A Comparative Study of CF And NCF In Children's Book Recommender System
Bojuan Niu, Jun Ma, Zezhou Yang · 2021
Recommender system has played a pivotal role in various fields and scenarios, but there are rare recommender systems for children's books aged 0–12 in China. In this paper, a deep learning named as the Neural Collaborative Filtering (shortly called NCF), is used to predict the list of recommended books for children. By comparing with the other traditional recommender algorithms, such as User-based collaborative filtering (briefly named User-CF) and Item-based collaborative filtering (briefly named Item-CF), it is found that NCF is more suitable for the recommendation of children's books than other methods. NCF has a higher accuracy than others. Through the experiment in this paper, NCF in Hit Ratio (HR) is 0.528 and 0.475 higher than User-CF and Item-CF respectively, in Normalized Discounted Cumulative Gain (NDCG) is 0.543 and 0.473 higher than the last two respectively and in Mean Average Precision (MAP) is 0.550 and 0.475 higher than the last two respectively. Therefore, among the recommender systems for children's books, the NCF model based on deep learning is fitted for the recommended scenes for children's reading, it could be optimized in the next step in further.