A Movie Recommendation System That Supports Privacy Protection And Diversity Of Results

Big data has penetrated into every field of human life and is playing a more and more important role in the decision-makings of people.However,the inherent characteristics of big data,such as big volume and sparse value,usually mean high expenses and time cost in finding the valuable data from massive candidates.On the other hand,users’ expectations on the high efficiency and accuracy of big data applications require that a recommender system can make realtime and accurate big data analyses and decision-makings for the users.The big gap between reality and users’ expectations blocks the successful deployment and implementation of big data applications in people’s production and life activities severely.Considering this,personalized and lightweight recommendation technologies are proposed accordingly.By analyzing user preferences,a recommender system can make personalized recommendations to the users quickly and accurately,which reduces the users’ service selection cost and decision-making burden significantly.Due to the advantages of “domain-independence” and “easy-to-explain recommended results”,Collaborative Filtering(CF)has gradually become one of the most commonly used technologies in current recommender systems.However,existing CF-based recommendation approaches still face many problems and challenges introduced as follows:(1)The efficiency of traditional CF-based recommendation approaches is not high;especially when the scale of data used to make recommendations is large,the time cost of recommendations is too large to satisfy the quick response requirements from partial users.(2)Traditional CF-based recommendation approaches often assume that the decision-making data for recommendations are centralized,without considering the multi-source property of decision-making data as well as the privacy leakage risks during the multi-source data integration process.(3)Traditional CF-based recommendation approaches often pay more attention to the accuracy of recommended results,while ignore the diversity of the recommended results.Therefore,a recommender system is prone to produce repeated or redundant recommended results of low quality,which reduces the users’ satisfactions with the recommended results significantly.This thesis mainly studies the above mentioned three scientific problems and applies the research outcome to a movie recommender system.Concretely,the main research work of this paper is introduced as follows:(1)In order to cope with the high time expenses of traditional CF-based recommendation approaches,we build user indices offline based on Locality-Sensitive Hashing(LSH)technique.Afterwards,according to the user indices,we quickly search for the similar neighbors of a target user,through which the time complexity of neighbor searching is dropped from O(m*n)to O(1)(m and n denote the number of users and the number of services,respectively);consequently,the efficiency of subsequent service recommendations based on similar neighbors are improved significantly.(2)In order to solve the problem of user privacy leakage during the multi-source data integration process,we utilize the LSH technique to transform the sensitive decision-making data for recommendations(e.g.,Quality of Service(QoS))into user indices with little or no privacy.Afterwards,we recruit the user indices with little or no privacy to search for similar neighbors and recommend appropriate services accordingly;this way,the real QoS data are protected and correspondingly,the private information of users implicit in QoS data is guaranteed.(3)As to the repeated or redundant functionalities of recommended results,we measure the functionality differences or distances of different services based on the services’ functional tags.Afterwards,we filter the recommended service list again based on the functionality differences of candidate recommended services,so as to achieve the diversified service recommendation while guaranteeing recommendation accuracy and further increase the users’ satisfaction with the recommended results.(4)With the research outcome of(1)-(3),we design and develop a movie recommender system based on B/S architecture.Concretely,we design the overall architecture of the movie recommender system based on the functional and non-functional requirements of users.Furthermore,we develop the functional modules of the movie recommender system with Java language in the Eclipse development environment.This way,the LSH technique and the functional diversification approach can be successfully appled to the movie recommender system,so that the movie recommender system can offer users an accurate recommended movie list and diverse movie alternatives,in a quick,efficient and privacy-preserving way.