Kyoung Jun Lee., Yujeong Hwangbo., Hokyoung Jung., Baek Jeong., & Jong Il Park. (2022). TransformRec: User-Centric Recommender System for e-Commerce Using Transformer. 23rd International Center for Electronic Commerce.
Abstract
We propose a new User-Centric recommender system using Transformer model called
TransformRec, which uses receipt data without personal information and identity and
considers only the relationships between tokenized product names. TransformRec
recommends a product based on its most recent receipt, which includes product names.
Although a receipt includes a product that the Transformer has not learned,
TransformRec can recommend a real product that is considered as most relevant to the
user’s last purchase. We used two commercial datasets, an e-commerce dataset and
Instacart dataset, and compared the performances of TransformRec, TransformRec
without tokenizing, and Word2Vec. The experimental results demonstrated that the
performance of TransformRec is superior to that of the other two models. Thus, we
conclude that it is possible to recommend a product without using user identity or
demographic information with higher performances. In addition, we confirmed that
reflecting the relationship among tokens can improve recommendation performance.
Keywords: TransformRec, User-Centric, Recommender System, e-Commerce
AI & BM Lab 