Research On Improved Film Collaborative Recommendation Algorithm Driven By Multi-layer Preferences

With the rapid development of information technology and the increasingly rich internet application,people’s life has become intelligent and colorful,but at the same time,it also brings the problem of information overload to people.Due to the huge amount of data,people will spend a lot of time to find the relevant information accurately from the vast amount of information.Recommendation algorithm to a certain extent,solve the problem,in which collaborative filtering is one of the widely used recommendation algorithm;Through the analysis of user interest,found in the user group of designated users alike(interest);The synthetic evaluation of these similar users on a particular project,the formation of preference for the specified user projects to forecast;That is to use the preferences of a group with similar interests and common experiences to recommend information of interest to target users.But at the same time,there are also some defects: the quoted groups are group groups with direct "interaction",and the reference scope is too limited because the mass groups without direct "interaction" are not considered.At present,most of the improved recommendation algorithms are improved by filling the matrix with preference information in advance and then calculating the similarity;The information obtained between users or items is symmetric,without taking into account the differences between individual users,resulting in different preferences for items,and the recommendation results fail to meet the personalized needs of users.To improve the shortcomings of collaborative filtering algorithm,this thesis holds that the influence of mass groups leads to popular preferences in recommendation,while individual differences of users lead to personalized preferences in recommendation.Therefore,based on the preference perspective,this thesis proposes a new collaborative filtering recommendation algorithm combining the popular preference coefficient and the personalized preference coefficient.It can effectively mine for users some products that are difficult to be found spontaneously,but are in line with users’ preferences,and at the same time improve the accuracy of the algorithm.The main work of this thesis is as follows:(1)Recommended project is actually built on the basis of a certain degree of user preferences,put forward the collaborative filtering algorithm based on project recommended for similarity is the basis of reflecting the preferences.However,most studies use the method of personalized filling matrix in advance to find similarity,which will increase the objectivity of objects and distinguish objects;The similarity between multiple users may reduce the bias of a single user.Therefore,this thesis uses the method of first calculating similarity and then modifying preference information to generate recommendations.(2)Collaborative filtering algorithm is the referenced group group type groups,considering the human behavior is influenced by mass groups namely bandwagon effect;According to the film’s score,number of movies are popular user preferences.This thesis calculated all user history records containing the number of films and score,and with its results as a popular preference coefficient.The rapid development of the Internet lead to human taste is multifarious,differ in thousands ways,depending on the type of film calculation of film properties of personalized user preferences.By calculating the user view shadow records contained in the characteristics of various properties of the films,and take the total proportion of the category as the weight of users’ personalized preference for movie types,based on this,the personalized preference coefficient of users is calculated.Finally,the popularity preference coefficient is combined with the personalized preference coefficient to improve the collaborative filtering recommendation algorithm to meet the needs of users.(3)This thesis designs a detailed algorithm validation experiment.The test dataset is Movielens,a recommended test dataset widely used in academia.In the experiment,the precision,recall,lift,depth,FPR and other aspects were verified.The experimental results show that the precision of the hierarchical preference hybrid recommendation model is much higher than that of the classical recommendation algorithm,which proves the effectiveness of the proposed algorithm.