Implementasi Convolutional Neural Network Dan Probabilistic Matrix Factorization Pada Sistem Rekomendasi Buku

Abstract

Abstrak
Sistem Rekomendasi dapat merekomendasikan buku pada user tertentu berdasarkan prediksi rating, isi
konten buku, ataupun metode lainnya. Banyak metode recommendation system yang digunakan seperti
Probabilistic Matrix Factorization, dimana konten yang sudah diberi rating akan sering
direkomendasikan. Namun pada Probabilistic Matrix Factorization memiliki kekurangan yaitu dalam
mengatasi data yang memiliki nilai rating yang jarang. Maka diperlukan suatu metode yang digunakan
untuk memahami konteks isi dari buku sehingga tidak hanya melihat dari rating saja namun dilihat juga
dari review suatu buku. Untuk mempelajari review maka diigunakan suatu metode yaitu Convolutional
Neural Network dengan cara memberikan suatu nilai vektor yang mengarah terhadap konteks buku
kepada Probabilistic Matrix Factorization suatu recommender system. Berdasarkan hasil pengujiannya,
metode tersebut dapat meningkatkan keakuratan data dengan MAE = 3,0114707. Sedangkan untuk
Probabilistic Matrix Factorization nilai MAE = 4,0185377. Dari nilai tersebut dapat dijelaskan bahwa
metode Convolutional Neural Network dan Probabilistic Matrix Factorization bekerja cukup baik untuk
data yang jarang memiliki rating..

Kata kunci : recommender system, Convolutional Neural Network, Probabilistic Matrix Factorization

Abstract
The Recommendation System can recommend books to certain users based on rating predictions, book
content, or other methods. Many system recommendation methods are used such as Probabilistic Matrix
Factorization, where content that has been rated will often be recommended. However, the Probabilistic
Matrix Factorization has the disadvantage of overcoming data that has a rare rating value. So we need a
method used to understand the context of the contents of the book so that it is not only seen from the
rating but also seen from a book review. To study the review, a method called Convolutional Neural
Network is used by giving a vector value that leads to the context of the book to the Probabilistic Matrix
Factorization of a recommender system. Based on the test results, this method can improve the accuracy of
the data with MAE = 3.0114707. As for the Probabilistic Matrix Factorization the MAE= 4.0185377. From
these values it can be explained that the Convolutional Neural Network and Probabilistic Matrix
Factorization methods work well enough for data that rarely has a rating.

Keywords: Recommender system, Probabilistic Matrix Factorization, Convolutional Neural Network