Research On Personalized Product Recommendation Based On Visual Auxiliary Information

The Internet has provided people with convenient services,and people’s daily behavior on the Internet generates a large amount of data.How to use these data more efficiently and perform personalized information filtering is very important.As an effective information filtering tool for handling unstructured data,visual-aware recommendation systems are now becoming a major research hotspot in the recommendation field.Visual-aware recommendation system aims to use product image data as auxiliary information,and through deep learning technology processing and combining with user-commodity interaction data,it can provide users with recommendation services that match their visual preferences.At present,there are still some shortcomings in visual-aware recommender systems,such as: lack of modeling the structural information of user-item interaction behavior graph,not good at deep mining the hidden aesthetic features in images,and lack of diversity in recommended items.To address the above problems,this paper carries out targeted research,including enhancing the semantic and aesthetic representation of images,deep mining of user-item interaction behavior graph structure information,and making full use of multimodal data to enrich the potential feature representation of users and items,etc.,to realize personalized recommendations for users.The main work accomplished is as follows:An aesthetic feature-aware visual recommendation system is proposed that fuses image aesthetics and behavioral interaction structure embeddings(ABVR for short).ABVR uses the pre-trained Vit model to extract the high-level visual features of the image-semantic category features,uses the aesthetic extraction network to mine the middle-level aesthetic visual features in the image--the color,shapes and other features of the items,and uses the graph convolution neural network(GCN)module to learn the multi-layer graph structure embedding features of user item interaction graph nodes,and finally associates and fuses the three types of features to achieve aesthetically enhanced visual recommendations.Extensive experiments are conducted on two real datasets to verify the effectiveness of the ABVR model in improving visual recommendation performance.A visual aesthetic and content semantic enhanced visual-aware recommendation approach is proposed(ASEVR for short),which firstly incorporates the semantic features of visual content extracted by Vit network,the semantic features of product desription text extracted by Albert,and the aesthetic features extracted by Aesthetic network into the user preference calculation,at the same time,it also captures the intrinsic connection between modalities using cross-modal attention for visual and text features.And then a two-layer attention network is imposed to distinguish the differences in users’ preferences among visual,textual and aesthetical features,and a multi-layer perceptron is adopted to further fit the interaction between users and items.Finally,the experimental results conducted on three real-world datasets show that the proposed method can effectively improve the recommendation performance.