Research On Reverse Recommendation Algorithm Based On Hadoop/Spark

The recommended algorithm is a hot topic in the field of machine learning.The main research objects of the proposed algorithm are users and items.On the one hand to solve user contradictions and help users find his favorite items;on the other hand to resolve item contradictions,so that items can be displayed in front of users who interested in it.Today's popular recommendations algorithms are basically user-centric and oriented,ignoring the mutual relationship between users and items.This article will break the traditional recommendation algorithm thinking,reverse thinking about the relationship between users and items,with the object as the center and orientation,Using the user-item connection to recommend the user for the item,that is,the reverse recommendation.The reverse recommendation on the one hand solves the long-tailed scene and the Matthew effect of the traditional recommendation algorithm;on the other hand,it helps third party to take the initiative,improve their competitiveness,and actively approve their products to users who are Suitable for it.This has important implications both in theoretical research and in practical applications.Reverse recommendation combined big data,Hadoop is used as data storage,and Spark,as a calculation engine,can respond quickly and in real time according to to changes in user The requirements.This paper designs a reverse recommendation algorithm based on real user purchase data,and designs the reverse recommendation algorithm.The algorithm improves the diversity and novelty of the recommended results,improves the coverage of recommended items,and conducts verification analysis through experiments.The main work of this article is as follows:(1)Climb an Amazon user to purchase a book record as a data set and process the data set ETL.Studied Hadoop,Spark,and distributed machine learning libraries Mahout and SparkML.(2)Research the current popular recommendation algorithm.Focus on analyzing the applicable scenarios and performance advantages and disadvantages of various recommendation systems and performance problems and disadvantages.A backward suggestion algorithm is proposed and a clustering algorithm is proposed for the sparsity and cold Start problems encountered in the recommendation algorithm,and then the reverse recommendation algorithm is combined.(3)It is mainly extracts features from the data,redesigns the algorithm,and performs secondary development and implementation of the combined reverse recommendation algorithm on Mahout and Spark ML.(4)Introduce the commonly used recommendation algorithm evaluation index,compare the performance of the reverse recommendation algorithm and other recommendation algorithms under each index,and make a conclusive analysis of the experimental results.