DOI: 10.30865/mib.v6i4.4571
Netflix Movie Recommendation System Using Collaborative Filtering With K-Means Clustering Method on Twitter
Muhammad Tsaqif Muhadzdzib Ramadhan*, Erwin Budi Setiawan School of Computing, Informatics, Telkom University, Bandung, Indonesia
Email: 1muhtsaqif@student.telkomuniversity.ac.id, 2erwinbudisetiawan@telkomuniversity.ac.id Email Author Correspondence: muhtsaqif@student.telkomuniversity.ac.id
Abstract−Nowadays, the development of technology is very rapid, so watching movies at home has become a means of entertainment. Netflix is one of the platforms for watching movies and provides various movie titles. However, because of the many movie titles, it makes it difficult for users to determine the movie they want to watch. The solution to this problem is to provide a recommendation system that can provide movie recommendations to watch. Collaborative filtering is a method that exists in the recommendation system by providing recommendations based on the ratings given by other users. Collaborative filtering is divided into two, namely based on items (item-based) and based on users (user-based). Twitter is a social media used to write posts called tweets. For this system, tweets serve as data that will be processed into ratings. This research was conducted using k-means clustering with collaborative filtering and collaborative filtering only. By using a dataset obtained from Twitter by crawling data and added with ratings from IMDb, Rotten Tomatoes, and Metacritic. Which resulted in a dataset with 35 users, 785 movie titles, and 6184 reviews. Then preprocessing the data with text processing, polarity, and labeling.
And get the dataset that will be used for this experiment. The results of this research test show that k-means clustering with collaborative filtering gets the best results with the best prediction of 2.8466, getting an MAE value of 0.5029, and an RMSE value of 0.6354
Keywords: Collaborative Filtering; User-Based; Item-Based; K-Means Clustering; Recommendation System
1. INTRODUCTION
The development of technology at this time has developed very rapidly, including movies which are a means of entertainment media for the community. However, because there are too many movie titles that have been circulating, it is difficult for people to determine the movie they want. Currently, people not only watch movies through theaters but also online streaming platforms, namely Netflix. Netflix is one of the online streaming platforms founded in 1999 as an online video store, has become the most widely used, and is still a rapidly growing American online streaming provider specializing in video on demand [1].
Social media is a platform that gives a big impact on the development of technology, one of the most popular social media is Twitter. Twitter is one of the social media used by various groups, by writing tweets, Twitter users usually provide information, express, and express opinions on things that are happening such as movies [2].
A recommendation system is a system that can help to overcome information overload by providing specific recommendations for users and it is hoped that these recommendations can fulfill the wants and needs of users [3].
Recommendation systems have several methods, namely content-based, collaborative filtering, and hybrid based [4]. One of the recommendation system methods that will be used is collaborative filtering. Collaborative filtering is a recommendation system that connects each user with the same preference for an item such as a movie based on the rating given by the user [5]. Collaborative filtering is divided into two, namely user-based and item-based.
Collaborative filtering is the most successful and popular algorithm for recommendation systems, but it has poor accuracy and a long running time so clustering is needed to overcome these problems [6]. With the problems that exist in collaborative filtering, we will use one of the clusterings to overcome this with k-means clustering.
The research that will be researched is based on research that has been done before. The use of references has the aim that the author gets knowledge about what he wants to research in the study. Here are some references to the results of research that has been done.
Based on the research of Arwin Halim, etc. with the title “Sistem Rekomendasi Film menggunakan Bisecting K-Means dan Collaborative Filtering”. Showing the error rate on the recommendation system has been calculated using the average value of MAE combination of bisecting K-Means and user-based CF is 1.63, lower than the average value of MAE combination of bisecting K-Means and item-based. In addition to the recommended method, the distribution of rating values on the dataset also greatly affects the MAE value. This is also shown in clusters 11 and 17 with uneven distribution of rating values, which will result in a higher error value in the recommendation system [5].
Based on the research of Yessica Putri Santoso, etc. with the title “Implementasi Metode K-Means Clustering pada Sistem Rekomendasi Dosen Tetap Berdasarkan Penilaian Dosen”. Shows that of the 70 data tested 39 lecturer data can be recommended as worthy of being a permanent lecturer and 31 lecturer data that is not worthy of being recommended as a permanent lecturer with an accuracy calculation result of 55.67% so it can be concluded that the K-Means algorithm is not suitable for this case [7].
Then, based on the research of Mu'tashim Billah, etc. with the title “Penerapan Collaborative Filtering, PCA dan K-Means dalam Pembangunan Sistem Rekomendasi Film”. From the research that has been tested, a movie recommendation system with K-Means Clustering and User-Based Collaborative Filtering has been
developed using the Principal Component Analysis method, by reducing the time complexity which produces a value of 1.061282 and the level of accuracy in the recommendation results has been calculated with an average MMR value of 0.44533417402269865 [8].
Then, based on the research of Rishabh Ahuja, etc. with the title “Movie recommender system using k- means clustering and k-nearest neighbor”. The results that have been tested from this research are a movie recommendation system using K-Means Clustering and K-nearest Neighbor by dividing into 3 clusters, namely 2, 19, and 68. Get an RMSE value of 1.23154 in 68 clusters, get an RMSE value of 1.233 in 19 clusters, and get the best value at an RMSE of 1.081648 in 2 clusters [9].
Then, based on the research of Riyan Alfa Rizkie and Muhammad Fachrurrozi with the title “Sistem Rekomendasi Wisata Kuliner Kota Palembang Menggunakan Metode Collaborative Filtering”. This shows that collaborative filtering has a relevance value of 80% with a Mean Absolute Error value of 0.723948146 using 18 food data, 100 training data, and 10 testing data [10].
The purpose of the research to be carried out is to produce a Netflix movie recommendation system using the collaborative filtering method with k-means clustering and get a performance value on the Netflix movie recommendation system using the collaborative filtering method with k-means clustering. It is hoped that with the Netflix movie recommendation system, people can find movies that match their interests and get a more accurate and precise performance value.
The organizational structure of the research paper is as follows. Section 2 describes the methods that will be used in the research, Section 3 discusses data crawling, data preprocessing, methods to be used such as collaborative filtering and k-means clustering, and shows the performance results obtained, Section 4 discusses the conclusions of the research conducted.
2. RESEARCH METHODOLOGY
2.1 Research Stages
The system design that will be built on the movie recommendation system applies two different methods. The first is using collaborative filtering and the second combines k-means clustering with collaborative filtering.
Figure 1. Collaborative Filtering
Figure 2. K-Means Clustering with Collaborative Filtering 2.2 Crawling Data
In the data crawling process, Twitter was crawled using the snscrape python library. The crawled data is the result of a tweet review from each user who can be trusted in reviewing movies. The data has been crawled based on movie titles available on the Netflix online streaming platform. The movie titles that have been crawled were movie titles from 2005-2021. Data retrieved in the form of id_tweet, username, date, tweet, and movie title.
After obtaining data containing movie reviews of movie titles on Netflix, reviews that contained movie reviews were selected. Then the best 1 tweet review regarding the discussion of these movie titles was selected.
DOI: 10.30865/mib.v6i4.4571
After that, the data will be added with rating values from websites that only specifically review movies such as IMDb, Rotten Tomatoes, and Metacritic according to the movie title on Netflix. The results of the data crawling process get results like Table 1.
Table 1. Crawling Data User Movie Title Total Data
35 785 6184
2.3 Preprocessing Data
Preprocessing is the initial stage in processing to select words in tweets, to produce more concise words, by selecting and removing words that are not needed [11]. In this research, preprocessing carried out on data that was originally in the form of reviews on Twitter is converted into a rating from 1-5 which can be used as a recommendation system. In the process of converting tweets into ratings, several stages are carried out, namely Text Processing, Polarity, and Labeling.
Text processing is a step to get more structured data in the process of selecting text data. At this stage, text cleaning is carried out which still contains elements of punctuation, numbers, emoticons, URLs, and hashtags.
Polarity is the process of identifying a text by knowing how negative and positive it is. Using polarity can be useful for predicting sentences that have positive or negative phrases, for example "this is the best movie" then the word "best" has a positive context [12]. For this research using polarity with the library from TextBlob. This library can help the processed text data to be good at identifying the meaning of words. Then the text data is converted to -1 and 1. Text data that has a polarity value close to -1 means that the rating will be made between 0- 2.4, while for text data that has a polarity value close to 1, the rating will be made between 2.6-5, and text data that produces a polarity value of 0, the rating will be 2.5.
Labeling for this research process is to identify the polarity result data to be checked again whether it is in accordance with the existing rating context, which is the result of text data into a rating with a value of 0 to 5.
2.4 K-Means Clustering
K-means clustering is an unsupervised learning algorithm, usually used in data mining. K-Means Clustering has the goal of dividing the values in N data into K number of clusters where each data is a cluster with the closest average distance [13]. Data that has the same characteristics will be combined into one cluster, while data that has different characteristics will be combined with other clusters [7]. The steps taken in K-Means Clustering by determining the number of clusters, displaying the maximum of users and the maximum of movies per cluster, and calculating the best cluster with Euclidean distances. Euclidean distances is a formula to calculate the distance to each cluster point, can be calculated as formula 1 [7] :
𝐷(𝑖, 𝑗) = √(𝑋1𝑖 − 𝑌1𝑗)2+ (𝑋2𝑖 − 𝑌2𝑗)2+ ⋯ + (𝑋𝑘𝑖 − 𝑌𝑘𝑗)2 (1) 2.5 Collaborative Filtering
Collaborative filtering is one of the methods used in recommendation systems that are used based on interactions between users and stored items which will be used to create a recommendation system [14]. Collaborative filtering is divided into two types user-based collaborative filtering and item-based collaborative filtering. User-based collaborative filtering is a method that provides item recommendations by comparing all items across all users to obtain Top-N user similarity [15]. Meanwhile, item-based collaborative filtering is a method that provides item recommendations by looking for other items that have Top-N similarity to other items. [15]. To get Top-N user similarity and Top-N item similarity, we use the cosine similarity method. Cosine similarity is a measure used to determine the similarity between two items, systematically the cosine angle between two vectors in three dimensions. [16].
Figure 3. Cosine Similarity 2.6 Mean Absolute Error (MAE)
The accuracy of the recommendation system can be seen from the Mean Absolute Error (MAE) value. MAE is the average of the error which is the difference between the actual rating value and the predicted rating value which is then absolved [17]. The following is the calculation of MAE as formula 2:
MAE = ∑𝑁𝑢=1|𝑃𝑢,𝑖−𝑅𝑢,𝑖|
𝑁 (2)
2.7 Root Mean Square Error (RMSE)
The accuracy of the recommendation system can be seen from the Root Mean Square Error (RMSE) value. RMSE can be determined from the magnitude of the prediction error rate, where the smaller the RMSE value, the more accurate the prediction results [18]. The following is the calculation of RMSE as formula 3:
RMSE = √∑𝑖∈1(𝑌𝑟𝑒𝑓−𝑌𝑝𝑟𝑒𝑑)2
𝑛 (3)
3. RESULTS AND DISCUSSION
In this research we apply two modeling techniques that will be adjusted to our objectives, scenario 1 which contains the calculation of recommendations using Collaborative Filtering (CF) only, and for scenario 2 contains the calculation of recommendations using K-Means Clustering with Collaborative Filtering. Then from each scenario that has been carried out, we test or measure the errors generated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) to see how much the improvement has increased.
3.1 Dataset
Before performing the recommendation system, we crawled data on Twitter with 785 movie titles and 30 Twitter accounts using the snscrape library. Then we select 1 tweet review that contains movie reviews on each movie title. The results of selecting the best 1 review got 3134 tweet reviews. Table 2 shows the results of crawling data with the selection of 1 review tweet containing movie reviews.
Table 2. Crawling Data
Id_tweet Username Date Tweet Title
5640 CenayangFilm 2015-02-06 11:29:00+00:00
cu... JUPITER ASCENDING JELEK
Jupiter Ascending
….. ….. ….. ….. …..
6850 hafilova 2016-01-07
13:55:05+00:00
mewek sendiri nonton "Room",
bagus! Room
After crawling the data by selecting the best 1 review, the next step is to do preprocessing. By doing text processing, namely cleaning text that still contains elements of punctuation, numbers, emoticons, URLs, and hashtags. Table 3 shows the results of text processing.
Table 3. Text Processing
Id_tweet Username Date Tweet Title
5640 CenayangFilm 2015-02-06
11:29:00+00:00 cu jupiter ascending jelek Jupiter Ascending
….. ….. ….. ….. …..
6850 hafilova 2016-01-07
13:55:05+00:00
mewek sendiri nonton room
bagus Room
After doing text processing, the next step is to do polarity by making the value into -1 to 1, and changing the rating from 0 to 5. Table 4 shows the results of polarity and rating.
Table 4. Polarity
Id_tweet ….. Tweet Polarity Rating
5640 ….. cu jupiter ascending jelek -0.7 0.75
….. ….. ….. ….. …..
6850 ….. mewek sendiri nonton room bagus 0.7 4.25
After obtaining a dataset that has been polarized, the next step is to combine it with ratings from websites namely IMDb, Rotten Tomatoes, and Metacritic. From the merged data there are 6184 reviews, 35 usernames, and 791 Netflix movie titles. The following results of data merging are in table 5.
Table 5. Result of Data Merging
Username Film idUser idFilm Rating
HabisNontonFilm Gunpowder Milkshake 1 2 2.65
….. ….. ….. ….. …..
DOI: 10.30865/mib.v6i4.4571
Username Film idUser idFilm Rating
Metacritic User Score G-Force 35 791 2.25
After getting the latest dataset, the next step is create a 2-dimensional pivot table containing idUser, movie, and rating. Table 6 shows the results of the pivot table
Table 6. Pivot Table Film
#Alive ….. Zombieland idUser
1 2.50 ….. 0
….. ….. ….. …..
35 0 ….. 4.30
3.2 Collaborative Filtering
After creating the pivot table, normalization of the dataset will be carried out such as removing duplicate words and filling the value 0 in the column that has the value Nan. Then the results of normalization will be as in table 7.
Table 7. Dataset Normalization idUser
1 ….. 35 Film
#Alive -0.1746 ….. 0
….. ….. ….. …..
Zombieland 0 ….. 0.3005
After normalizing the dataset, the similarity value will be found using cosine similarity. Then the results of cosine similarity for user-based are in table 8 and for item-based are in table 9.
Table 8. Used Based Similarity idUser
1 2 ….. 35
idUser
1 1 -0.0150 ….. 0.0152 2 -0.0150 1 ….. 0.0920
….. ….. ….. ….. …..
35 0.0152 0.0920 ….. 1
Table 9. Item Based Similarity Film
#Alive #FriendButMarried 2 ….. Zombieland Film
#Alive 1 -0.0077 ….. 0.4231
#FriendButMarried 2 -0.0077 1 ….. -0.2572
….. ….. ….. ….. …..
Zombieland 0.4231 -0.2572 ….. 1
After getting the similarity value of each item and user, the next step is to find the top 10 of the item/film and find the top 10 of the user. For this research, we used the movie 'Twilight' and user '15', so the results that will be displayed for the movie 'Twilight' and user '15' are in table 10.
Table 10. Top 10 CF
NO Top 10 Items on the movie 'Twilight' Top 10 users on user '15' 1 The Twilight Saga: Breaking Dawn: Part 2 6
2 Green Lantern 20
3 The Next Three Days 34
4 Pottersville 27
5 6 Underground 9
6 The Smurfs 2 35
7 Cahaya Dari Timur Beta Maluku 31
8 Olympus Has Fallen 4
9 The Twilight Saga: Eclipse 32
NO Top 10 Items on the movie 'Twilight' Top 10 users on user '15'
10 The Last Letter From Your Lover 26
The next step is to calculate predictions using the weight sum equation from the similarity value process that was previously obtained with cosine similarity. Prediction calculations to get predictions of ratings that have been determined by movies and users that will be recommended to users who have not watched the movie. The results obtained in the prediction calculation with the movie 'Twilight' and user '15' amounted to 1.724021206968171.
After obtaining the results of the prediction calculation, the recommendations for items are displayed to the user with the highest similarity. The results that will be displayed for recommendations can be seen in table 11.
Table 11. Recommendation Result CF NO Hasil Rekomendasi film untuk user ‘15’
1 Marriage Story
Then, for the final step, measuring performance through errors using MAE and RMSE, where if the resulting value is close to 0, it means it is close to accurate. The results for collaborative alone get an MAE value of 2.762112297727458 and an RMSE value of 3.773938001358683.
3.3 K-Means Clustering
To get the optimal cluster, it is needed to determine it with the elbow method, where calculations will be made to get a comparison with the Sum of Square Error (SSE) of each cluster. Therefore, the number of clusters that will be taken based on the elbow position. Based on Figure 4, it can be concluded that the number of clusters is 6.
Figure 4. Elbow Method
After getting the optimal number of clusters, clustering is done on the dataset. Then see which cluster is the best, and the best cluster is in cluster 1. Then normalization will be carried out to fill in the value 0 in the column that has Nan's value. Can be seen the normalization results in table 12.
Table 12. Normalization on K-Means idUser
2 ….. 9 Film
Black Panther -0.1930 ….. 0
….. ….. ….. …..
Clash of the Titans 0.1532 ….. 0
After normalization, the similarity value will be found using Euclidean distances. Then the results of Euclidean distances on users are in table 13 and for item-based are in table 14.
Table 13. Euclidean Distances User idUser
2 3 ….. 9
idUser
2 0 25.215 ….. 21.400
3 25.215 0 …. 23.682
….. ….. ….. ….. …..
9 21.400 23.682 ….. 0
DOI: 10.30865/mib.v6i4.4571
Table 14. Euclidean Distances Item Film
Black Panther Thor: Ragnarok ….. Clash of the Titans Film
Black Panther 0 0.8863 ….. 0.7339
Thor: Ragnarok 0.8863 0 …. 0.5907
….. ….. ….. ….. …..
Clash of the Titans 0.7339 0.5907 ….. 0
The next step is to find the top 10 of items and find the top 10 of users. For this research we used the movie 'Twilight' and user '15’ was used, so the results that will be displayed for the movie 'Twilight' and user '15' are in table 15.
Table 15. Top 10 K-Means
NO Top 10 Items on the movie 'Twilight' Top 10 users on user '15' 1 Spider-Man: Into the Spider-Verse 14
2 The Theory of Everything 3
3 The Conjuring 2
4 Marriage Story 12
5 Lincoln 22
6 Train to Busan 1
7 Extraction 0
8 Cloud Atlas 5
9 The Social Network 19
10 Sinister 2 20
The next step is to calculate predictions using the weight sum equation from the similarity value process that was previously obtained with Euclidean distances. Prediction calculation to get a prediction rating that has been determined by the movie and user that will be recommended to users who have not watched the movie. The results obtained in the prediction calculation with the movie 'Twilight' and user '15' amounted to 2.8466656053555686.
After getting the results of the prediction calculation, display item/film recommendations to the user with the highest similarity. The results that will be displayed for recommendations can be seen in table 16.
Table 16. Recommendation Result K-Means NO Movie recommendation results for user '15'
1 The Irishman
2 Dilan 1991
3 Ant-Man and the Wasp
4 Five Feet Apart
Then, for the final step, measuring performance through error using MAE and RMSE, where if the resulting value is close to 0, it means it is close to accurate. The results for k-means clustering with collaborative filtering get an MAE value of 0.502972045212329 and an RMSE value of 0.6354298021241583.
4. CONCLUSION
Based on research that has been done by combining k-means clustering with collaborative filtering and collaborative filtering only, it is used for recommendation systems. By using a dataset obtained from Twitter by crawling data and added with ratings from IMDb, Rotten Tomatoes, and Metacritic. Which resulted in a dataset with 35 users, 785 movie titles, and 6184 reviews. Then preprocessing the data with text processing, polarity, and labeling. And get the dataset that will be used for this experiment. After that, testing the dataset by combining k- means clustering with collaborative filtering and collaborative filtering only. It was found that the rating prediction generated from k-means clustering with collaborative filtering has a greater result than collaborative filtering only, which is 2.8466. Then the MAE and RMSE values generated by k-means clustering with collaborative filtering are smaller with the resulting MAE value of 0.5029 and for the resulting RMSE of 0.6354 which can be interpreted as better than collaborative filtering only, because accuracy/performance can be seen from the average value of MAE and RMSE errors. If the value is closer to 0, the better the accuracy/performance obtained. Therefore, it can be concluded that k-means clustering with collaborative filtering has better results than collaborative filtering only.
Therefore, it is hoped that future research can improve the performance of the recommendation system with a
larger dataset. In addition, it may be possible to combine with other methods such as content-based or use other clustering algorithms to create a recommendation system.
REFERENCES
[1] G. Kumar, S. Rathod, and A. Laha, “Sentiment Analysis on Micro-Blogs,” SSRN Electron. J., vol. 4, no. 11, pp. 121–
126, 2021, doi: 10.2139/ssrn.3867142.
[2] L. R. Dharmawan, I. Arwani, and D. E. Ratnawati, “Analisis Sentimen pada Sosial Media Twitter Terhadap Layanan Sistem Informasi Akademik Mahasiswa Universitas Brawijaya dengan Metode K- Nearest Neighbor,” J. Pengemb.
Teknol. Inf. dan Ilmu Komput., vol. 4, no. 3, pp. 959–965, 2020, [Online]. Available: http://j-ptiik.ub.ac.id/index.php/j- ptiik/article/view/7099.
[3] V. L. Jaja, B. Susanto, and L. R. Sasongko, “Penerapan Metode Item-Based Collaborative Filtering Untuk Sistem Rekomendasi Data MovieLens,” d’CARTESIAN, vol. 9, no. 2, p. 78, 2020, doi: 10.35799/dc.9.2.2020.28274.
[4] P. G. Padti, K. Hegde, and P. Kumar, “Hybrid Movie Recommender System,” vol. 4, no. 7, pp. 311–314, 2021.
[5] A. Halim, H. Gohzali, D. M. Panjaitan, and I. Maulana, “Sistem Rekomendasi Film menggunakan Bisecting K-Means dan Collaborative Filtering,” Citisee, vol. 1, no. 3, pp. 37–41, 2017.
[6] P. Phorasim and L. Yu, “Movies recommendation system using collaborative filtering and k-means,” Int. J. Adv. Comput.
Res., vol. 7, no. 29, pp. 52–59, 2017, doi: 10.19101/IJACR.2017.729004.
[7] Y. P. Santoso, M. Marlina, and H. Agung, “Implementasi Metode K-Means Clustering pada Sistem Rekomendasi Dosen Tetap Berdasarkan Penilaian Dosen,” J. Inform. Univ. Pamulang, vol. 3, no. 4, p. 228, 2018, doi:
10.32493/informatika.v3i4.2133.
[8] M. Billah, M. A. Zartesya, D. S. Prasvita, S. Komp, and M. Kom, “Penerapan Collaborative Filtering , PCA dan K-Means dalam Pembangunan Sistem Rekomendasi Film,” no. April, pp. 579–587, 2021.
[9] R. Ahuja, A. Solanki, and A. Nayyar, “Movie recommender system using k-means clustering and k-nearest neighbor,”
Proc. 9th Int. Conf. Cloud Comput. Data Sci. Eng. Conflu. 2019, pp. 263–268, 2019, doi:
10.1109/CONFLUENCE.2019.8776969.
[10] R. A. Rizkie and M. Fachrurrozi, “Sistem Rekomendasi Wisata Kuliner Kota Palembang Menggunakan Metode Collaborative Filtering,” Generic, vol. 12, no. 1, pp. 1–3, 2020, [Online]. Available:
http://generic.ilkom.unsri.ac.id/index.php/generic/article/view/101.
[11] D. Darwis, E. S. Pratiwi, and A. F. O. Pasaribu, “Penerapan Algoritma Svm Untuk Analisis Sentimen Pada Data Twitter Komisi Pemberantasan Korupsi Republik Indonesia,” Edutic - Sci. J. Informatics Educ., vol. 7, no. 1, pp. 1–11, 2020, doi: 10.21107/edutic.v7i1.8779.
[12] A. P. Gopi, R. N. S. Jyothi, V. L. Narayana, and K. S. Sandeep, “Classification of tweets data based on polarity using improved RBF kernel of SVM,” Int. J. Inf. Technol., 2020, doi: 10.1007/s41870-019-00409-4.
[13] M. Garanayak, S. N. Mohanty, A. K. Jagadev, and S. Sahoo, “Recommender system using item based collaborative filtering (CF) and K-means,” Int. J. Knowledge-Based Intell. Eng. Syst., vol. 23, no. 2, pp. 93–101, 2019, doi:
10.3233/KES-190402.
[14] I. Yoshua and H. Bunyamin, “Pengimplementasian Sistem Rekomendasi Musik Dengan Metode Collaborative Filtering,”
J. Strateg. …, vol. 3, pp. 1–16, 2021, [Online]. Available:
https://www.strategi.it.maranatha.edu/index.php/strategi/article/view/220.
[15] G. Ferio, R. Intan, and S. Rostianingsih, “Sistem Rekomendasi Mata Kuliah Pilihan Menggunakan Metode User Based Collaborative Filtering Berbasis Algoritma Adjusted Cosine Similarity,” J. Infra, vol. 7, no. 1, pp. 1–7, 2019.
[16] S. Pawar, P. Patne, P. Ratanghayra, S. Dadhich, and S. Jaswal, “Movies Recommendation System using Cosine Similarity,” vol. 7, no. 4, pp. 342–346, 2022.
[17] A. Pamuji, “Sistem Rekomendasi Kredit Perumahan Rakyat Dengan Menggunakan Metode Collaborative Filtering,”
Fakt. Exacta, vol. 10, no. 1, pp. 1–9, 2017.
[18] A. N. Khusna, K. P. Delasano, and D. C. E. Saputra, “Penerapan User-Based Collaborative Filtering Algorithm,”
MATRIK J. Manajemen, Tek. Inform. dan Rekayasa Komput., vol. 20, no. 2, pp. 293–304, 2021, doi:
10.30812/matrik.v20i2.1124.
