Research On Recommendation Algorithms With Item Images

With the advent of the era of big data,"information overload" has become an urgent problem to be solved.In the face of complex data,it takes a lot of time for users to find the information they are interested in,so personalized recommendation technology comes into being.According to the needs of the target users,the recommendation technology can effectively filter the information,alleviating the problem of “information overload” and gradually becoming an indispensable part of daily life.Collaborative filtering algorithm is the most commonly used and most effective algorithm in personalized recommendation,but the traditional collaborative filtering algorithm has problems such as sparse data,cold start,and poor scalability.Firstly,in view of the difficulty of extracting image features by traditional image feature extraction algorithm,this paper uses the advantage of convolutional neural network in deep learning to extract image features,and uses Vgg16 convolutional neural network to extract the features of project images and extract them to The characteristics are used to calculate the similarity of the project image.Secondly,in order to solve the problem of sparse data in collaborative filtering algorithm,consider using Slope One filling algorithm to fill data missing values,but the traditional Slope One data filling algorithm does not consider the characteristics of the project itself,and proposes a new weighted Slope One data.The filling algorithm introduces the project image similarity into the weighted Slope One algorithm formula,and obtains a new scoring prediction formula,which uses the new calculation formula to fill the data and effectively improve the reliability of the filled data.On the basis of the filling,the project-based collaborative filtering algorithm is used to improve the recommendation quality.Thirdly,in order to solve the problem of project similarity deviation obtained by project scoring in collaborative filtering algorithm,an improved project similarity algorithm is proposed.Because the collaborative filtering algorithm does not consider the attribute of the project itself,it is proposed to linearly combine the similarity of the project image with the traditional project-based project similarity to obtain a new project similarity calculation method.It is a good complement to the similarity calculation error caused by the sparseness of the scoring data,which effectively improves the accuracy of the recommendation.Comparing the various improved recommendation algorithms proposed in this paper with the traditional recommendation algorithms on the MovieLens movie dataset,the comparison results show that the improved algorithm proposed in this paper has higher accuracy in predicting scoring than the traditional algorithm.Improve the accuracy of the forecast and improve the quality of the recommendation.The improved algorithm proposed in this paper has a certain improvement effect on the recommendation accuracy of the movie,which can help users find the movie of their interest more effectively.Further,considering the images of different application scenarios,a convolutional neural network for extracting corresponding scene images is trained to improve the accuracy of image similarity calculation.