A Privacy-preserving Movie Recommendation System In A Distributed Environment

In big data era,the varieties and volume of information have increased dramatically,which results in increasingly severe information overload.It is rather difficult for users to obtain valuable information efficiently.In this situation,recommender system comes into being and collaborative filtering-based recommendation algorithms are applied widely.However,in the distributed environment,there are the following disadvantages of traditional collaborative filtering-based algorithms:(1)This kind of algorithms usually assumes that data being utilized in making recommendations is just from one platform,without considering the fact that in the distributed environment,data is usually distributed in various platforms and the problem of users'privacy breach when merging data from these platforms.(2)In the distributed environment,data from different platforms update rather fast.When traditional collaborative filtering-based algorithms are applied to calculate similarity between data from these platforms,it tends to cause huge time cost and low efficiency of recommendation.In this way,users'fast response demand can't be satisfied.In addition,there is usually a tradeoff relationship between data privacy and data availability.It's another challenging task to improve the accuracy of recommendation while guaranteeing data privacy.In view of above-mentioned issues,in this thesis,we modified traditional collaborative filtering-based recommendation algorithm and utilized Locality-Sensitive Hashing(LSH)technique.We proposed two privacy-aware recommendation algorithms in the distributed environment and we further applied our proposed algorithm SR_chain-LSH to movie recommendation system so as to make personalized recommendation.The specific research in this thesis is introduced as follows:(1)According to the problem of users'privacy leakage in the process of merging data from multiple platforms,we utilized LSH technique to protect users'privacy.In this thesis,we made specific statement about basic concepts of LSH,mathematical calculation of LSH functions in cosine distance and basic concepts of amplified LSH.(2)On the basis of protecting users'privacy,according to the problem of improving the accuracy and efficiency of recommendation,we made research about amplified LSH and proposed amplified LSH recommendation algorithm SR_{Amplified-LSH}.In the algorithm,it is the first step to transform original users'data into less private users'indices through mathematical calculation with LSH functions.Then,we define users's similar relationship through designing“AND-OR-AND”amplified strategy,in order to improve the accuracy of finding similar users.Finally,according to similar users'historical behaviors,we utilized collaborative filtering technique to make recommendations to target users.In this thesis,we set up large amounts of experiments to validate the feasibility and effectiveness of the proposed algorithm SR_{Amplified-LSH}.And experimental results show that SR_{Amplified-LSH} is more accurate and efficient than compared methods.(3)In terms of probability analysis,we made further research about amplified LSH,proposed enhanced and amplified LSH recommendation algorithm SR_chain-LSH.In this thesis,we firstly analyzed the relationship between probability and distance in LSH technique and then specifically analyzed the relationship in“AND-OR-OR-AND”amplified strategy.In the algorithm,we frstly build users'indices through utilizing LSH functions.Then,we construct similar users'matrix by designing“AND-OR-OR-AND”amplified strategy,so as to improve the accuracy and efficiency of finding similar users.Finally,we make recommendations to target users based on their similar users'historical records.In this thesis,we set up comparative experiments to validate the effectiveness of algorithm SR_chain-LSH.In addition,we made comparison between algorithm SR_chain-LSH and algorithm SR_{Amplified-LSH}.And the experimental results show that SR_chain-LSH is more accurate and efficient than SR_{Amplified-LSH}.What's more,the algorithm SR_chain-LSH is more suitable to address data sparsity problem in recommendation.(4)We designed and implemented a movie recommendation system which was based on B/S architecture.And we applied the improved algorithm SR_chain-LSH to the movie recommendation system,in order to make personalized recommendation functionality.In terms of the design of the system,we made analysis about the feasibility of the system,requirements of the system,architecture of the system,functionality of the system and database of the system.At last,we utilized Java language to implement functional modules of the movie recommendation system.