Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).

An Exploration of Machine Learning

DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

in Movie Recommendation

ASVSNM M anjusha Pachigulla Ajay Kumar Singireddy Sreevani

Computer Science and Engineering Computer Science and Engineering Computer Science and Engineering
Koneru Lakshmaiah Education Koneru Lakshmaiah Education Koneru Lakshmaiah Education
Foundation Foundation Foundation
Vaddeswaram,India Vaddeswaram,India Vaddeswaram,India
[email protected] [email protected] [email protected]

Chatradi Tejaswini Dr. K.Padmanaban Dr. A.Dinesh Kumar

Computer Science and Engineering Computer Science and Engineering Computer Science and Engineering
Koneru Lakshmaiah Education Koneru Lakshmaiah Education Koneru Lakshmaiah Education
Foundation Foundation Foundation
Vaddeswaram,India Vaddeswaram,India Vaddeswaram,India
[email protected] [email protected] [email protected]

Abstract—In today’s culture, technology plays a crucial role Consequently, the evolution of movie
for any kind of recommendations. In this digital World, recommendation systems based on algorithms has the
Algorithms are used in movie recommendation systems to offer potential to significantly improve user serialized dramas
films to users based on their watching history or ratings. With while also driving income for content producers.
the advent of digital services and the massive quantities of
information they accumulate on user preferences, these There are several algorith ms for movie
systems have grown in popularity in recent years. This system recommendation systems, each with its own set of strengths
employs machine learning to do sentiment analysis on movie and flaws. Co llaborative filtering algorith ms generate
reviews to improve the user experience. This article also suggestions based on the watching behaviors of similar
compares NB, S VM, and KNN on metrics such as Accuracy, users, whereas content-based screening strategies make
Precision, Recall, and F1 S core. We present an overview of the recommendations utilizing the characteristics of the films
many algorithms utilized in the creation of movie themselves. In an effort to increase suggestion accuracy,
recommendation systems, encompassing text categorization, hybrid techniques that mix adaptive algorithms and
knowledge filtering, and hybrid techniques, in this review. We evidenced filtering have also been developed.
also examine the problems and limits of these algorithms, as
well as possible future research objectives in this subject. In this review, we present an overview of the many
algorith ms utilized in the creation of mov ie reco mmendation
Keywords—Support Vector Machine (SVM), Naive Bayes systems, while also providing a discussion of their problems
and K-Nearest Neighbors Algorithm. and limits. We also discuss some potential future areas fo r
study in this area.
I. INT RODUCT ION
II. LIT ERAT URE REVIEW
The Internet has developed swiftly since its
inception. Because of the abundance of data accessible via In recent years, there have been a substantial amount
the internet, it is difficult to obtain important info rmation of study on movie reco mmendation systems, with many
quickly and easily. Fortunately, reco mmendation systems algorithms and techniques being created and evaluated.
assist to overcome this difficulty.
One of the most prominent ways is particle swarm
Nowadays, Movie optimizat ion techniques have optimization, wh ich provides suggestions based on
become an essential aspect of the entertainment business in comparable users' viewing behaviors. There are t wo types of
recent years. With the growth of streaming p latforms and collaborative filtering: sensation collaborative filtering and
the massive quantities of data collected on user preferences, framework collaborative filtering. These produce values
these systems have the ability to assist viewers find new based on the complete user-item dataset, whereas model-
films that they may appreciate while also driving inco me fo r based collaborative filtering algorith ms develop a latent
content producers. representation of the data before making reco mmendations.
Despite their usefulness, these algorithms have a The merits of these methods are Noise reduction and
number of obstacles and constraints that must be addressed. Interpretation, the demerits are overfitting, interaction
The "cold start" problem, primarily alludes to the difficulty effects.
in producing suggestions for clients that have not yet Another method is content-based filtering, whereby
contributed sufficiently data for the system to effect ively produces suggestions based on film qualities such as genre,
forecast their preferences, is one of the challenges. Another actors, and narrative. These algorithms often offer
difficulty is balancing the diversity and originality of suggestions based on a feature wavelet coefficients of the
recommendations with about their ultimate correctness. By movie and approaches such as cosine similarity.
understanding the interaction between the agents helping
multiple users, the cold start challenge can be solved.

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 890

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

By comb ining the qualities of both methodologies, these F. Data Transformation

approaches have the ability to increase suggestion accuracy. Depending on the nature of your data, you may
The cosine similarity approach is used to need to apply various modifications to your features before
recommend movies. The distance function is computed using them with your preferred reco mmendation algorith m.
using a normalized popular score, and the K-Nearest For example, you could wish to apply a log transformation
Neighbors approach is used to increase accuracy. to numerical characteristics that are skewed.
KNN has several advantages, including its G. Feature Engineering
simp licity, flexib ility, and capability to deal with inter and In order to optimize the reco mmendation model,
continuous dependent variable. The simp licity of NB, as you may wish to generate new features by merg ing or
well as its capability to handle missing data and highly altering current features.
dimensional data, are among its advantages. SVM's
advantages include its capability to process high - These are just a few examp les of pre-processing
dimensional data as well as its fluid ity via the use of various processes to think about while developing a movie
kernel functions. recommendation system with SVMs, KNN, or NB. The
particular preparation procedures you'll need to perform will
III. DAT A ENGINEERING get to optimized for the reco mmendation algorithm you've
chosen.
Movie Database management system, projection IV. A NALYSING THE M ODEL
methods, user option analysis in cognitive science all have
roots in recommender system research. In the mid-1990s,
recommender systems developed as a separate study field. In the study, a dataset was utilised to develop the
The basic goal of Sentiment Classification is to research mutual framework, which is emp loyed to recommend titles.
users' behavior in order to learn their p references and The Sentiment Classification algorithms K-Nearest
interests. Existing Reco mmender Systems are classified into Neighbors (KNN), Naive Bayes (NB), and Support Vector
seven fundamental types and their sub - classes, which we Machine (SVM) were then trained on another dataset.
will look at in depth in this section. A. K – Nearest Neighbors Model:
A. Data Cleaning
Before you start creating your recommendation A straightforward supervised machine learning
model, ensure sure your dataset is clean and free of mistakes technique for classification and regression is K-Nearest
or inconsistencies. This might include dealing with missing Neighbors (KNN). Based on the rat ings that other users
numbers, fixing mistakes, and deleting duplicates. have given a movie, KNN can be used in the context of
movie reco mmendation to estimate the result that a user
B. Feature Selection would give to that film.
You must determine which features will be used as Calcu lating the distance between the movie for
input to your recommendation model. This might include which we want to forecast the rating and every other movie
movie -specific informat ion (e.g., genre, year of release, in the dataset is the fundamental concept underpinning
actors, etc.) as well as user-specific features (e.g., age, KNN. Then, in order to predict the rating for the movie in
gender, ratings history). question, we choose the K movies that are the most similar
C. Data Filtering to the one in question. We do this by looking at the ratings
that users have given to those films.
In the actual world, data obtained via explicit
feedbacks such as movie ratings might be quite scarce, with The d istance between movies can be calculated in a
most data points coming fro m really popular goods (movies) few different ways, such as utilizing cosine similarity or the
and highly engaged people. A large nu mber of lesser-known Euclidean distance. Cosine similarity is measured as the dot
products (movies) have no rat ings at all. Let's plot the product of the ratings vectors of the two movies divided by
frequency distribution of movie ratings. the product of their magnitudes, whereas Euclidean distance
is determined as the square root of the sum of the squares of
D. Normalization the variances between the ratings of the two movies.
Normalizing your characteristics so that they are all
on the same scale may be beneficial. This is especially ∑ ) ) ) )
crucial when utilizing algorith ms that are sensitive to the
magnitude of the input characteristics, such as KNN o r √ ∑ ) ∑ )
SVMs. It can be achieved by figuring the functional
dependencies between the columns in a table and splitting it
up into smaller tables based on these dependencies . where x is the average of x (p, q), t (p, q) is the
dendro grammatic d istance in between ith and jth
E. Train / Test Split: observation, and t is the average of t (p, q).
To analy ze the performance of your
recommendation model, div ide your dataset into training Fundamentally, M inko wski distance is a distance
and testing sets for the algorithms . metric that could be emp loyed in reco mmendation systems
to determine the similarity of two individuals or two
products. It is characterized as the 5th root of the sum o f the
absolute differences of their coordinates increas ed to the
power of p, and is a generalizat ion of both the Euclidean and

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 891

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

Manhattan distances. When p equals 1, the Minkowski

distance has now become the Manhattan distance, and when ∑ ) ̅)
p equals 2, the Minkowski distance has become Euclidean ̂ ̅
∑ )
distance.

dist(p, q) = ∑r =1d|pr qr|f)1/f Where ̅ is the mean rating of user p.

Once the relationship in between movie and all Like this , the KNN model performance with the
other movies has been discovered, we may choose the K different datasets to optimise.
movies that are the closest equivalent to the one that is
query and apply user ratings for those movies to forecast the
rating for the movie in question. B. Naïve Bayes Model (NB):
We can either weight the ratings by the distance
between the mov ies, so that movies that are closer to the One of the supervised learning technique is known
movie in question have a stronger influence on the forecast, as naive Bayes algorithm and it works on the principle o f
or we can simply average the rat ings of the K nearest Bayes Theorem.
neighbor to estimate a movie's rating.
It is mostly involved in deep learning with a large
Using the ratings that other users have given to training set. Being a classification algorith m, it makes
movie A and other movies, for instance, we can estimate the predictions on the probability that an item will occur.
rating that a user will g ive movie A. To compute the
prediction, When dealing with large amounts of data, the naive
Bayesian classification technique has highly accurate
we can use the formula shown below: results, and the algorithm is primarily used to predict
unrated data. The trouble of statistical likelihood
) classification is the inspiration for Naive Bayes.
=
) The items to be classified are chosen to give under
the condition that the subscriber feature items are satisfied,
and even the probability of an event of every segment there
where weight i is the weight we wish to assign to under afflict ion of that kind of event that occurs is obtained,
that rating, which is established according to the distance then in order to achieve the classification items are
between movie A and also its ith nearest neighbor, and rating classified through into categories with a high possibility o f
i is the rating that movie A received fro m its ith nearest people who seek, and are also used as the result of
neighbor. prediction.
A co mmon method for creating a reco mmendation For examp le, based on historical data of a specific
system is the K-Nearest Neighbors (KNN) approach. A user watching a movie in a movie theatre or on the Internet,
sluggish, non-parametric classification algorith m, the KNN it is discovered that five movies appease the circumstance
approach trains the dataset whenever a prediction is there under presumption of acco mplishing adverse
required. The Psychometric o r cosine similarity parameters of something like the user, and the first movie
measurements are used in the KNN technique to identify the has the highest probability of an event, recommend the first
correlation users or items. movie with the highest chance of occurring to this
destination use, depending on the Naive Bayesian algorithm.
∑
) In the large number theorem, if the sample data are
√∑ ∑ independent of each other and susceptible to the identical
allocation of condit ional probability events, the recurren ce
Where rth is the user x-provided rating for item y. of the happening can sometimes be used rather than the
probability of its occurrence.
The model's performance is dependent on
determining the most optimal value o f k; too large a value Furthermore, under the circu mstance of naive
would be too prohib itively costly and support vector Bayesian theory, that seems to be, the quantities amongst
regression, resulting in poor prediction performance. A each feature of the specimen are irrespective of each other.
value for 'k' that is too small results in premature The challenge is focused on calculating the
convergence and poor performance on different datasets. To conditional probability P(A|B). Because the random samp le
figure out the ideal size of k, we g raph the accuracies for a data A does not comprise a single pendant there in large
reasonable period of k and select the optimu m k as the one dataset, the whole report includes the Movie reviews
that produces the lowest error rate. statistical model, A exemp lifies the user's relevant data, and
Following the specification of k, we co mpute the indeed the user's necessary data includes not only the name
expected overall score for target consumer I on item j using character trait, but also data such as gender, age,
their arithmetic mean. emp loyment, and favorite movie type, that seems to be, A
embodies a plurality of data attributes and s ignifies a user's
attribute vector. In order to calcu late the prediction erro r, the
sample must contain all possible permutations of

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 892

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

characteristics. If there are no attribute criteria that are

independent of one another, all correlation values are zero.
C. Support Vector Machine (SVM):
Indiv idual attribute values are independent of one
another in this formula. SVMs are a sort of Supervised Machine Learning
technique that may be used for classification and regression
applications. In the case of movie reco mmendation, an SVM
) ) ) might be trained to anticipate whether a user would
) ∏ ) appreciate a specific mov ie based on factors like as genre,
) )
cast, or director.
The SVM used to train the data for the Movie
Reco mmendation System. This training procedure involves
 Performance of Evaluation Index: tweaking SVM parameters such as kernel type and
regularizat ion term to attain the best efficiency on training
A variety of recommendation algorithms, such as data. Once trained, the SVM could be utilized to determine
search engine recommendation, collaborative filtering-based whether a new, previously unseen movie would be en joyed
selection, or naive Bayes -based recommendation system by a certain user.
emp loying ML, can be applied in the customized The "kernel technique," which entails transferring
recommendation system. The performance of the network the user input data so that it becomes linearly separable, is
will not be the same for the classification method one method for mathematically representing an SVM. This
implemented in various algorith ms, as will the is accomplished by the use of a kernel function, which
recommended results and the system operating efficiency. might be a linear separable. The kernel used will be
The system's performance must then be analyzed determined by the characteristics of the data and the
in order to pick the suggested algorith m that is appropriate intended model complexity.
for the system, so considerably imp roving the efficiency and Assume we do have dataset of movies, each of
accuracy of the suggestion. In this study, the system's which is represented by a feature vector x Rn, where n
performance is primarily assessed using four evaluation indicates the number of variables (e.g., genre, cast, director).
indices: Accuracy, Precision, Recall Rate, and F1-score. We also include the binary label y 0-1 that indicates if the
The t wo-dimensional mult iple regression is the user enjoyed or hated each movie. Following that, the SVM
most generally used assessment model pred iction ability model is established as follows:
prior to the evaluation index co mputation, in accordance
with the general approach of data mining theory. f(x) = w ^Tx + b

TABLE I. ACTUAL CLASS AND THE PREDICTION CLASS

POSITIVE NEGATIVE
where w Rn denotes the weight matrix and b R
denotes the bias term. The SVM searches the higher
POSITIVE T rue Positive False Negative
(T P) (FN) dimensional space hyperplane that best separates positive
NEGATIVE False Positive T rue Negative and negative samples, as specified by the labels p.
(FP) (T N)
We may use a technique called "big marg in
classification" to optimize the SVM model, which entails
Precision = maximizing the range in between hyperplane and the closest
)
instances (called the "support vectors"). This is
Negative Prediction Value = accomplished by dealing with the fo llo wing optimization
)
method:
Recall =
)

Specificity =
)
Minimize ||weight||^2 subject p a (w ^T a x + b) ≥ 1 for
all x
Accuracy =
)

where ||weight|| is the weight vector's Euclidean rule

Positive and negative categorization accuracy did and a x is the mult idimensional array for the a-th movie. The
not attain comparable levels, but instead differed in linearization co mponent refers to the desired function
comparison with the original research. As a result, the ||weight||2, which assists in avoiding overfitt ing by
average categorization accuracy suffers. Even with the café punishing models with excessive weights .
and feedback review that serves as a specific topic in this
assertion, when viewpoint analysis is executed using a After training the SVM, we can utilize it to predict
conventional supervised Machine learning algorith m such as whether a new mov ie x new would be liked by a certain user
SVM and NB, the mean appro ximate rate decreases by by computing the function f(x new). If f(x new) is more than
demonstrating a variation classification. Thus, in this thesis, zero, we anticipate that the user will enjoy the film; if f(x
it enhanced NB algorithms are presented to reduce the gap new) is less than zero, we estimate that perhaps the user will
for the approximation rate of classifications and its mean. despise the film.

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 893

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

Then, in order of importance of resemblance, sort these

webpages. Choose the best number o f webpages and
Hierarchy Method for Recommendation using SVM: promote them to the user.
The suggested method is separated into four steps:
FINDINGS A ND RESULT S
 Create a One-Class SVM model:
The motive of this paper is to identify the most
Construct a 1-Class SVM user preferences schema reliable algorith m to integrate with our models so that our
for user utilising webpages set Q = {q 1 , q 2 , q 3 , …q z} that outcome to be optimized solution. We have conducted a
have already been read. Try the optimization method below: research regarding the work procedure and limitations for
each algorith m for wh ich we considered, fro m various
sources and gained various insights od training data sets for
‖ ‖ ∑ the algorithms of Machine learning.
There is a process to assemble the information of
SVM algorith m for testing the data. First, data is collected
⟨ ⟩ and process it after processing train the SVM model and test
the data. If error arises when training the model again find
Wu as the vector of attributes for the user. the error and test the data for retriev ing better results the
algorithm.
The flow chart describes a bout the KNN, NB and
 Determining the similarity in between the user model SVM algorith ms wh ich the training datasets gives the
and each domain model: outputs in a format to give the Optimized solution for the
Movie Recommendation System as illustrated in the fig[1].
Websites in the domain are represented as P(i) = {p 1(i),
p 2(i),…..p i(i)}. And l1+l2+l3+….+l k = n. n is the entire
websites we have. Let's say that we having k domains and
subdomains. The domains collection can indeed be written as Start
D = {d 1, d 2, d 3, ….d k} Create a One-Class SVM domain
model on every domain. Try the optimization problem
below:
Data Collection

‖ ‖ ∑

Data Pre-Processing
⟨ ) ⟩
Accuracy

 Domains with similarities and compute the relevance: SVM Prediction Model

The identity in between domain Di and user might be Testing the data
measured as:
Train SVM Model

)
‖ ‖‖ ‖
Completed testing
data
 Choosing webpages with identical and for the user
preferences:

Sort the user preferences similarities in an order. End

Choose the top number of do mains that most closely match
the user's interests. Sort the commonalities in the domain's
websites and user interest. The similarity may be co mputed Fig[1]: Flow Chart
as follows:
) The main goal for the article is that SVM is the
viability of Movie Reco mmendation System for the best
algorith m. A useful method for the Movie Reco mmendation
System is Support Vector Machine which has the highest
accuracy among the K-Nearest Neighbor Algorith m, Naïve

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 894

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

Bayes Algorith m and the Support Vector Machine It should be noted that the survey results may differ
Algorith m of having of 94.68%. Which the table and the based on the data utilised and the exact execution of the
graph shows about the accuracy, suspicious, legitimate, algorith ms. To g ive mo re accurate suggestions, it may also
phishy and average that ultimately describes about the be advantageous to examine other algorith ms or aggregate
Supervised learning Models of Machine learning as shown the findings of different algorithms.
in the fig[2], fig[3] and fig[4].
A reco mmendation including the size and
complexity of the dataset, the unique needs of the
recommended task, and the computational resources
available. It is crit ical to thoroughly assess the performance
of several algorith ms on your unique dataset to discover
which one performs best.
P HISHING ASSESSMENT RESULTS
Additionally, it is worth considering other factors
S. No: Values NB KNN SVM such as the scalability and interpretability of the algorithms,
1 Suspicious 52% 56% 21% as well as their computational requirements and runtime,
2 Legitimate 28% 35% 87%
3 Phishy 64% 78% 75%
when selecting an algorithm for movie recommendation.
4 Average 45% 65% 69% Consequently, despite SVM appears on the surface
to be the right alternative based on the results of this survey,
Fig[2]: T abular Form before making a final selection, it is critical to thoroughly
consider the individual criteria and limits of each movie
recommendation situation.
100 Having said that, SVMs may be a viable choice for
movie suggestion in some cases. If you have a huge and
80 complicated dataset with numerous characteristics, for
60 examp le, an SVM may be able to train a more accurate
decision boundary than a simpler method like KNN or NB.
40 SVMs, on the other hand, may be more co mputationally
intensive and need more time and resources to train,
20 requiring you to evaluate these factors against the
0
requirements of Movie recommendation System.
1 2 3 4 Future Work:
Naïve Bayes KNN SVM A movie reco mmendation system might be
improved by using a broader range of data sources. It may be
improved by tailoring suggestions further. There are several
Fig[3]: Graph algorithms that might be used to generate suggestions.
Collaborative filtering, matrix factorizat ion, and deep
learning are three of the most successful ways. The
100 recommendation system might be improved by adding user
80 input and making it more user-friendly by upgrading the user
interface.
60
REFERENCES
40
20 [1] Katarya, R., & Verma, O. P. (2017). An effective collaborative movie
recommender system with cuckoo search. Egyptian Informatics
0 Journal, 18(2), 105-112.
1 2 3 4 [2] Nassar, N., Jafar, A., & Rahhal, Y. (2020). A novel deep multi-
criteria collaborative filtering model for recommendation system.
Knowledge-Based Systems, 187, 104811.
NB KNN SVM
[3] Pavitha, N., Pungliya, V., Raut, A., Bhonsle, R., Purohit, A., Patel, A.,
& Shashidhar, R. (2022). Movie Recommendation and Sentiment
Analysis Using Machine Learning. Global T ransitions Proceedings.
Fig[4]: Line Graph
[4] Zhang, C., Duan, X., Liu, F., Li, X., & Liu, S. (2022). Three-way
naive Bayesian collaborative filtering recommendation model for
smart city. Sustainable Cities and Society, 76, 103373.
CONCLUSION [5] Kang, H., Yoo, S. J., & Han, D. (2012). Senti-lexicon and improved
Based on the survey statistics, the support vector Naïve Bayes algorithms for sentiment analysis of restaurant reviews.
mach ine (SVM) method outperforms the k-nearest Expert Systems with Applications, 39(5), 6000-6010.
neighbour (KNN) and Naïve Bayes(NB) algorith ms for [6] Gonzales, R. M. D., & Hargreaves, C. A. (2022). How can we use
movie selection. In regards to accuracy, precision, recall, artificial intelligence for stock recommendation and risk
management? A proposed decision support system. International
and F1 score, SVM tends to exceed the other two Journal of Information Management Data Insights, 2(2), 100130.
algorith ms, indicating that it is mo re successful at [7] Wang, Z., Ren, G., Sun, M., Ren, J., & Jin, J. S. (2015). Ranking
recommending movies to viewers. highlight level of movie clips: A template based adaptive kernel SVM
method. Journal of Visual Languages & Computing, 27, 49-59..

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 895

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

[8] Min, S. H., & Han, I. (2005, July). Recommender systems using [17] Peling, I. B. A., Arnawan, I. N., Arthawan, I. P. A., & Janardana, I. G.
support vector machines. In International Conference on Web N. (2017). Implementation of Data Mining T o Predict Period of
Engineering (pp. 387-393). Springer, Berlin, Heidelberg. Students Study Using Naive Bayes Algorithm. Int. J. Eng. Emerg.
[9] Li, H., Cui, J., Shen, B., & Ma, J. (2016). An intelligent movie T echnol, 2(1), 53.
recommendation system through group-level sentiment analysis in [18] Rahat, A. M., Kahir, A., & Masum, A. K. M. (2019, November).
microblogs. Neurocomputing, 210, 164-173. Comparison of Naive Bayes and SVM Algorithm based on sentiment
[10] Cui, L., & Shi, Y. (2014). A method based on one-class SVM for analysis using review dataset. In 2019 8th International Conference
news recommendation. Procedia Computer Science, 31, 281-290. System Modeling and Advancement in Research Trends (SMART)
(pp. 266-270). IEEE.
[11] Wu, C. S. M., Garg, D., & Bhandary, U. (2018, November). Movie
recommendation system using collaborative filtering. In 2018 IEEE [19] Pop, I. (2006). An approach of the Naive Bayes classifier for the
9th International Conference on Software Engineering and Service document classification. General Mathematics, 14(4), 135-138.
Science (ICSESS) (pp. 11-15). IEEE. [20] Sathyadevan, S., Sarath, P. R., Athira, U., & Anjana, V. (2014,
[12] Phorasim, P., & Yu, L. (2017). Movies recommendation system using August). Improved document classification through enhanced Naive
collaborative filtering and k-means. International Journal of Advanced Bayes algorithm. In 2014 International Conference on Data Science &
Computer Research, 7(29), 52. Engineering (ICDSE) (pp. 100-104). IEEE.
[13] Ruotsalo, T ., Weber, S., & Gajos, K. Z. (2022). Active tag [21] Yang, W., Cui, X., Liu, J., Wang, Z., Zhu, W., & Wei, L. (2015,
recommendation for interactive entity search: Interaction October). User's interests-based movie recommendation in
effectiveness and retrieval performance. Information Processing & heterogeneous network. In 2015 International Conference on
Management, 59(2), 102856. Identification, Information, and Knowledge in the Internet of Things
(IIKI) (pp. 74-77). IEEE.
[14] Marjanović, M., Kovačević, M., Bajat, B., & Voženílek, V. (2011).
Landslide susceptibility assessment using SVM machine learning [22] Agrawal, S., & Jain, P. (2017, February). An improved approach for
algorithm. Engineering Geology, 123(3), 225-234. movie recommendation system. In 2017 International Conference on
I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I-SMAC) (pp.
[15] Pradhan, A. (2012). Support vector machine-a survey. International
336-342). IEEE.
Journal of Emerging Technology and Advanced Engineering, 2(8),
82-85. [23] Shreya, A., & Pooja, J. (2017). An improved approach for movie
recommendation system. In International conference on I-SMAC (IoT
[16] Feng, W., Sun, J., Zhang, L., Cao, C., & Yang, Q. (2016, December). in Social, Mobile, Analytics and Cloud).
A support vector machine based naive Bayes algorithm for spam
filtering. In 2016 IEEE 35th International Performance Computing [24] Roy, D., & Dutta, M. (2022). A systematic review and research
and Communications Conference (IPCCC) (pp. 1-8). IEEE. perspective on recommender systems. Journal of Big Data, 9(1), 59.

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 896