Research Of E-commerce Recommendation System Based On Learning To Hash

With the rapid development of information technology,the Internet has become an indispensable part of people's life and production,and e-commerce with its convenient and affordable advantages to attract the people of all ages and works in the world.The network transaction has become a popular way to trade.In the information expansion,it is an important way for users and businesses to get a win-win situation that provide users with good recommendation.Therefore,the application of recommender system in the field of e-commerce has an important role and broad prospects.The main contents of this paper are as follows:(1)In the information age,data is an important element and learning to hash is an effective strategy for analyzing and mining data.This paper combines the method of learning to hash and traditional collaborative filtering algorithm.And proposed a recommendation algorithm based on learning to hash.Hashing map the data in the original space into the binary space by two steps of projection and quantization phase,which can effectively reduce the data storage space.In the two step of hashing,using the principal component analysis(PCA)to reduce the dimensionality of the original data,and then using the k-means clustering algorithm in quantization;In the end,using collaborative filtering to calculate and forecast the recommendation score,forming a list of items,and taking the top N items of the list as the final recommendation that presenting to the user.In this paper,we use the hit rate(HR)and the average reciprocal hit rank(ARHR)as the evaluation method,the experiment shows that this method can effectively carry out personalized recommendation;(2)Hash method usually uses Hamming distance as similarity measure,but this paper in order to keep the similarity structure of data in the original space using real-value Manhattan distance which corresponding to binary code as similarity measure.Analysis and comparison of the Hamming distance and Manhattan distance in measuring the data after quantization phase,illustrates the feasibility of using Manhattan distance.And through data experiments prove that the Manhattan distance lower than the traditional recommendation system using cosine similarity and Pearson correlation to measure the similarity of time complexity;(3)This paper puts forward the way to improve the recommendation algorithm by adding implicit feedback data.The experiment shows that the proposed method can improve the effect of the original algorithm on the implicit feedback data.