Collaborative Recommendation System in Users of Anime Films

The recommendation system is one method to know the preference consumer by showing the potential object. This recommendation also helps the consumer gets the preference object. One of the popular objects in the recommendation is anime film. In this case, we conduct research to recommend anime films based on ratings of previously watched films. Collaborative filtering is a technique that consists of calculating similarities, predictions, and recommendations. This study is taken from dataset Kaggle which consists of 73,516 users and 12,294 anime. A user’s history will be matched with the whole user’s history with alternating least squares (ALS) method. The anime will be recommended based on that results. This method is expected to help millions of users find the desired anime.


Introduction
With the development of technology, all forms of activities can be done digitally. Some information in cyberspace more and more with the growth of the internet in real-time which records every information received. As a consequence, much information is uncertain. It makes consumers or users are hard to find their preference because the information can be irrelevant to them. The recommendation system aims to resolve this problem. with a recommendation system, one can quickly access relevant information without searching manually. Many websites today benefit from a recommendation system for promoting and selling their products such as movies, hotels, restaurants, food and so forth. The recommendation can be built based on user profiles social media or track history in online stores which knows the user's behavior.
The recommendation system helps users to determine items that might be used. In terms of anime and manga, anime and manga versions are increasingly appearing with an interesting story and series quality. this will make it difficult for users to determine which movie to watch first. Collaborative filtering is a technique that is often used because it only relies on ranking data from previous users to be able to predict recommended films [1] In this paper, we propose the use of a recommendation system using collaborative filtering (CF) to recommend anime. CF [2], which is one of the most popular techniques used wildly for recommendation systems. A simple recommendation system is used to measure the similarity between shows, users and helps to predict whether users will enjoy certain anime.

Related work
Many types of research about recommendations have been conducted using collaborative, content-based and hybrid both of them. Dev, in his research, uses a big data platform with MapReduce to develop a collaborative reconciliation system [3]. This system is able to reduce costs by removing the redundant computational process. Kumar [4] proposes a film recommendation system using collaborative filters that focus on the ratings given by users to provide recommendations. This allows the user to choose his choice from a given set of attributes and then recommends him to list the films based on the cumulative weights of the different attributes use algorithms k-means to recommend films based on the highest order. Shahjalal et al [5] implement a reconciliation system that combines both collaborative and userbased filtering approaches based on items. Researchers use the KUNN algorithm, a new algorithm for collaborative screening of one class, a setting that includes many applications. Table 1. Movie ratings by users.

Problem formulation
To know this problem clearly, table 1 is given as an example problem. It consists of five movies (P, Q, R, S, and T) and four users (A, B, C, and D). User A gives a rating for movies P, Q, S, and T with 5, 5, 1 and 1. User A likes movie P and Q very much with giving the highest score, 5. Contrarily, He/she doesn't like movies S and T with the lowest score 1. As a short, it shows that the more user like, the score will be more with 5 as the highest score. User A does not have a rating for movie R. It means user A has not watched this movie yet. With the same logic, Table 1 also shows the rating for the other users B, C and D for movies P, Q, R, S and T. The system is built to recommend for a user which movie should be watched by using the history data. It can be said the system will offer to each user which movies should be watched by sorting from the highest movie rating [6]. Figure 1 shows the process to build this recommendation system.

Mean normalization
For new users who are registered will not have preferences related to the film because it has not been recorded, because of the supported activities. So, it is quite difficult to use the required film. This is the initial part of the debate which is known as a cold start problem. To overcome this, it is calculated from the average value of each film calculated from the film rating. So that the film rating will be normalized with films that have an average of zero. this will increase efficiency much better.

Similarity matrices
Interactions on users and movies are expressed in a matrix ranking that can be transformed. Where the node corresponds to users and items, and an edge that connects users and items. so that structural steps can be used in the bipartite graph to identify the environment and then determine the nearest neighbor of the target user. SimRank is a well-known tool and proven effective in reducing sparsity problems. In this paper, we include SimRank for calculating the similarity of users in the collaborative filtering model.

3.4.Recommend top movies
The results of minimizing the X matrix and user ratings by using the parameter θ and films with the X feature will get the formula θ TX. The approach is to calculate the difference in features between films. so for small differences, it will be recommended for users. For example, small differences | Xi -Xj | between films i and j mean they are similar. In this study, Top Anime will be recommended based on the friendliness of the anime from the largest to the biggest difference.

Dataset
The film dataset used in our experiment was obtained from the anime recommendation database "Kaggle". Data provide two files namely anime data and rating on anime. This data consists of a relation between 73,516 users and 12,294 anime. Each user is able to add anime to their completed list and give it a rating and this data set is a compilation of those ratings.

Model evaluation
The collaborative filtering method is a method that guests the rating of the user's data based on the similarity history of the other user preference. Collaborative filtering is implemented by using matrix factorization [7].        Figure 9. Build the recommendation model using ALS.

Conclusion
A better anime recommendation system based only on user watch history. A simple recommendation system can measure the similarity between performances, users and help predict whether users will enjoy certain anime. However, due to limitations calculating very large data sets, 100k data sets selected. Using a much more powerful machine, if a 1 million data set could be experimenting with. This paper shows a very simple but efficient recommendation system.