Dtree2vec: A High-Accuracy and Dynamic Scheme for Real-Time Book Recommendation by Serialized Chapters and Local Fine-Grained Partitioning

With the rapid development of social computing technologies and online reading platforms, the proportion of e-books, especially online serialized novels, has been increasing. Identifying ways to add updated serialized books to readers' real-time recommendation lists has become an urgent problem to be solved. While serialized books are still underwritten and in the process of production, their features and categories can constantly evolve and change, lacking complete text content information and complete global scoring data. Therefore, this paper proposes a dynamic DTree2Vec scheme for serialized books that models text of varying degrees of completion to achieve unified semantic feature representation to measure the semantic relevance of books. At the same time, it ensures recommendation quality by tracking the update status of serialized books and by adding the subsequent chapter content in real time. This scheme establishes a dynamic hierarchical tree structure for serialized books and applies a cosine-type local reconstruction model to reconstruct new semantic features of books. In addition, the fine-grained factor of chapter partitioning is introduced to the reconstruction process to adjust the proportion of global and local features to better represent the semantic features of books. We analyzed effects of the content of serialized chapters on the recommendation results and used semantic features of the reconstruction to represent the content of serial novels in real time. The experiment results prove that the proposed DTree2Vec scheme can achieve higher degrees of recommendation accuracy when dealing with unfinished serialized books, effectively alleviating dynamic capture problems and real-time recommendations of book recommendations.