Intelligent Recommendation Algorithm of Consumer Electronics Products With Graph Embedding and Multi-Head Self-Attention in IoE
Lin Li, Libin Jia, Sattam Al Otaibi
Abstract
IoE connects and interacts with the widest range of people, data and things. With the progress of IoE and the popularity of mobile devices, the demand for consumer electronics is increasing. Various consumer electronic products emerge in an endless stream, and users often have no way to start when facing many products. Therefore, the product recommendation system came into being. This paper proposes an intelligent recommendation model for consumer electronics products (CEPR) in IoE. First, CEPR uses graph embedding to solve data sparsity and cold start. This introduces commodity auxiliary information, it uses random walk to model the commodities and obtain the embedding. Data sparsity can be addressed by transforming high-dimensional feature into low-dimensional feature. The auxiliary information can solve cold start. Secondly, CEPR introduces multi-head self-attention, which combines location coding to model the dependency relationship between users’ historical purchases of goods. AUGRU is embedded in CEPR to filter the evolution path of user interest and obtain accurate user interest representation. Finally, a comprehensive experiment is conducted for the CEPR. Experiments verify the effectiveness of applying this to consumer electronics products recommendation. CEPR achieves 77.2%/80.7% HR, 30.6%/32.1% Recall, and 10.6%/12.5% NDCG on two datasets, which surpass the compared methods.