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

net/publication/344627182

Movie Recommendation System using Cosine Similarity and KNN

Article in International Journal of Engineering and Advanced Technology · June 2020

DOI: 10.35940/ijeat.E9666.069520

CITATIONS READS

9 5,560

5 authors, including:

Gaurav Srivastav
Datta Meghe Institute of Medical Sciences (Deemed University)
10 PUBLICATIONS 17 CITATIONS

SEE PROFILE

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

Effective Sensory Communication using GEAR Protocol View project

All content following this page was uploaded by Gaurav Srivastav on 13 October 2020.

The user has requested enhancement of the downloaded file.

 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-9 Issue-5, June 2020

Movie Recommendation System using

Cosine Similarity and KNN
Ramni Harbir Singh, Sargam Maurya, Tanisha Tripathi, Tushar Narula, Gaurav Srivastav

Abstract—Over the past years, the internet has broadened the Recommendation systems are primarily using three
horizon of various domains to interact and share meaningful approaches [6]. In content-based filtering, we do profiling
information. As it is said that everything has its pros and cons based on what type of content any user is interested in and
therefore, along with the expansion of domain comes
using the collected information, it recommends items.
information overload and difficulty in extraction of data. To
Another one is collaborative filtering, where we make
overcome this problem the recommendation system plays a vital
role. It is used to enhance the user experience by giving fast and clusters of similar users and use that information to make
coherent suggestions. This paper describes an approach which recommendations. Hybrid systems are the one which takes
offers generalized recommendations to every user, based on into account both above stated approaches to deal with
movie popularity and/or genre. Content-Based Recommender operational data more concisely [7]. Our goal is to provide
System is implemented using various deep learning approaches. accurate recommendations with less computational
This paper also gives an insight into problems which are faced in complexity.
content-based recommendation system and we have made an
effort to rectify them.
II. RELATED WORK
Keywords: Recommendation System, Content-Based
Recommender System, Deep learning Some of the common approaches of recommender
system are:
I. INTRODUCTION
1. Content-based filtering
Advancement in technology is reaching new heights every 2. Collaborative filtering
day and due to which we can see enormous growth in 3. Hybrid filtering
information. To deal with such large data we use machine A. Content Based Filtering
learning that automates analytical model building [1]. The
early classification of machine learning is divided into three This approach filters the items based on the likings of the
broad categories: Supervised learning, Unsupervised user. It gives result based on what the user has rated earlier.
learning and Reinforcement learning [2]. We use computers The method to model this approach is the Vector Space
to make predictions to help us achieve better results using Model (VSM). It derives the similarity of the item from its
various computational statistics. Tasks can be performed description and introduces the concept of TF-IDF (Term
without being explicitly programmed to do so [3]. It Frequency-Inverse Document Frequency) [28].
becomes a tedious task to extract the relevant information.
frequency occurrence of term t in document
Search engines solve the problem to some extent but it does Tf(t)= totalnumberoftermsindocumen 𝑡
not solve the personalization problem. Recommendation
System framework plays a vital role in today‟s internet totalnumberofdocument
surfing, be it buying a product from an e-commerce site or If(t)= 𝑙𝑜𝑔10numberofdocumentscontainingtermt
watching a movie on some video-on-demand service [4]. In
our everyday life, we depend on recommendations given by
other people either by word of mouth or reviews of general The similarity between item vectors can be computed by
surveys. People often use recommender systems over the three methods:
1. Cosine similarity
web to make decisions for the items related to their choice.
2. Euclidian distance
Recommendation systems are software tools and techniques
3. Pearson‟s correlation
whose goal is to make useful and sensible recommendations COSINE SIMILARITY
to a collection of users for items or products that might
interest them [5]. In other words, the recommender system Cosine similarity among two objects measures the angle of
or recommendation systems belongs to a class of cosine between the two objects. It compares two documents
information filtering system that aims at predicting the on a normalized scale. It can be done by finding the dot
„preference‟ or „rating‟ given to an item. product between the two identities.

Revised Manuscript Received on May 29.2020.

Gaurav Srivastav, Inderprastha Engineering College, AKTU
Ramni Harbir Singh, Inderprastha Engineering College, AKTU
Sargam Maurya, Inderprastha Engineering College, AKTU
Tanisha Tripathi, Inderprastha Engineering College, AKTU Tushar
Narula, Inderprastha Engineering College, AKTU

Published By:
Retrieval Number: E9666069520/2020©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.E9666.069520 556
& Sciences Publication
 Movie Recommendation System using Cosine Similarity and KNN

(SVD). It also speeds up the generation of result and

produces a reliable and efficient result.
Synonym: When many words have a similar meaning then
they are known as synonyms. In this problem, the
system sometimes fails to understand the difference between
two similar words and cannot produce the desired output.
E.g.: movie and film have the same meaning but the
recommendation system considers them as different words
and because of this it fails to give the accurate output.
Singular Value Decomposition (SVD), especially Latent
Semantic Indexing is capable of solving the synonymy
problem [12]. In other recommendation systems,
[25]
recommendations are based on the ratings and genres given
As the above diagram shows, the angle between v1 and v2
by other users due to which it became difficult to give
is. Lesser the angle between the two vectors more is the
accurate results. This is called collaborative filtering. This is
similarity. It means if the angle between two vectors is
a common method for a recommendation system. To make
small, they are almost alike each other and if the angle
things easier for people and come over this drawback, we
between the two vectors is large then the vectors are very
have used another method for the recommendation which is
different from each other.
content-based filtering. This method uses ratings and genres
B. Collaborative Filtering
given by the user itself. It gives recommendations based on
It depends upon the users who have similar interests and the movies watched by the user earlier. This helps us in
gives the result based on all the users. giving accurate suggestions because the ratings and genres
User-based: In user-based collaborative filtering, it is are given by the user only and depend entirely on the user‟s
considered that a user will like the items that are liked by choice, not on any other person‟s choice. Previous
users with whom have comparable taste. researches lack the accuracy of true recommendations due to
their dependency on other users. Here it is important to
Item-based: Item-based collaborative-filtering is mention that recommendations can be based on a particular
different, it expects the users to like items that are user and they assure to give recommendations on the mark
related to items that he has liked earlier. as they depend on the likes and dislikes of a particular user.
C. Hybrid Filtering The proposed system is capable of storing a large amount of
data and give efficient results. Our recommendation system
Hybrid filtering can be known as a combination of
searches for the best movies that are similar to the movie
collaborative filtering and content-based filtering. It is the
that we have watched based on genre and gives the result.
most common and popular technique today. It avoids the
weakness of every single recommender technique.
III. RESULT AND DISCUSSION
There are some problems related to the recommendation
system. They are: In the deep learning framework recommendation system, we
Cold-start problem: When a user registers for the first time, have used Cosine Similarity and Content-based filtering to
he has not watched any movie. So, the recommendation predict our result and recommend a movie to the user by
system does not have any movie based on which it can give running the code in python using NumPy and panda
results [8]. This is called the cold-start problem [9]. libraries.
Recommendation system goes through this problem as a A. Experiment Result
result of no previous record. This happens with every new The formula used to measure how similar the movies are
user once. based on their similarities of different properties.
Data sparsity problem: This problem occurs when the user Mathematically, it shows the cosine of the angle of two
has rated very few items based on which it is difficult for the vectors projected in a multidimensional space. The cosine
recommendation system to give accurate results. In this similarity is very beneficial since it helps in finding similar
problem, the results given are not much similar to the objects.
expected result. Sometimes, the system fails to give
successful results and generates weak recommendations
[10]. Also, data sparsity always leads to coverage problems.
Scalability: Another problem associated with
recommendation is scalability. In this, the encoding goes
linearly on items. The system works efficiently when the
data set is of limited size [11]. As the data set increases, it
becomes difficult for the recommendation system to give
accurate results based on varying genres of movies. The
scalability problem can be solved by dimensionality
reduction technique such as singular value decomposition

Retrieval Number: E9666069520/2020©BEIESP Published By:

DOI: 10.35940/ijeat.E9666.069520 Blue Eyes Intelligence Engineering
557 & Sciences Publication
 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-9 Issue-5, June 2020

Fig: This is the Cosine similarity formula which is

used for the recommendation of movies

Fig: Movies with different genres are takes and their

similarity is taken with the help of cosine similarity

IV. CONCLUSION
We have illustrated the modelling of a movie
recommendation system by making the use of content-based
filtering in the movie recommendation system. The KNN
algorithm is implemented in this model along with the
principle of cosine similarity as it gives more accuracy than
the other distance metrics and the complexity is
comparatively low too
Recommendations systems have become the most essential
fount of a relevant and reliable source of information in the
world of internet. Simple ones consider one or a few
Fig: To implement Cosine similarity we take an example parameters while the more complex ones make use of more
of 2 movies of different genre adventure and comedy parameters to filter the results and make it more user
friendly. With the inclusion of advanced deep learning and
other filtering techniques like collaborative filtering and
hybrid filtering a strong movie recommendation system can
be built. This can be a major step towards the further
development of this model as it will not only become more
efficient to use but also increase the business value even
further.

REFERENCES
1. Gediminas Adomavicius and Alexander Tuzhilin. Toward the next
generation of recommender systems: A survey of the state-of-the-art
Fig: Cosine similarity and possible extensions. Knowledge and Data Engineering, IEEE
The angle theta between the two movies will determine the Transactions on, 17(6):734–749, 2005.
similarity between the two movies. The theta ranges from 0- 2. Ricardo Baeza-Yates, Berthier Ribeiro-Neto, et al. Modern
information retrieval, volume 463. ACM Press New York, 1999
1. If the value of the theta is near 1 then it is most similar 3. ShumeetBaluja, Rohan Seth, D Sivakumar, Yushi Jing, Jay Yagnik,
and if it's near to 0 then it is least similar. The movie will be Shankar Kumar, Deepak Ravichandran, and Mohamed Aly. Video
recommended if it is close to 1 otherwise there would be no suggestion and discovery for youtube: taking random walks through
the view graph. In Proceedings of the 17th international conference on
similarity between them. It will recommend the best movies World Wide Web, pages 895–904. ACM, 2008.
to the user according to the Cosine similarity. After the 4. Xu Hailing, Wu Xiao, Li Xiaodong, and Yan Baoping.
cosine similarity, we have used a normalised popular score Comparison study of internet recommendation system. Journal of
Software, 20(2):350–362, 2009.
through which we get our function of computing distance. 5. T. E. D. Mining, “Enhancing teaching and learning through
Then by using the KNN functionality, we have found the educational data mining and learning analytics: An issue brief,” in
nearest neighbour which will be recommended to the user. Proceedings of a conference on advanced technology for
education, 2012.
6. Nakagawa and T. Ito, “An implementation of a knowledge
recommendation system based on similarity among users‟
profiles,” in Sice 2002. roceedings of the Sice Conference, 2002,
pp. 326–327 vol.1.
7. T. K. Quan, I. Fuyuki, and H. Shinichi, “Improving the accuracy
of recommender system by clustering items based on the stability
of user similarity,” in International Conference on Computational
Intelligence for Modelling Control and Automation, 2006, p. 61.

Published By:
Retrieval Number: E9666069520/2020©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.E9666.069520 558
& Sciences Publication
 Movie Recommendation System using Cosine Similarity and KNN

8. M. Muozorganero, G. A. Ramezgonzlez, P. J. Muozmerino, and AUTHORS PROFILE

C. D Kloos, “A collaborative recommender system based on
space-time similarities,” vol. 9, no. 3, pp. 81–87, 2010.
9. 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. ACM, 2001, pp. 285–295. Ramni harbir singh. Final year student of computer
10. G. Wang, “Survey of personalized recommendation system,” science engineering in Inderprastha Engineering College,
Computer Engineering & Applications, 2012. batch of 2016-20. Completed internship in core java from
11. Albadvi and M. Shahbazi, “A hybrid recommendation technique CEBS, Noida. Also has certification in python.
based on product category attributes,” Expert Systems with
Applications, vol. 36, no. 9, pp. 11 480–11 488, 2009.
12. F. R. Hernandez and N. Y. G. Garcia, “Distributed processing
using cosine similarity for mapping big data in Hadoop,” IEEE
Latin America Transactions, vol. 14, no. 6, pp. 2857–2861, 2016.
13. Anant Gupta, Dr.B.K.Tripathy.A generic hybrid recommender Sargam Maurya, final year student of computer science in
system based on neural networks. Advance Computing Inderprastha Engineering College of APJ Abdul Kalam
Conference (IACC), 2014 IEEE International. 21-22 February Technical University batch 2016-20. She has a certified
2014. training of Ethical Hacking from internshala and Python
14. Graham L. Giller (2012). ”The Statistical Properties of Random from coding ninjas.
Bit-streams and the Sampling Distribution of Cosine Similarity”.
Giller Investments Research Notes (20121024/1).
15. Nandhini Rengaraj. C.M.Kavitha. .Sabitha Loganathan.
N.Muthurasu. A study of existing systems of recommendation
engines. National Conference of Emerging Computing
Technologies and Applications. Vel Tech University. Avadi,
Chennai. 05-06 April 2018.
16. Robert Bell, Yehuda Koren, and Chris Volinsky. Modelling Tanisha Tripathi, final year student of computer science
relationships at multiple scales to improve the accuracy of large in Inderprastha Engineering College of APJ Abdul Kalam
recommender systems. In Proceedings of the 13th ACM SIGKDD Technical University batch 2016-20. She has completed
international conference on Knowledge discovery and data her internship in android from Ducat, Noida and also has
mining, pages 95–104. ACM, 2007. a certification in python.
17. Suvir Bhargav. Efficient features for movie recommendation
systems. 2014.
18. Dietmar Jannach, Markus Zanker, Alexander Felfernig, and
Gerhard Friedrich. Recommender systems: an introduction.
Cambridge University Press, 2010.
19. Joseph A Konstan. Introduction to recommender systems:
Algorithms and evaluation. ACM Transactions on Information
Systems (TOIS), 22(1):1–4, 2004. Tushar Narula, final year student of computer science in
20. Vilakone, P., Park, D., Xinchang, K. et al. An Efficient movie Inderprastha Engineering College of APJ Abdul Kalam
recommendation algorithmbased on improved k-clique. Hum. Technical University batch 2016-20. He has done training
Cent. Comput. Inf. Sci. 8, 38 (2018). in Java and Web Development from Webtek Labs Pvt.
21. Kelvin Luk, Introduction to TWO approaches of content-based Ltd and training in Python from coding blocks.
recommendation system
22. https://en.wikipedia.org/wiki/Recommender_system
23. https://medium.com/@bkexcel2014/building-movie-
recommender-systems-using-cosine-similarity-in-python-
eff2d4e60d24
24. https://www.kaggle.com/rounakbanik/movie-recommender-
systems
25. https://towardsdatascience.com/movie-recommender-system-part- Gaurav Srivastav is working as an Asst. Professor in
1-7f126d2f90e2 Inderprastha Engineering College,AKTU. He has
26. https://campus.datacamp.com/courses/introduction-to-natural- completed his B. Tech, M. Tech, and currently pursuing
language-processing-in-r/representations-of-text?ex=12 PHD. He has worked for Ericsson and Nokia in RF
27. https://www.researchgate.net/publication/11563559_Comparative_ department. His area of expertise is machine learning, e-
methods_for_the_analysis_of_continuous_variables_Geometric_in learning.
terpretations
28. https://towardsdatascience.com/calculate-similarity-the-most-
relevant-metrics-in-a-nutshell-9a43564f533e?gi=79783aadb374
29. https://www.ijcaonline.org/archives/volume124/number3/22082-
2015904111
30. https://www.forbes.com/sites/mitsubishiheavyindustries/2019/01/2
3/how-hydrogen-fuel-cells-can-power-the-world/#19e6cdb05695
31. https://github.com/jeknov/movieRec/blob/master/movieRec_descr.
Rmd
32. https://www.sciencedirect.com/science/article/pii/S000437021500
137X
33. https://www.researchgate.net/publication/266655527_Choice-
Based_Preference_Elicitation_for_Collaborative_Filtering_Recom
mender_Systems

Retrieval Number: E9666069520/2020©BEIESP Published By:

DOI: 10.35940/ijeat.E9666.069520 Blue Eyes Intelligence Engineering
559 & Sciences Publication
View publication stats