Research On Semantic Description Of Commodity Image Based On Target Word Vector

With the process of global economic integration and the development of the Internet open information network,E-commerce has become an important industry in the modern service industry.The accurate semantic description of the automatic product image has important application value to improve the retrieval precision and the user's shopping experience.At present,product information obtained by image semantic description based on image classification and target detection technology using depth learning is single,It has been difficult to deal with the diversity of today's e-commerce products and the variability of the characteristics of the elements.In the search engine based e-commerce platform information retrieval has also been greatly restricted.In recent years,the application and development of image description in the field of electronic commerce platform commodity retrieval has been paid attention to,and great progress has been made,but there are still many limitations.In order to meet the needs of variety and comprehensiveness of commodity information description,this paper proposes a semantic description algorithm of commodity image based on target word vector,aiming at the existing problems and the achievements of in-depth learning.The main research contents are as follows:1.The accurate classification of commodities and the feature extraction of the target region are the important prerequisites of commodity description.In this paper,the convolution neural network is chosen as the tool of image feature extraction,which lays a good foundation for the subsequent image feature acquisition.2.According to the characteristics of the title description of the store merchandise image,a large number of text description corpus is analyzed,and then the corresponding word vector language model is constructed to mine the text feature to obtain the fine-grained feature of the product.3.Using the method of feature association,the text feature and image feature are used to construct a more advanced semantic concept through association rules,and the semantic description of commodity image is formed.The information analysis process that combines text features improves the limitations and accuracy of the existing methods.The article compares the advantages and disadvantages of different word vector language models.To design the most suitable network model for text features and design the most reasonable association rules for feature associations.The network parameters are continuously optimized during the training process.The proposed model architecture can effectively generate a semantic description of the product image,which is obviously superior to the existing model in the evaluation criteria.The quantity and precision of commodity information describing the statement meet the needs of the actual application market.