Aiming at the problems that the existing sequence recommendation algorithm is difficult to fully exploit the intrinsic relationship between sequences and the deeper relationship between items

a novel sequence recommendation algorithm is proposed. The algorithm is mainly divided into two main parts

namely item embedding and preference learning. Firstly

a new item embedding method is created based on knowledge graph to obtain an unified representation of each item. Then

through self-attention mechanism to learn the context information of sequence

the user’s personalized preference vector is obtained

and the next recommendation is generated. Contrast experiments are set up on the real datasets MovieLens and MovieTweetings

and the effectiveness of the algorithm is verified by comparison with advanced algorithms of the same type. The experimental results show that the proposed algorithm has a significant improvement in the recommended index HR and the average reciprocal MRR

with an average increase of 5. 62 % and a maximum increase of 7. 74 %.