Research On Collaborative Filtering Recommendation Algorithm For Fusion Item Visual Features

With the rapid development of the Internet,information on the Internet has also shown a rapid growth trend,and the problem of information overload has become more serious.The recommendation system has become an important solution to the problem of information overload.Collaborative Filtering(CF)is one of the mainstream models,and has been widely applied in academia and industry.Collaborative filtering models the user's preference function for the item from the user's implicit feedback on the item(eg,item purchase history,browsing log,etc.)to generate the final recommendation result,while ignoring the item's related attribute information,such as the item.The visual characteristics,the labeling of the items,etc.,thus limiting the performance of the model to some extent.This thesis considers that the visual features of items have an important impact on user preference modeling.To this end,this thesis proposes a collaborative filtering recommendation algorithm that fuses the visual features of items,and combines the visual features of the items with the user-item interaction history to jointly model the user-item interaction function.This thesis has carried on the in-depth research on the problem of the visual features of the items.The main research work is as follows:(1)The traditional collaborative filtering recommendation algorithm only considers the user's interaction information with the item,thereby limiting the expression effect of the model.In this thesis,the visual features of the item are integrated into the collaborative filtering algorithm based on user-item interaction.The linear regression method is used to model the influence of the visual features of the item on the user's preference.The experimental results show that the effect of the model has been improved significantly and verified.Visual features do have a significant impact on user preferences.(2)Among many collaborative filtering recommendation algorithms,matrix factorization is the simplest and most effective method.However,the matrix factorization model models the linear user function of the key user-item interaction function using the vector inner product,which limits the expression of the model to some extent.To this end,based on incorporating the visual features into user modeling,this thesis considers using deep neural network to model the user-item interaction function to capture the nonlinear high-order interaction between users and items.The experimental results show that the proposed deep collaborative filtering recommendation model based on the framework of BPR can effectively extract the visual features of items and capture the nonlinear high-level interaction between users and items,thus achieving more general and accurate recommendations.