Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

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

net/publication/344633647

Various Techniques for Movie Recommendation System

Article · October 2020

CITATIONS READS

0 406

2 authors, including:

Naveen Kumari
Punjabi University, Patiala
26 PUBLICATIONS   28 CITATIONS   

SEE PROFILE

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

Comparison between artificial View project

Study of various multi keyword search in cloud computing View project

All content following this page was uploaded by Naveen Kumari on 13 October 2020.

The user has requested enhancement of the downloaded file.


Journal of Interdisciplinary Cycle Research ISSN NO: 0022-1945

Movie Recommendation System


2 Author(s): Yogesh Kumar, Ms. Naveen Kumari
Guide Name: Ms. Naveen Kumari (Astt. Prof.)
yogi.peshlo@gmail.com
naveencse2k4@gmail.com
Punjabi University Regional Centre for IT & Management, Mohali

Abstract— The recommendation system plays an essential before watching movie. A different recommendation
role in the modern era and used by many prestigious schemes have been presented includes collaborative
applications. The recommendation system has made the filtering, content-based recommender system, and hybrid
collection of apps, creating a global village, and growth recommender system. However, several issues are raised
for abundant information. This paper represents the with users posted reviews.
overview of Approaches and techniques generated in the
Collaborative Filtering based recommendation system [1]. II. RELATED WORK
The recommendation system derived into Collaborative Zhang, R. et. al. [1] presented a new model family
Filtering, Content-based, and hybrid-based approaches. termed Markovian factorization of matrix process (MFMP).
This paper classifies collaborative filtering using various On one hand, MFMP models, such as timeSVD++, are
approaches like matrix factorization, user-based capable of capturing the temporal dynamics in the dataset,
recommendation, item-based recommendation. This and on the other hand, they also have clean probabilistic
survey also tells the road map for research in this area. We formulations, allowing them to adapt to a wide spectrum of
extract aspect-based specific ratings from reviews and also collaborative filtering problems. Two simple example
recommend reviews to users depends on user similarity models in this family are introduced for the prediction of
and their rating patterns. Finally, validating the proposed movie ratings using time-stamped rating data. The
movie recommendation system for various evaluation experimental study using MovieLens dataset demonstrates
criteria, and also the proposed system shows better result that the two models, although simple and primitive, already
than conventional systems. have comparable or even better performance than
Keywords: kNN, Sentiment Analysis, Opinion Mining, timeSVD++ and a standard tensor factorization model.
Product Review Zhang, J.; et al. [2] proposedhigh-efficient
recommendation algorithm, which exploits users’ profile
I. RECOMMENDATION SYSTEM attributes to partition them into several clusters. For each
Recommender systems are more popular and increase the cluster, a virtual opinion leader is conceived to represent the
production costs for many service providers. Today the whole cluster, such that the dimension of the original
world is an over-crowded so that the recommendations are useritem matrix can be significantly reduced, then a
required for recommending products or services. However Weighted Slope One-VU method is designed and applied to
recommender systems minimize the transaction costs and the virtual opinion leader-item matrix to obtain the
improves the quality and decision making process to users recommendation results. Compared to traditional
[1], [5], [6]. It is applied in various neighboring areas like clusteringbased CF recommendation schemes, our method
information retrieval or human computer interaction (HCI). can significantly reduce the time complexity, while
It gathers huge amount of information about user’s achieving comparable recommendation performance.
preferences of several items like online shopping products, Furthermore, we have constructed a real personalized web-
movies, taxi, TV, tourism, restaurants, etc. It stores based movie recommendation system, MovieWatch, opened
information of different ways either positive or negative it to the public, collected user feedback on
manner. It captures users review for watched movies, recommendations, and evaluated the feasibility and
traveled places, and purchased products. When compare accuracy of our system based on this real-world data
demand from the shopping products, service providers Haruna, K. et. al. [3] derived an alternative method
(travel, and restaurants), movie recommendation system based on the concept of co-occurrence from multi-soft sets
design a big problem since other recommendation systems to handle conflict situations and first used an illustrative
require fast computation and processing service from example of a movie selection problem to demonstrate the
service providers and product distributors. To recommend proposed approach and provide an extensive elaboration
movies, first collects the ratings for users and then using a publicly available dataset and provided a new
recommend the top list of items to the target user [2]. In measure based on support, strength, certainty and coverage
addition to this, users can check reviews of other users of soft set on movie selection problem and findings have

Volume XII, Issue IX, September/2020 Page No:847


Journal of Interdisciplinary Cycle Research ISSN NO: 0022-1945

revealed that the proposed approach achieved less Shaik, I. et al. [9] focused on K-means clustering
computational time when compared with the rough set- algorithm by analyzing the E-commerce big data. In this
based approach of up to 8.05%. research, geographical location and unique identification
Kalra, N.et al. [4] proposed a technique using trust number of the customer are considered as constraints for
factor extracted with help of ratings given so that quality clustering. Mining such data is a challenging task. One of
can be improved and better predictions can be done. A the important tasks of mining is to group similar objects or
novel-technique has been proposed for recommender system similar data into cluster which is very much useful for
using film-trust dataset and its effectiveness has been analysis and prediction. K-means clustering method is a
justified with the help of experiments. Traditional popular partition based approach for clustering data as it
recommender systems consider that all the users are leads to good quality of results.
independent and identical, its an assumption which leads to Mehra, J.; and Thakur, R.S.; [10] defined that clustering
a total ignorance of social interactions and trust among user. is one of the most important tasks in the active areas of Web
Trust relation among users ease the work of recommender Usage Knowledge Discovery. It assures to handle the
systems to produce better quality of recommendations. difficulty of information overload on the Internet while
Pavithra, M. et al. [5] designed and implement a movie many users are connected on the social media. Clustering is
recommendation system. There are different genres, cultures utilized for grouping information into comparative access
and languages to choose from in the world of movies. Such design for discovering client interest. There are two
a system can suggest a set of movies to users based on their drawbacks of FCM algorithm, firstly the requirements of no.
interest, or the popularities of the movies. On an average of of clusters c and secondly assigning the primary relationship
one year movie survey 600 movies are released in matrix. Due to these two drawbacks the FCM algorithm is
Hollywood. For streaming movie services like Netflix, hard to decide about the suitable no. of cluster and this
recommendation systems are essential for helping users find algorithm is insecure. The determination of desirable
new movies to enjoy. So far, a decent number of works has preliminary cluster is an important problem, therefore a new
been done in this field. But there is always room for technique called PDFCM algorithm is described.
renovation.
Nagamanjula, R. and Pethalakshmi, A. [6] proposed a III. PROPOSED WORK
movie recommender system based on new user similarity Start Input data set Pre-Process data set. i.e. remove
metric and opinion mining. The primary objective of this unnecessary data Extract features of input data set Use
paper is to find the type of opinions (positive, negative or preference region or input text Apply fitness function as
neutral) for movies and also suggest top-k recommendation given in eq 1 Calculate best weight according to eq 1 Apply
list for users and extract aspect-based specific ratings from improved kNN[15] classification
reviews and also recommend reviews to users depends on Apply proposed classification Generate and proposed results
user similarity and their rating patterns. Finally, validating End
the proposed movie recommendation system for various Start
evaluation criteria, and also the proposed system shows
better result than conventional systems Input data set
Mhetre, R. et al. [7] focused on movie recommendation
and the technique currently present for this is the
collaborative filtering technique. Of the collaborative Pre-Process data set. i.e. remove unnecessary data
filtering techniques, the matrix factorization algorithms
namely Alternative Least Square and Singular Value
Decomposition are implemented to predict or recommend Extract features of input data set
the movies. Further to improve the processing and time
computation for a large dataset we have used Apache Spark
along with Elastic search and the accuracy is compared Use preference region or input text
between the two algorithms for different values of testing
subsets.
Xi, W. et al. [8] proposed a novel recommendation Apply fitness function as given in eq 1
algorithm based on Back Propagation (BP) neural network
with Attention Mechanism (BPAM). In particular, the BP
neural network is utilized to learn the complex relationship Calculate best weight according to eq 1
of the target users and their neighbors. Compared with deep
neural network, the shallow neural network, i.e., BP neural
network, can not only reduce the computational and storage Apply improved kNN[25] classification
costs, but also prevent the model from over-fitting. In
addition, an attention mechanism is designed to capture the
global impact on all nearest target users for each user. Apply proposed classification

Generate and proposed results

End

Volume XII, Issue IX, September/2020 Page No:848


Journal of Interdisciplinary Cycle Research ISSN NO: 0022-1945

Fig 2: Flow Chart of positive outcomes that ought to have been returned. The
Step1. Initialization: run pre-determined generations kNN F1 score can be deciphered as a weighted normal of the
to generate initial population; exactness and review, where a F1 score achieves its best an
Step2. Parameter input: ask the DM to input his/her incentive even from a pessimistic standpoint at 0
preference information; F = 1/((a)(1/P) + (1-a)(1/R))
Step3. Determining reference region: Performance Evaluation on the basis of parameters for
Step4. Updating: update the population in the preference an Existing Technique
region by MOEA/D-DE until the stopping criterion is meet; TP FP Precisio Recall F- ROC
Step5. Interaction: If the DM is not satisfied with Rate Rate n Measure Area
reference information, go to Step2 to reset the reference
information; If the DM wants to get better results in this 0.45 0.009 0.902 0.45 0.621 0.888
reference region, go to Step4; otherwise, output the results.
1 0.55 9.11 1 0.954 0.888
(eq.
1) 0.917 0.467 9.25 0.917 0.904 0.888
Where N is the population size in the reference region,
and f i t denotes the objective value of individual i at the tth
generation. In this case, the function f is one with the
minimum value of the individual i. Further, δ means the Performance Evaluation on the basis of parameters for
average improvement degree of function value that the an Proposed Technique
current individuals are compared with their parent TP FP Precisio Recall F- ROC
individuals. The precision of the solutions presented to the Rate Rate n Measure Area
DM can be controlled by setting the value of δ. In other
words, the parameter δ is determined by the DM, when the
condition of derta ≤ δ is satisfied, the interaction happens. 0.5 0.007 0.923 0.5 0.649 0.904
Evaluation Metrics
Accuracy: it's an outline of systematic errors, a live of
applied mathematics bias; as these cause a distinction 0.993 0.5 0.921 0.993 0.966 0.904
between a result and a "true" worth, ISO calls this
exactitude.
Accuracy = (TP+TN)/(TP+TN+FP+FN)
Precision: it's an outline of random errors, a live of 0.921 0.428 0.921 0.921 0.911 0.904
applied mathematics variability.
Precision(P) = TP / (TP + FP)
Recall: Recall (also called sensitivity) is that the fraction
of relevant instances that are retrieved over total relevant
instances within the image. each exactness Associate in
Nursingd recall {are|ar|area unit|square live} thus supported
an understanding and measure of relevancy.
Recall (R) = TP / (TP + FN)
True positive (TP) = the amount of cases properly known
as true
False positive (FP) = the amount of cases incorrectly
known as true
True negative (TN) = the amount of cases properly
known as false
False negative (FN) = the amount of cases incorrectly
known as false

F-Measure: In statistical examination of parallel


arrangement, the F1 score (likewise F-score or F-measure)
is a measure of a test's precision. It considers both the
accuracy p and the review r of the test to process the score:
p is the quantity of right positive outcomes partitioned by
the quantity of every single positive outcome, and r is the
quantity of right positive outcomes isolated by the quantity

Volume XII, Issue IX, September/2020 Page No:849


Journal of Interdisciplinary Cycle Research ISSN NO: 0022-1945

of a word is further classified according to the classes.


Finally, top-k movies are recommended for target user. The
results suggested by our proposed system are leading and
block buster movies and the system is helpful for millions
and billions of users around the globe. Here the accuracy of
classification is improved using NbSVM classifier and also
meet the requirement of the users. We have tested our
proposed system on Movie Lens Dataset and also our work
shows better performance than ALS and SEHRS.

V. REFERENCES
[1]. Zhang, J.; Wang, Y.; Yuan, Z.; Jin, Q.; “Personalized
Real-Time Movie Recommendation System: Practical
Prototype and Evaluation”, Tsinghua Science And
Technology, vol: 25, 2020, pp: 180-191
[2]. Zhang, R.; Mao, Y.; “Movie Recommendation via
Markovian Factorization of Matrix Processes”, IEEE,
vol: 7, 2019, pp: 13189-13199
[3]. Haruna, K.; Ismail, M.A.; Suyanto, M; Gabralla, L.A.;
Bichi, A.B.; Danjuma, S.; Kakudi, H.A.; Haruna, M.S.;
Zerdoumi, S.; Abawajy, J.H.; Herawan, T.; “A Soft Set
Approach for Handling Conflict Situation on Movie
Selection”, IEEE, vol: 7, 2019, pp: 116179-116194
[4]. Kalra, N.; Yadav, D.; Bathla, G.; “SynRec: A
Prediction Technique using Collaborative Filtering and
Synergy Score”, International Journal of Engineering
and Advanced Technology, vol: 8, 2019, pp: 457-463
Proposed approach delineated a way to adapt discriminative [5]. Pavithra, M.; Sowmiya, S.; Tamilmalar, A.; Raguvaran,
re ranking to boost the performance of the generative S.; “Searching an Optimal Algorithm for Movie
models for grounded learning. Specifically, we have a Recommendation System”, International Research
tendency to dig into the matter of steering instruction Journal of Engineering and Technology, vol: 6, 2019,
following mentioned in last chapter and aid 2 PCFG models pp: 216-221
delineated earlier with the framework of discriminative re- [6]. Nagamanjula. R.; Pethalakshmi, A.; “A Novel Scheme
ranking. standard ways of discriminative re-ranking need for Movie Recommendation System using User
gold-standard references so as to guage candidates and Similarity and Opinion Mining”, International Journal
update the model parameters within the coaching part of re- of Innovative Technology and Exploring Engineering,
ranking. However, grounded learning issues don't have vol: 8, 2019, pp: 316-322
gold-standard references naturally available; so, direct [7]. Mhetre, R.; Priya, G.; “Movie Recommendation Engine
application of standard re-ranking approaches don't work. using Collaborative Filtering with Alternative Least
Instead, we have a tendency to show however the weak Square and Singular Value Decomposition
superintendence of response feedback (e.g., roaring task Algorithms”, International Journal of Advanced
completion within the steering task) will be used as another, Research in Computer and Communication
through an experiment demonstrating that its performance is Engineering, vol: 8, 2019, pp: 88-92
comparable and even more practical compared to coaching [8]. Xi, W.; Huang, L.; Wang, C.; Zheng, Y.; Lai, J.;
on gold-standard take apart trees. changed Re-ranking “BPAM: Recommendation Based on BP Neural
algorithmic rule for Grounded learning. In re-ranking, a Network with Attention Mechanism”, Proceedings of
baseline generative model is 1st trained and it generates a the Twenty-Eighth International Joint Conference on
collection of candidate outputs for every coaching example. Artificial Intelligence, 2019, pp: 3905-3911
[9]. Shaik, I.; Nittela, S.S.; Hiwarkar, T.; Nalla, S.; “K-
IV. CONCLUSION means Clustering Algorithm Based on E-Commerce
In this paper, we intend to present a movie recommendation Big Data”, International Journal of Innovative
system based on the combination of opinion mining and Technology and Exploring Engineering, vol: 8, 2019,
user similarity analysis. This system helps to recommend pp: 1910-1914
top-k movies for target user. In this work, we collect [10]. Mehra, J.; Thakur, R.S.; “Probability Density Based
reviews of users for movies and pre-process data with Fuzzy C Means Clustering for Web Usage Mining”,
certain major pre-processing steps. The pre-processed are International Journal of Innovative Technology and
given for explicit and implicit aspect extraction. The aspect Exploring Engineering, vol: 8, 2019, pp: 169-173

Volume XII, Issue IX, September/2020 Page No:850


Journal of Interdisciplinary Cycle Research ISSN NO: 0022-1945

[11]. M. Govindarajan”Sentiment Analysis of Movie


Reviews using Hybrid Method of Naive Bayes and
Genetic Algorithm “International Journal of Advanced
Computer Research, Vol.3, Issue-13, December-2013..
[12]. Dhanashri Chafale, Amit Pimpalkar” Sentiment
Analysis on Product Reviews Using Plutchik’s Wheel
of Emotions with Fuzzy Logic ” An International
Journal of Engineering & Technology , Vol. 1, Issue
No. 2, December, 2014.
[13]. G. Vaitheeswaran , L. Arockiam “Hybrid Based
Approach to Enhance the Accuracy of Sentiment
Analysis on Tweets ” IJCSET ,Vol 6, Issue 6, June ,
2016.
[14]. Pravin Keshav Patil, K. P. Adhiya “ Automatic
Sentiment Analysis of Twitter Messages Using Lexicon
Based Approach and Naive Bayes Classifier with
Interpretation of Sentiment Variation ” International
Journal of Innovative Research in Science, Engineering
and Technology, Vol. 4, Issue 9, September 2015.

Volume XII, Issue IX, September/2020 Page No:851


View publication stats

You might also like