Improvement And System Implementation Of Collaborative Filtering Recommendation Algorithm Based On Low Rank Matrix Decomposition

With the development of information network technology,e-commerce has developed very well,and online shopping has become an indispensable part of people's lives.And on the shopping website,there are so many goods,how can I find the items I like? It is based on this that the recommendation system was born,which played a significant role in recommending personalized items for us.At present,the recommendation systems applied in the e-commerce system mainly include the following: content-based recommendation,recommendation of collaborative filtering algorithm,and mixed recommendation.This paper mainly introduces the collaborative filtering algorithm based on matrix decomposition model and its optimization.Collaborative filtering algorithm is most commonly used in e-commerce.It is divided into neighborhood-based collaborative filtering and collaborative filtering based on matrix decomposition model.The research of matrix decomposition technology in collaborative filtering algorithm is also one of the cutting-edge technologies.The low rank matrix decomposition eliminates the redundant information and some invalid information of the original scoring matrix,and retains the effective information of the scoring matrix to the greatest extent,which is in line with the idea of matrix dimensionality reduction,which is greatly calculated.The reduction also saves storage space costs,so low rank matrix decomposition has a good track in the field of recommendation systems.This paper starts with the research of collaborative filtering algorithm for analyzing low rank matrix decomposition,consults the relevant data,compares the research status at home and abroad,and determines the direction and method of this paper.The content and innovation of this paper are as follows:First,this paper integrates the k-means clustering algorithm under the collaborative filtering algorithm of low rank matrix decomposition.After the scoring matrix is decomposed using the low rank matrix,the K-means clustering algorithm is used to classify the items into k categories according to the item feature matrix,and then each user's preference for the category,ie,category preference,is calculated in each category.In large-scale shopping sites such as Amazon,Taobao,etc.,there are many items and kinds of books,clothing,mobile phones,computers,beauty,etc.In each category,users have their own preferences.These preference preferences can be implicitly fed back through the scoring matrix.Users often buy items of a certain category and often give praise to a certain category of items.These feedbacks are category preferences.In the processing details,the scoring matrix is normalized by the item mean,and the average score of the item is used as the score of the new user,which can deal with the cold start problem of the new user joining the system.Second,considering that the user's interest preferences will vary over time,a time factor is added,with a time factor as a weight to limit the impact of longer-term scoring data on the model.Third,the algorithm of this paper is verified by experiments.The experimental results show that compared with the traditional algorithm,the optimized collaborative filtering algorithm has significant improvement in recommendation accuracy and recommendation quality.Finally,based on this improved algorithm,this paper designs and implements a shopping mall to purchase a website system.The system's functions include browsing products,selling recommendations,guessing your favorite,adding shopping carts,back-office management,etc.,adding personalized recommendation functions,which can bring a good experience to users.