Book2Vec: Representing Books in Vector Space Without Using the Contents

摘要

This paper presents book2vec, a neural network based embedding approach for creating book representations. In this work, a well-known method from natural language processing domain, namely word2vec, is applied to a dataset of the books read by different users from the Goodreads website. Unlike previous works, we use non-textual features, considering only the book IDs. We represent the books read by each user as a sentence where the books' IDs are the words in the sentences. The results show that this approach can find meaningful representation of the books.