Hierarchical Recommendation Algorithm Incorporated with Book Descriptions

As more and more abundant information integrated into the recommendation system, the recommended effect is getting better and better. But not all of these side information have positive influence. For the book recommendation system, the descriptions always play the role of the first window for users to quickly overview a book. These descriptions is a kind of side information containing rich and refined semantic information. Therefore, based on the book-crossing data set, we crawls the descriptions of 107552 books, and make further recommendation by using the metapath2vec++ algorithm and LDA(Latent Dirichlet Allocation) algorithm. At the same time, aiming at the problem that users who contribute more scores are difficult to make effective recommendation in VSM(vector space model) because of the traditional vector addition method, a user similarity calculation algorithm based on the wasserstein distance is proposed, and the recommendation is based on these similar users. Through experiments, the accuracy improved 16.3% and F1 score improved 22% among the users with more than 200 rating items.