Research On Intelligent Recommendation System Based On Flow Data Cube

With the development of information technology,the data has grown exponentially.The rapid and accurate mining of information in massive data has become a research hotspot.The recommendation system has emerged as a result.Information mining under the big data environment has become a relatively active area for the research of recommendation systems.The existing recommendations are often calculated offline,and the recommendation results are regularly updated.The recommendation system lacks real-time performance,and cold start and data sparseness problems commonly exist in recommendation systems.In the era of information-based,how to quickly and accurately respond to user needs needs to be solved.The traditional stand-alone model requires a lot of time to perform iterative calculations of the recommendation algorithm and it is difficult to meet the current business needs.By comparing with existing big data processing frameworks,Spark big data calculation processing engine can improve the performance of the recommendation system with its advantage of memory-based computing.This paper focuses on the combined recommendation algorithm based on streaming data by the framework of Spark platform,and uses the book recommendation as an example for streaming implementation.It mainly includes the following two aspects:(1)Research the Parallelization of Recommendation Algorithms based on Streaming Data.Analyze the Fuzzy C-Means(FCM)algorithm and the Alternating Least Squares(ALS)algorithm based on the research of related technologies in distributed computing and recommendation systems,and make the design for parallelization under Spark framework.(2)Implement parallelization of recommendation algorithm under streaming data.In order to solve the problem of cold start and data sparsity of the traditional recommendation algorithm,this paper combine the FCM clustering algorithm with the ALS matrix decomposition algorithm to design and implement the parallel recommendation algorithm AAF(ALS AND FCM).For new users,users with similar attributes are grouped together to recommend new users based on the existing recommendation relationships in the cluster.For old users,fill in missing items of user ratings in the cluster cluster by ALS algorithm,then calculate the score matrix to obtain Top-N recommendation.The first step of the algorithm is data preprocessing.The T-C model convert the borrowing time and the number of borrowings into user-book scores.Then build a user-book scoring matrix to implement the hybrid recommendation algorithm AAF,and finally optimize the system performance.The experimental results verify that the AAF algorithm is superior to the single recommendation model in the Spark cluster environment.The related indicators of the algorithm are basically consistent with the stand-alone environment,and the accuracy loss problem is not obvious.With the increase of the data volume,the running time after the algorithm fusion is greatly shortened,the computational efficiency is significantly improved,and the performance requirements of the real-time recommendation are met.