Optimization And Implementation Of Recommendation Algorithm Based On Matrix Decomposition

Recommendation algorithm is a hot technology emerging in the era of big data explosion.Matrix decomposition,as a key technology,has become a research hotspot in the industry.However,due to problems such as sparse scoring matrix prediction and model building efficiency of massive historical behavior data of users,the existing strategy algorithm cannot give consideration to both,resulting in low recommendation results and user satisfaction.Therefore,it is an effective way to improve recommendation results and user satisfaction by comprehensively considering users' personalized characteristics and combining Spark big data platform when building the recommendation model.By studying the existing recommendation algorithm strategies,the research results obtained in this thesis mainly include:1.For the sparse matrix scoring prediction problem,this thesis adopts the improved matrix algorithm to solve the recommendation accuracy problem;As most scoring matrices tend to be sparse and dimensionality is also increasing rapidly,the prediction accuracy and calculation time of current matrix decomposition are limited.This thesis proposes a matrix decomposition model based on user characteristics,which can effectively improve the accuracy of prediction score and reduce the number of iterations.Through experiments on actual data and comparison with existing recommendation algorithms,the experimental results show that the method proposed in this thesis can well predict the user rating.2.Aiming at the problem of model construction efficiency of massive data,this thesis proposes an improved matrix algorithm based on Spark to solve this problem.Can be seen from the experimental data is small,the efficiency of local single building model is superior to the parallel computing,but with the increase of the scale,single machine operation is obviously better than the effect of parallel computing,but found that the parallel algorithm in the experiment of ALS in solving process consumes a lot of memory,cause the efficiency is very low,in order to solve this problem,combined with the actual demand of ALS algorithm was improved,the experimental results show that the proposed method can well solve the problem of constructing model efficiency.3.This thesis adopts the Spring MVC framework to design and implement the recommendation system,and visually presents the UFMF recommendation algorithm model proposed in this thesis.