Deep Interest Network Based Book Recommends-A Case Study of College Reader
Kui Zhang, Jianguang Hu, xing xin · 2023
Data providers widely use Recommendation Systems (RS) to facilitate users getting required knowledge in an information exploration context. Prediction based on item ranking or comments is usually the book recommendation task, in which lots of work has been proposed in a commercial context. However, there are no comments or even ranking data under the college library context. Moreover, the volume of the user dataset is vast and spans years. Different from the recommendation for classic bookselling, the recommendation for the college readers needs to meet their study progress and also follow their reading interesting drift. So, if the existing model can be applied in the library context and how these models performed under the context of the sparse datasets need to be verified. This study applied an interest-oriented ranking algorithm to the book recommendation for college students. Furthermore, a case study was carried out on the BeiHang University library dataset, which consists of 2.32 million user records. An experiment of six groups of different sequence lengths was conducted to validate the ability of DIN in interesting capturing. The result shows that the DIN can perform well in AUC, which on average achieved 76%. Meanwhile, the rapid decrease of the AUC according to the increasing number of sequence lengths also indicates the limitation of long-term interesting capturing.