Research On Recommendation Algorithms Based On User Interests

Today’s users are in an era of information explosion.All users are faced with a large amount of information,and it is very difficult to select relevant information that is closely related to them.It has also become more difficult for platforms to provide users with information that users are interested in.E-commerce platforms,community group buying platforms,especially video platforms,facing the increasing number of users and related data,in order to improve the efficiency of the platform,it is necessary to dig out effective information which is very important for personalized recommendation for users.Require.All video platforms hope to be able to grasp the dynamic information of each user in real time and implement more accurate personalized recommendations.Personalized recommendation can not only improve the efficiency of the platform,but also bring the ultimate experience to users.Since the traditional collaborative filtering algorithm evaluation system does not take into account the differences in the scores of different users,it still needs to be further improved in terms of accuracy,precision,and differences.There are three main research contents in this paper,which are how to divide the proportion of datasets in the recommendation algorithm to improve the traditional similarity calculation method,and integrate the neural network with the item-based recommendation algorithm and the SVD-based recommendation algorithm.The research on the recommendation algorithm based on user interest is carried out in this paper from the three perspectives of data set processing,similarity improvement,and fusion of two different recommendation algorithms.The detailed research points are as follows:(1)In the recommendation algorithm based on collaborative filtering,the data set is divided into a test set and a training set.The proportion of data set division will have a certain impact on the efficiency and accuracy of the recommendation algorithm.In the third chapter,the recommendation algorithm based on collaborative filtering is simulated under three different proportions,and three different proportions are obtained MAE through the simulation.According to the size of the MAE value to determine the ratio of the division of the data set and the test set in the next two chapters.(2)In the traditional recommendation algorithm,the original Pearson similarity calculation only considers the impact of the user’s rating on the similarity,while ignoring the role of the user’s evaluation system in the recommendation algorithm.Therefore,in order to meet the needs of precision in the collaborative filtering algorithm for movie rating recommendation,this paper introduces the user difference factor,improves the calculation method of Pearson similarity,and builds a simple evaluation system that can describe users.Therefore,in the similarity calculation of the traditional collaborative filtering algorithm,there is a certain deviation in the evaluation system for different users.For the experimental simulation under different similarity calculation methods,the experimental results obtained through the simulation show that the improved collaborative filtering algorithm for similarity calculation can reduce the MAE value.(3)In this paper,the neural network is integrated with the item-based recommendation algorithm and the SVD model,and the improved similarity calculation method is used.Compared with the previous algorithms,the recommendation algorithm based on the neural network-SVD model has a lower RMSE value,with a maximum drop of 3.99% and a minimum drop of 1.33%.And the recommendation algorithm based on neural network-item performs better in time efficiency.