than user-based. The Computer supported cooperative work (1994, October) (pp. 175-
main aim was to improve the regular recommendation 186).
algorithm and to provide better results. The research work was 15. J. A. Konstan, B. N. Miller, D. Maltz, J. L. Herlocker, L. R.
successful as it has been able to fulfill our aim of the project. Gordon, and J. Riedl, “ GroupLens: applying collaborative filtering
In the future, more features can be included to datasets (year to Usenet news,” Communications of the ACM, vol. 40, no. 3, pp.
77-87.
of release, actor, genre, casting details, etc) to make 16. W. Liang, G. Lu, X. Ji, J. Li, and D. Yuan, “Difference factor’
recommendations more reliable and innovative. The content- KNN collaborative filtering recommendation algorithm,”
based filtering and collaborative filtering can be combined to In International Conference on Advanced Data Mining and
minimize the errors and improve the performance as a hybrid Applications (2014, December). Springer, Cham, pp. 175 -184.
approach. 17. J. S. Breese, D. Heckerman, and C. Kadie, “ Empirical analysis of
predictive algorithms for collaborative filtering,” In Proceedings of
the Fourteenth conference on Uncertainty in artificial intelligence
(UAI’98), Morgan Kaufmann Publishers Inc., San Francisco, CA,
REFERENCES USA, pp. 43–52.
1. C. S. M. Wu, D. Garg, and U. Bhandary, “Movie Recommendation
System Using Collaborative Filtering,” In 2018 IEEE 9th
International Conference on Software Engineering and Service
Science (ICSESS), pp. 11-15, IEEE, 2018 Nov.
2. R. E. Nakhli, H. Moradi, and M. A. Sadeghi, “Movie
Recommender System Based on Percentage of View,” In 2019 5th
Conference on Knowledge Based Engineering and Innovation
(KBEI), pp. 656-660, IEEE.
3. H. W. Chen, Y. L. Wu, M. K. Hor, and C. Y. T ang, “Fully content-
based movie recommender system with feature extraction using
neural network,” In 2017 International Conference on Machine
Learning and Cybernetics (ICMLC) , vol. 2, pp. 504-509, Jul,
2017, IEEE.
4. P. Phorasim and L. Yu, “Movies recommendation system using
collaborative filtering and k-means,” International Journal of
Advanced Computer Research, vol. 7, no. 29, p.52, 2017.
5. M. Bahadorpour, B. S. Neysiani, and M. N. Shahraki,
“ Determining Optimal Number of Neighbors in Item-based kNN
Collaborative Filtering Algorithm for Learning Preferences of New
Users,” Journal of Telecommunication, Electronic and Computer
Engineering (JTEC), vol. 9, no. 3, pp.163-167, 2017.
6. A. R. Lahitani, A. E. Permanasari, and N. A. Setiawan, “Cosine
similarity to determine similarity measure: Study case in online
essay assessment,” In 2016 4th International Conference on Cyber
and IT Service Management, pp. 1-6, IEEE, Apr, 2016.
7. D. Pathak, S. Matharia, and C. N. S. Murthy, “ORBIT : Hybrid
movie recommendation engine,” In 2013 IEEE International
Conference ON Emerging Trends in Computing, Communication
and Nanotechnology (ICECCN), pp. 19-24, IEEE, Mar, 2013.
8. B. Sarwar, G. Karypis, J. Konstan, and J. Riedl, “Item-based
collaborative filtering recommendation algorithms,”
In Proceedings of the 10th international conference on World
Wide Web, pp. 285-295, Apr, 2011.
9. R. Van Meteren, and M. Van Someren, “Using content -based
filtering for recommendation,” In Proceedings of the Machine
Learning in the New Information Age: MLnet/ECML2000
Workshop, vol. 30, pp. 47-56, May 2000.
10. L. E. Molina and S. Bhulai, Recommendation System for Netflix,
2018.

978-1-7281-4108-4/20/$31.00 ©2020 IEEE 420

Authorized licensed use limited to: Carleton University. Downloaded on August 08,2020 at 11:21:42 UTC from IEEE Xplore. Restrictions apply.
View publication stats