融合知识图谱的时尚推荐方法研究
III
Research on Fashion Recommendation Method
Combining Knowledge Graph
ABSTRACT
The development of Internet technology has been for more than several decades, and the era
of big data has gradually begun. How to analyze and discover the value of data has become an
important research topic. Under the sweep of the wave of informatization and intelligence, various
fields are seeking their own artificial intelligence. This paper will focus on the fashion e-commerce
field, starting from fashion data and combining with existing computer technology, to propose a
personalized fashion recommendation solution. The work and innovations of this paper are as
follows:
(1) A fashion data augmentation framework that combines deep learning and natural language
processing technology is proposed. In the recommendation field, richer information generally
provides more accurate recommendation services, and existing fashion e-commerce datasets
generally have quality problems. Therefore, this paper addresses this issue and proposes this
augmentation framework, which includes two submodules of data filtering and attribute
augmentation. Through this framework, the quality of dataset is improved, and the accuracy of
recommendations is also improved.
(2) Introduce knowledge graph technology to solve the cold start problem. Almost all
recommendation systems suffer from cold start problem. The common solution is to add side
information, such as context, social network and other information. This paper introduces
knowledge graph technology to solve this problem. Based on the characteristics of the dataset, this
paper establishes a user-product knowledge graph, and introduces a common graph algorithm to
generate a recommendation list, which effectively alleviates the cold start problem.
(3) Finally, based on the user distribution characteristics of the dataset, a differentiated
recommendation strategy is proposed. For inactive users, the algorithm is run in the knowledge
graph to generate a recommendation list; for active users with sufficient data, the factorization
machine technology is introduced in this paper, and the factorization machine and knowledge graph
are combined to generate a more accurate recommendation list. The experimental results show that
the recommendation strategy proposed in this paper has achieved good results.
The fashion data augmentation framework proposed in this paper is an integration of the
current cutting-edge technology, which has certain portability; and the establishment of user-product
knowledge graph is to explore the combination of knowledge graph and traditional recommendation
methods; Finally, the differentiated recommendation strategy based on user distribution
characteristics which combines the above two point, proposes a complete fashion e-commerce
solution, which provides a feasible idea for the exploration of fashion recommendation.