CONTENT-BOOSTED COLLABORATIVE FILTERING FOR RATING PREDICTION USING ARTIFICIAL NEURAL NETWORK

• Main Author: Rizki Hasanah, Nurlaili
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42812
• Institution: Institut Teknologi Bandung

Recommender system is a popular technology that answers the need of finding relevant information immediately. Collaborative filtering as one of the methods in recommender system has been widely used because of its various advantages. Still, this method has main problem related to data sparsity which causes lack of accuracy in the recommendation results. Content-boosted Collaborative Filtering (CBCF) which incorporates content-based prediction in the recommendation process is one hybrid approach that has been developed previously. Modifying this approach by utilizing Artificial Neural Network can be an effective alternative because of its ability to properly model relationships between features. This research tried to formulate an optimal method for dealing with data sparsity problem in recommender system, that is by using Multi-Layer Perceptron of ANN for feature-learning phase in CBCF. These features/characteristics are used as input for model prediction to fill the sparse matrix. In this research, the features utilized were derived from metadata of reviews information. The proposed method is then tested using Amazon and Yelp dataset. The experimental results show that the proposed CBCF with Artificial Neural Network method reduces errors in recommendation with difference of up to 0.243 in MAE and 0.279 in RMSE. It also improves the accuracy of ratings classification by 3.9 to 6.5% from pure Collaborative Filtering. In addition, the inability of identifying neighbors due to data sparsity was also addressed using this method with error value decreased by up to 0.332. Thus, the proposed method has been proven to be a new alternative with improved accuracy that can be universally used for recommendation.