Personalized Book Recommendation Based on a Deep Learning Model and Metadata

Reading books is one of the widely-adopted methods to obtain knowledge. Through reading books, one can obtain life-long knowledge and maintain them. Additionally, if multiple sources of information can be obtained from various books, then obtaining relevant books is desirable. This can be done by book recommendation. There are, however, a number of challenges in designing a book recommender system. One of the challenges is to suggest relevant books to users without accessing their actual content. Unlike websites or blogs, where the crawler can simply scrape the content and index the websites for web search, book contents cannot be accessed easily due to copyright laws. Because of this problem, we have considered using data such as book records, which contains various metadata of a book, including book description and headings. In this paper, we propose an elegant and simple solution to the book recommendation problem using a deep learning model and various metadata that can infer the content and the quality of books without utilizing the actual content. Metadata, which include Library Congress Subject Heading (LCSH), book description, user ratings and reviews, which are widely available on the Internet. Using these metadata are relatively simple compared to approaches adopted by existing book recommender systems, yet they provide essential and useful information of books.