Design And Implementation Of Movie Recommendation System Based On Lambda Architecture

Movies are closely related to people’s lives.Currently,movie-related websites on the market can be roughly divided into two categories: one represented by Tencent Video is a kind of resource website,which mainly provides video resources and recommendations of popular movies;the other one represented by Douban Movie is a kind of judging website,which is mainly used as a movie evaluation platform and also has the function of recommending high-quality movies.The above two recommendations are non-personalized and all users see the same content,which cannot meet the diverse needs.Facing massive films,if users do not have a clear goal,it is difficult to find the movies they are interested in from the non-personalized recommendation lists.In order to solve the problem of "difficulty in selection" and help users quickly find the movies they are really interested in,this dissertation proposed a personalized recommendation scheme based on the linkage calculation between offline matrix factorization algorithm and real-time matrix factorization algorithm,designed and implemented a movie recommendation system covering three dimensions of playback,evaluation and recommendation based on the Lambda big data processing architecture and explored the form and operating effect of film product that integrates multiple modes.The playback and evaluation of movies are basic functions.Except for adding movies to wishlist,users can also play,rate,comment and tag them.Movie recommendation is the core function and four recommendation methods are designed in this dissertation: the personalized offline recommendation adopts bias-SVD matrix decomposition algorithm to analyze rating data of all users and accurately recommends movies that meet current user’s preference;the personalized real-time recommendation adopts a modified real-time matrix decomposition algorithm to analyze current user’s recent playback behavior and recommends movies that meet current user’s recent preference;the non-personalized statistical recommendation adopts a statistical algorithm to analyze rating data of all movies and recommends popular movies to expand users’ vision;the semi-personalized similar recommendation adopts an improved content recommendation algorithm based on TF-IDF algorithm to analyze current user’s behavior records and recommends personalized movies similar to a certain movie for current user.After the development and deployment on the big data processing platform,the overall operation of the system is good.As a movie recommendation platform,it can also be used as a movie playback platform and a movie evaluation platform.The system can effectively collect users’ behavior data and display the calculated personalized movie lists,which greatly improved users’ efficiency in selecting movies and saved their time.