Research On The Algorithm Of Clothes Matching Based On Deep Learning

In recent years,with the rapid development of mobile Internet and e-commerce,online shopping has become an important part of people's daily life.In practical application scenarios,when users buy a piece of clothing,often e-commerce platform will automatically recommend its collocation clothes.The quality of recommendation directly affects the user experience,and further affects the development of the ecommerce platform.The problem of clothing collocation has become a hot research issue in recent years.The existing clothing collocation models are usually learned by only utilizing clothing images,making it difficult to extract the clothing attributes such as gender,color,style and so on.However,these attributes can be obtained coveniently in the text description of clothing.Due to the latest advances in the text and language coding of the existing deep learning model,two clothes collocation models based on text description are proposed to solve this problem.Specifically,a clothes collocation model based on long short-term memory network and compatible matrix is first employed.The model uses long short-term memory network to model text information and learns collocation through compatible matrix.Based on this model,we then propose a refined model by fusing the image and text information.In this paper,three large-scale data sets of Amazon,Taobao and Polyvore were used to evaluate the performance of our proposed model.Specifically,we use the above data sets to construct matching and non-matching text pairs,so as to get positive and negative samples.The performance of the collocation model is evaluated by the accuracy of the classification results.The experimental results show that the two models proposed in this paper outperform the existing image based deep learning models and the text based traditional machine learning models.In the two models proposed in this paper,the clothing matching model based on the long short-term memory network and compatibility matrix is the best,with high collocation accuracy,which can be applied to the actual e-commerce platform.