BERT and CNN integrated Neural Collaborative Filtering for Recommender Systems

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE BERT and CNN integrated Neural Collaborative Filtering for Recommender Systems Abdullah Al Munem Department of Computer Science and Engineering East West University Dhaka, Bangladesh abdullahalmunem@gmail.com Sumona Yeasmin Department of Computer Science and Engineering East West University Dhaka, Bangladesh sumonasumu930@gmail.com Mohammad Rezwanul Huq Department of Computer Science and Engineering East West University Dhaka, Bangladesh mrhuq@ewubd.edu Abstract— Every day, a significant number of users visit the internet for different needs. The owners of a website generate profits from the user interaction with the contents or items of the website. A robust recommendation system can increase user interaction with a website by recommending items according to the user's unique preferences. BERT and CNN-integrated neural collaborative filtering (NCF) have been proposed for the recommendation system in this experiment. The proposed model takes inputs from the user and item profile and finds the user's interest. This model can handle numeric, categorical, and image data to extract the latent features from the inputs. The model is trained and validated on a small sample of the MovieLens dataset for 25 epochs. The same dataset has been used to train and validate a simple NCF and a BERT-based NCF model and compared with the proposed model. The proposed model outperformed those two baseline models. The obtained result for the proposed model is 0.72 recall and 0.486 Hit Ratio @ 10 for 799 users on the MovieLens dataset. This experiment concludes that considering both categorical and image data can improve the performance of a recommendation system. Keywords—recommendation system, neural collaborative filtering, deep learning, cnn, bert, ncf I. INTRODUCTION The number of active users of the internet has been increasing. People generate a lot of data while browsing. To utilize the information, many online services, such as social media sites, streaming services, and e-commerce, have widely adopted recommender systems to increase user interaction and profit. Every user has some personal preferences, and recommending items based on user preferences can improve user experience and make them more engaged with the website. Effective recommendations can lead to significant profit increase [1,2]. The most commonly used approach for recommendation systems is Content-Based Filtering (CBF) and Collaborative Filtering (CF) [3]. Since CBF recommends items based on the item's features and the user's existing preferences, it cannot explore the users' new interests. CBF only considers item-item or user-user similarities for recommendations. To address the limitation of CBF, CF provides recommendations based on users’ similarities between items and users simultaneously. Thus, CF can explore new interests for a particular user. CF can consider both explicit and implicit feedback from the user. But the data generated from CF has a high sparsity because most of the users will not interact with most of the items. Thus, it is computationally inefficient and resource hungry. To address this limitation, neural network-based CF called neural collaborative filtering (NCF) has been introduced which embeds the user-item pair and represents the latent features using dense vectors [4]. In this study, a hybrid NCF has been proposed to make the recommendation system more efficient and effective. The proposed model can consider a wide range of data that can be generated from the items and by the users. Bidirectional Encoder Representations from Transformers (BERT) and Convolutional Neural Network (CNN) has been interacted with the proposed model to handle both categorical and image data. II. RELATED WORK Traditional methods such as content-based recommendation systems [5] and Collaborative Filtering (CF) [6] have been used as a recommendation system for several years. Due to the extensive success of deep neural networks in various fields, it has been adopted in recommendation systems. Neural Collaborative Filtering (NCF) has shown tremendous improvement in recommending items based on user’s unique preferences. Xiangnan et al. [4] proposed the NCF model, which outperforms the state-of-the-art models. Matrix Factorization (MF) is the most popular kind of collaborative filtering. Utilising a fixed inner product of the user-item matrix, user-item interactions are learned. Since calculating inner product is a computationally expensive task, NCF replaces the user-item inner product with a neural architecture. In order to learn about user-item interactions, NCF used a multi-layer perceptron and tried to articulate and generalize MF inside its framework. Since the proposed model of this study is developed based on the concept of NCF, this Related Works section only focused on NCF-based recommendation systems. Juarto

This content is AI-processed based on open access ArXiv data.