Research On Image Automatic Comment Method Based On Attention Automatic Coding Mechanism

With the popularity of e-commerce,online tourism and online shopping are also booming.More and more consumers comment on products or services through the Internet based on their consumption experience.The images for online travel or shopping sites,on the site are crucial.Merchants need to infer users' preferences and shopping psychology by learning the comments posted by users,and then build user portraits Under this background,this paper proposes an automatic image review model to automatically conduct image reviews.In the real scene,the judgment of users on the image is highly subjective,it is difficult to form a universal standard.Therefore,this paper uses the deep learning method to mine image attributes that affect user judgment,and find a method how to effectively integrate these attributes to evaluate the image.The automatic image review algorithm proposed in this paper is an intercrossed research domain of computer vision,natural language processing and machine learning.The model in this paper can simulate the user to automatically generate short comments on the image,so that the merchant can adjust the image according to the user's preference.The main work of this paper is as follows:1.This paper proposes the image short comment model for the first time benefits from the method and model of image description.Refer to the encoder-decoder model algorithm,this paper adds the Attention automatic coding mechanism to form the image automatic comment model.The model is trained and parameterized with the image comment data set produced in this paper,and short comments are generated automatically for the photos.Meanwhile,the model was adjusted to achieve the best effect based on evaluation indexes such as BLEU and CIDEr.2.The images and commentary data are crawled from the Fengniao photo Forum and 58 automobile website,and then the collected data were cleaned and screened according to error images and comments in each post.Finally,high frequency words are deleted according to word frequency statistics,images and comments are corresponded,and image annotation is made to form image comment data set according to the professional vocabulary.