【作者】 王超；

【导师】 杨悦；

【作者基本信息】 哈尔滨工程大学 ， 计算机科学与技术， 2022， 硕士

【摘要】 目前,信息技术飞速发展,与人们相关的数据也越来越多,用户面对繁杂的信息,一时难以找到自己感兴趣的内容,推荐系统(Recommendation system)应运而生。协同过滤推荐算法易于理解和应用,在内容推荐上得到了广泛的应用。然而,协同过滤推荐算法过于依赖用户的历史数据,利用历史数据产生矩阵,通过矩阵分解从而产生相应的对应关系,数据量繁杂且不全面,致使分解的矩阵过度稀疏,在推荐时难以找到最接近的用户集合,相似性的计算也比较困难。同时协同过滤推荐算法,产生的推荐结果并没有考虑数据之间的语义关系,导致产生的推荐结果缺乏语义性。针对协同过滤存在的数据稀疏、缺乏语义性和推荐效率低的问题,本文采用在协同过滤算法的基础上利用聚类的方法提升推荐效率,同时融入知识表示向量方法,计算实体相似度,提升推荐语义性。本文主要从以下两个方面解决协同过滤算法存在的不足。针对协同过滤推荐算法中用户-项目矩阵数据量过大,进行推荐时时间复杂度较高,采用聚类的方法加以优化,利用用户的历史数据,得到用户-项目矩阵,采用Canopy和Kmeans对数据进行处理,通过Canopy得到聚类中心的个数,再将用户分成若干个聚类簇,减少了后期各个项目相似度的计算量,提升了推荐效率。针对推荐算法在推荐时没有考虑用户之间的语义关系,提出一种基于知识表示的协同过滤推荐方法,利用实体的知识表示向量计算相似度,首先建立知识图谱,利用知识表示向量来计算实体之间的相似性,根据相似性大小对稀疏矩阵进行填充,同时将原始数据中隐藏的实体间信息利用起来,提升了推荐结果的语义性,改善了存在的数据稀疏问题。最后在Movielens数据集上进行实验,结果表明本文提出的算法与推荐算法进行比较,在数据处理上降低了时间复杂度,提升了准确率,均值降低,有效地提升了推荐效率。更多还原

【Abstract】 At present,with the rapid development of information technology and more and more data related to people,it is difficult for users to find the content they are interested in in in the face of complex information,Recommendation system(recommendation system)came into being.Collaborative filtering recommendation algorithm is easy to understand and apply,and has been widely used in content recommendation.However,the traditional collaborative filtering recommendation algorithm relies too much on the user’s historical data,uses the historical data to generate a matrix,and generates the corresponding relationship through matrix decomposition.The amount of data is complex and incomplete,resulting in the decomposition of moments The array is too sparse,so it is difficult to find the closest user set and calculate the similarity.At the same time,the traditional collaborative filtering recommendation algorithm does not consider the semantic relationship between data,resulting in the lack of semantics of the recommendation results.Aiming at the problems of sparse data,lack of semantics and low recommendation efficiency in traditional collaborative filtering,this paper uses the clustering method to improve the recommendation efficiency based on the traditional collaborative filtering algorithm,and integrates the knowledge representation vector method to calculate the entity similarity and improve the recommendation semantics.This paper mainly solves the shortcomings of the traditional collaborative filtering algorithm from the following two aspects.In view of the large amount of user item matrix data in the traditional collaborative filtering recommendation algorithm and the high time complexity of recommendation,the clustering method is used to optimize,the user item matrix is obtained by using the user’s historical data,the data is processed by canopy and K-means,the number of cluster centers is obtained by canopy,and then the users are divided into several cluster clusters,It reduces the amount of calculation of each item similarity in the later stage and improves the recommendation efficiency.For the traditional recommendation algorithm does not consider the semantic relationship between users when recommending,a collaborative filtering recommendation method based on knowledge representation is proposed.The similarity is calculated by using the knowledge representation vector of entities.Firstly,the knowledge map is established,the similarity between entities is calculated by using the knowledge representation vector,and the sparse matrix is filled according to the similarity,At the same time,the inter entity information hidden in the original data is used to improve the semantics of the recommendation results and improve the existing problem of data sparsity.Finally,experiments are carried out on movielens data set.The results show that the proposed algorithm reduces the time complexity,improves the accuracy,reduces the mean value and effectively improves the recommendation efficiency compared with the traditional recommendation algorithm.更多还原