WO2020168992A1 - 一种商品推荐方法、装置、设备以及存储介质 - Google Patents

一种商品推荐方法、装置、设备以及存储介质 Download PDF

Info

Publication number
WO2020168992A1
WO2020168992A1 PCT/CN2020/075477 CN2020075477W WO2020168992A1 WO 2020168992 A1 WO2020168992 A1 WO 2020168992A1 CN 2020075477 W CN2020075477 W CN 2020075477W WO 2020168992 A1 WO2020168992 A1 WO 2020168992A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
product
users
rating
recommended
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2020/075477
Other languages
English (en)
French (fr)
Inventor
张莉
李泽鹏
周伟达
王邦军
章晓芳
屈蕴茜
赵雷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou University
Original Assignee
Suzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou University filed Critical Suzhou University
Publication of WO2020168992A1 publication Critical patent/WO2020168992A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Recommending goods or services

Definitions

  • the present invention relates to the field of information processing technology, in particular to a method, device, equipment and computer-readable storage medium for recommending commodities.
  • the recommendation system is a personalized information recommendation system that recommends information and products that interest users to users based on their information needs and interests.
  • the recommendation system conducts personalized calculations by studying the user's interest preferences, and the system discovers the user's points of interest, thereby guiding users to discover their own information needs.
  • Recommender systems are now widely used in many fields, among which the most typical field with good development and application prospects is e-commerce.
  • some traditional similarity measurement methods such as cosine similarity, Pearson Correlation Coefficient (PCC), and matrix factorization (Matrix factorization) algorithms have been widely used in product recommendation algorithms.
  • PCC Pearson Correlation Coefficient
  • Matrix factorization matrix factorization
  • the purpose of the present invention is to provide a product recommendation method, device, device and computer readable storage medium to solve the problem of poor recommendation performance on sparse scoring data.
  • the present invention provides a method for recommending commodities, including:
  • the score data of the product to be recommended by the neighbor users of the target user is obtained to recommend the product.
  • the obtaining rating data of different users for different commodities includes:
  • r ij If the value of r ij is 0, it means that the user has not rated the product, and if the value of r ij is non-zero, it means that the user has rated the product.
  • the calculating the user similarity between the different users according to the rating data of the products by different users includes:
  • A(u g ,u i ) represents the degree of consistency between user u g and user u i
  • the calculation method is: V g represents a rating vector based on the user u g , The superscript T denotes the transpose, V i represents a vector of the user based on the rating of u i; C (u g, u i ) represents a trade-off factor, calculated as: I g is the set of items evaluated by the user u g , and I i is the set of items evaluated by the user u i ; PCC(u g ,u i ) is to calculate the Pearson similarity between two users, the calculation method is:
  • r gj is the rating of the item m j by the user u g , Indicates the average value of the non-zero ratings of the product by the user u i , Represents the average value of the non-zero ratings of the user u g on the product; r ij is the rating of the user u i on the item m j ; I g ⁇ I i represents the set of products that the user u g and the user u i have jointly evaluated.
  • the obtaining the scoring data of the target user for the product to be recommended according to the score data of the neighboring users of the target user for the product to be recommended includes:
  • m j ⁇ M ⁇ r tj 0 ⁇
  • r ca user u c represents the score of product m a
  • S(u t , u c ) represents the user similarity between target user u t and neighboring user u c
  • NK is the number set of neighboring users of target user u t , including the target user's user selected from high to low similarity K nearest neighbors users
  • W ca m a treatment recommended product for neighbor users u c scores trust and W ca decay itear, decay is the decay rate, itear number of iterations , Each round of newly added points corresponds to a trust coefficient, and each round is multiplied by
  • This application also provides a product recommendation device, including:
  • the obtaining module is used to obtain rating data of different users for different products, and the rating data is used to characterize the user's preference for the products;
  • the calculation module is used to calculate the user similarity between the different users according to the rating data of the products by different users;
  • the recommendation module is used to obtain the score data of the target user for the product to be recommended according to the score data of the neighboring users of the target user to recommend the product, so as to recommend the product.
  • the acquisition module is used to:
  • r ij If the value of r ij is 0, it means that the user has not rated the product, and if the value of r ij is non-zero, it means that the user has rated the product.
  • calculation module is used for:
  • A(u g ,u i ) represents the degree of consistency between user u g and user u i
  • the calculation method is: V g represents a rating vector based on the user u g , The superscript T denotes the transpose, V i represents a vector of the user based on the rating of u i; C (u g, u i ) represents a trade-off factor, calculated as: I g is the set of items evaluated by the user u g , and I i is the set of items evaluated by the user u i ; PCC(u g ,u i ) is to calculate the Pearson similarity between two users, the calculation method is:
  • r gj is the rating of the item m j by the user u g , Indicates the average value of the non-zero ratings of the product by the user u i , Represents the average value of the non-zero ratings of the user u g on the product; r ij is the rating of the user u i on the item m j ; I g ⁇ I i represents the set of products that the user u g and the user u i have jointly evaluated.
  • the recommendation module is used for:
  • m j ⁇ M ⁇ r tj 0 ⁇
  • r ca user u c represents the score of product m a
  • S(u t , u c ) represents the user similarity between target user u t and neighboring user u c
  • NK is the number set of neighboring users of target user u t , including the target user's user selected from high to low similarity K nearest neighbors users
  • W ca m a treatment recommended product for neighbor users u c scores trust and W ca decay itear, decay is the decay rate, itear number of iterations , Each round of newly added points corresponds to a trust coefficient, and each round is multiplied by
  • This application also provides a product recommendation device, including:
  • Memory used to store computer programs
  • the processor is used to implement the steps of any one of the aforementioned commodity recommendation methods when the computer program is executed.
  • the present application also provides a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of any one of the aforementioned commodity recommendation methods are implemented.
  • the commodity recommendation method provided by the present invention obtains the rating data of different users for different commodities, and the rating data is used to characterize the user's degree of preference for the commodity; according to the rating data of different users on the commodity, the user similarity between different users The degree is calculated; according to the scoring data of the products to be recommended by the neighbor users of the target user, the scoring data of the products to be recommended by the target user is obtained to recommend the products.
  • This application is based on the product recommendation method of neighbor propagation and adopts an iterative idea to bring each round of scoring data into the next round of prediction scoring, making the scoring matrix denser and more accurate. Therefore, this application can better improve the performance of product recommendation.
  • this application also provides a product recommendation device, equipment, and computer-readable storage medium with the above technical effects.
  • FIG. 1 is a flowchart of a specific implementation manner of the commodity recommendation method provided by this application;
  • FIG. 2 is a structural block diagram of a product recommendation device provided by an embodiment of the present invention.
  • FIG. 3 is a block diagram of the product recommendation device provided by this application.
  • the core of the present invention is to provide a commodity recommendation method, device, equipment and computer-readable storage medium to solve the above technical problems.
  • FIG. 1 A flowchart of a specific implementation manner of the commodity recommendation method provided by the present application is shown in FIG. 1, and the method includes:
  • Step S101 Obtain scoring data of different users for different commodities, and the scoring data is used to characterize the user's degree of preference for the commodity;
  • Step S102 Calculate the user similarity between the different users according to the rating data of the products by different users;
  • ACCCC Accordance and Compromise based Pearson Correlation Coefficient
  • A(u g ,u i ) represents the degree of consistency between user u g and user u i
  • the calculation method is: V g represents a rating vector based on the user u g , The superscript T denotes the transpose, V i represents a vector of the user based on the rating of u i; C (u g, u i ) represents a trade-off factor, calculated as: I g is the set of items evaluated by the user u g , and I i is the set of items evaluated by the user u i ; PCC(u g ,u i ) is to calculate the Pearson similarity between two users, the calculation method is:
  • r gj is the rating of the item m j by the user u g , Indicates the average value of the non-zero ratings of the product by the user u i , Represents the average value of the non-zero ratings of the user u g on the product; r ij is the rating of the user u i on the item m j ; I g ⁇ I i represents the set of products that the user u g and the user u i have jointly evaluated. For example, the user u g score is ⁇ 1,3,4,2,1,1,2 ⁇ , then It is equal to 2.
  • Step S103 Obtain the score data of the target user for the product to be recommended according to the score data of the neighboring users of the target user to recommend the product.
  • m j ⁇ M ⁇ r tj 0 ⁇
  • NK is the number set of neighbor users of the target user u t , including K neighbor users selected from high to low similarity to the target user, and then select users who have rated product a from these users.
  • r ca user u c represents the score of product m a
  • S(u t , u c ) represents the user similarity between target user u t and neighboring user u c
  • NK is the number set of neighboring users of target user u t , including the target user's user selected from high to low similarity K nearest neighbors users
  • the commodity recommendation method provided by the present invention obtains the rating data of different users for different commodities, and the rating data is used to characterize the user's degree of preference for the commodity; according to the rating data of different users on the commodity, the user similarity between different users The degree is calculated; according to the scoring data of the products to be recommended by the neighbor users of the target user, the scoring data of the products to be recommended by the target user is obtained to recommend the products.
  • This application is based on the product recommendation method of neighbor propagation and adopts an iterative idea to bring each round of scoring data into the next round of prediction scoring, making the scoring matrix denser and more accurate. Therefore, this application can better improve the performance of product recommendation.
  • This embodiment of the application selects the Moußs data set ML_100k data for testing.
  • the data set consists of 942 users, 1682 movies, and a total of score records up to As many as 100,000.
  • the score range of this data set is ⁇ 1,2,3,4,5 ⁇ , 1 means a hate movie, 5 means a favorite movie.
  • the sparseness of the data set reaches 93.7% to form a 942*1682 rating matrix, in which the unrated movies are represented by 0.
  • r ij ⁇ ⁇ 0, s ⁇ represents the rating of the product m j by the user u i . If the value of r ij is 0, it means that the user has not evaluated the product, and if r ij takes other values, it means that the user has evaluated the product, and the value is the score.
  • the size of the scoring value indicates how high or low the user likes the product. In this embodiment, s is 5.
  • ACCCC Accordance and Compromise based Pearson Correlation Coefficient
  • A(u g ,u i ) represents the degree of consistency between u g and neighboring users u i
  • the calculation method is:
  • V g represents a rating vector based on user u g ;
  • C(u g ,u i ) represents a compromise factor, considering the percentage of common ratings between two users, the calculation method is:
  • I g is a collection of items evaluated by the user u g .
  • PCC(u g ,u i ) is to calculate the Pearson similarity between two users, the calculation method is:
  • r gj is the rating of user u g on item m j
  • the target user u t recommend an item m a ⁇ ⁇ m j
  • m j ⁇ M ⁇ r tj 0 ⁇ , i.e. the prediction target user u t m a product score of
  • decay is the decay rate
  • itear is the number of iterations.
  • Each round of newly added points must have a trust factor, which is multiplied by the decay rate for each round.
  • the value of K is 40, decay is 0.9, and itear is 15 times.
  • the data set is randomly divided into five training sets and test sets.
  • the comparison method includes the present invention, user-based PCC similarity (abbreviated as UCF-PCC) and commodity-based Pearson similarity algorithm (abbreviated as ICF-PCC) and non-iterative prediction similarity (ACPCC) .
  • UCF-PCC user-based PCC similarity
  • ICF-PCC commodity-based Pearson similarity algorithm
  • ACPCC non-iterative prediction similarity
  • IR gt is the set of movies liked by the user on the test set
  • IR gp is the set of movies recommended to the user. The results are shown in Table 1. It can be seen that the recommended performance of the present invention is significantly better than other comparison methods.
  • the product recommendation device provided by the embodiment of the present invention will be introduced below.
  • the product recommendation device described below and the product recommendation method described above can be referred to each other.
  • FIG. 2 is a structural block diagram of a product recommendation device provided by an embodiment of the present invention.
  • the product recommendation device may include:
  • the obtaining module 100 is configured to obtain rating data of different users for different commodities, and the rating data is used to characterize the user's degree of preference for the commodity;
  • the calculation module 200 is configured to calculate the user similarity between the different users according to the rating data of the products by different users;
  • the recommendation module 300 is configured to obtain the score data of the target user for the product to be recommended according to the score data of the neighboring users of the target user for the product to be recommended, so as to recommend the product.
  • the acquisition module in the recommendation device provided in this application is used to:
  • r ij If the value of r ij is 0, it means that the user has not rated the product, and if the value of r ij is non-zero, it means that the user has rated the product.
  • the calculation module in the recommendation device provided in this application is used to:
  • A(u g ,u i ) represents the degree of consistency between user u g and user u i
  • the calculation method is: V g represents a rating vector based on user u g ; C(u g ,u i ) represents a compromise factor, and the calculation method is: I g is the set of items evaluated by the user u g , PCC(u g ,u i ) is to calculate the Pearson similarity between two users, the calculation method is:
  • r gj is the rating of the item m j by the user u g , Represents the average value of the non-zero ratings of the product by the user u i .
  • the recommendation module in the recommendation device provided in this application is used to:
  • m j ⁇ M ⁇ r tj 0 ⁇
  • the product recommendation device of this embodiment is used to implement the aforementioned product recommendation method. Therefore, the specific implementation of the product recommendation device can be seen in the previous embodiment of the product recommendation method, for example, the acquisition module 100, the calculation module 200, and the recommendation module 300. , Are respectively used to implement steps S101, S102, and S103 in the foregoing commodity recommendation method. Therefore, for the specific implementation, refer to the description of the respective parts of the embodiment, and details are not described herein again.
  • the commodity recommendation method provided by the present invention obtains the rating data of different users for different commodities, and the rating data is used to characterize the user's degree of preference for the commodity; according to the rating data of different users on the commodity, the user similarity between different users The degree is calculated; according to the scoring data of the products to be recommended by the neighbor users of the target user, the scoring data of the products to be recommended by the target user is obtained to recommend the products.
  • This application is based on the product recommendation method of neighbor propagation, and adopts the idea of iteration to bring the scoring data of each round to the next round of prediction scoring, making the scoring matrix more and more dense, and also making the prediction score more and more accurate. Therefore, this application can better improve the performance of product recommendation.
  • this application also provides a product recommendation device.
  • the product recommendation device structure block diagram provided in this application the device includes:
  • the memory 11 is used to store computer programs
  • the processor 12 is configured to implement the steps of any one of the aforementioned commodity recommendation methods when executing the computer program.
  • the present application also provides a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of any one of the aforementioned commodity recommendation methods are implemented.
  • this application obtains the rating data of different users for different products, and the rating data is used to characterize the user's preference for the product; according to the rating data of different users to the product, the user similarity between different users is calculated; According to the scoring data of the products to be recommended by the neighboring users of the target user, the scoring data of the products to be recommended by the target user is obtained to recommend the products.
  • This application is based on the product recommendation method of neighbor propagation and adopts an iterative idea to bring each round of scoring data into the next round of prediction scoring, making the scoring matrix denser and more accurate. Therefore, this application can better improve the performance of product recommendation.
  • the steps of the method or algorithm described in the embodiments disclosed in this document can be directly implemented by hardware, a software module executed by a processor, or a combination of the two.
  • the software module can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or all areas in the technical field. Any other known storage medium.

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种商品推荐方法、商品推荐装置、商品推荐设备以及计算机可读存储介质。商品推荐方法包括：获取不同用户分别针对不同商品的评分数据，评分数据用于表征用户对商品的喜欢程度（S101）；根据不同用户对商品的评分数据，对不同用户之间的用户相似度进行计算（S102）；根据目标用户的近邻用户对待推荐商品的评分数据得到目标用户对待推荐商品的评分数据，以对商品进行推荐（S103）。基于近邻传播的商品推荐方法，采用迭代的思想，将每一轮的评分数据带入到下一轮预测评分当中，使得评分矩阵越来越稠密，也致使预测评分越来越精确，可以更好地提高商品推荐的性能。

Description

一种商品推荐方法、装置、设备以及存储介质
本申请要求2019年2月18日提交中国专利局、申请号为201910123041.2、名称为“一种商品推荐方法、装置、设备以及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中：
技术领域
本发明涉及信息处理技术领域，特别是涉及一种商品推荐方法、装置、设备以及计算机可读存储介质。
背景技术
互联网的出现和普及给用户带来了大量的信息，满足了用户在信息时代对信息的需求，但随着网络的迅速发展而带来的网上信息量的大幅增长，使得用户在面对大量信息时无法从中获得对自己真正有用的那部分信息，对信息的使用效率反而降低了，这就是所谓的信息超载问题。
解决信息超载问题一个非常有潜力的办法是推荐系统，它是根据用户的信息需求、兴趣等，将用户感兴趣的信息、产品等推荐给用户的个性化信息推荐系统。和搜索引擎相比推荐系统通过研究用户的兴趣偏好，进行个性化计算，由系统发现用户的兴趣点，从而引导用户发现自己的信息需求。
推荐系统现已广泛应用于很多领域，其中最典型并具有良好的发展和应用前景的领域就是电子商务领域。对于个性化推荐系统一些传统的相似度度量方法比如余弦相似度，Pearson相似度(Pearson Correlation Coefficient,PCC)以及矩阵分解(Matrix factorization)算法已经广泛应用于商品推荐算法中。然而用户和商品的数量的不断增加，评分矩阵的稀疏性也越来越明显。按照相似度计算以及矩阵分解已不能更好地提升推荐性能。
发明内容
本发明的目的是提供一种商品推荐方法、装置、设备以及计算机可读 存储介质，以解决在稀疏的评分数据上推荐性能较差的问题。
为解决上述技术问题，本发明提供一种商品推荐方法，包括：
获取不同用户分别针对不同商品的评分数据，所述评分数据用于表征用户对商品的喜欢程度；
根据不同用户对商品的评分数据，对所述不同用户之间的用户相似度进行计算；
根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据，以对商品进行推荐。
可选地，所述获取不同用户分别针对不同商品的评分数据包括：
采用集合U＝{u 1,...,u l}和M＝{m 1,...,m n}分别代表用户集合与商品集合，l和n分别表示用户数量和商品数量，输入商品评分矩阵
Figure PCTCN2020075477-appb-000001
其中r ij∈{0,s}表示用户u i对商品m j的评分；
若r ij取值为0则表示用户对商品未评价，若r ij取非零值则表示用户对商品评价的分值。
可选地，所述根据不同用户对商品的评分数据，对所述不同用户之间的用户相似度进行计算包括：
对任意两个用户u g和u i，采用S(u g,u i)＝A(u g,u i)C(u g,u i)PCC(u g,u i)计算二者之间的用户相似度；
其中，A(u g,u i)表示用户u g和用户u i的一致性程度，计算方式为：
Figure PCTCN2020075477-appb-000002
V g表示基于用户u g的评分向量，
Figure PCTCN2020075477-appb-000003
中的上角标T表示转置，V i表示基于用户u i的评分向量；C(u g,u i)表示折衷因素，计算方式为：
Figure PCTCN2020075477-appb-000004
I g为用户u g评价过的物品集合，I i为用户u i评价过的物品集合；PCC(u g,u i)是计算两个用户之间的Pearson相似度，计算方式为：
Figure PCTCN2020075477-appb-000005
其中，r gj是用户u g对物品m j的评分，
Figure PCTCN2020075477-appb-000006
表示用户u i对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000007
代表用户u g对商品非零评分的平均值；r ij是用户u i对物品m j的评分；I g∩I i表示用户u g和用户u i共同评价过的商品集合。
可选地，所述根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据包括：
在目标用户为u t∈U，给所述目标用户u t推荐待推荐商品m a∈{m j|m j∈M∧r tj＝0}时，预测目标用户u t对所述待推荐商品m a的评分
Figure PCTCN2020075477-appb-000008
为：
Figure PCTCN2020075477-appb-000009
其中，m a∈{m j|m j∈M∧r tj＝0}代表用户未曾评分的商品，∧是条件连接符号，r ca表示用户u c对商品m a的评分，
Figure PCTCN2020075477-appb-000010
表示目标用户u t对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000011
表示近邻用户u c对商品非零评分的平均值，S(u t,u c)表示目标用户u t和近邻用户u c的用户相似性，NK是目标用户u t的近邻用户数集，包含与目标用户的用户相似度从高到低选取的K个近邻用户，W ca为近邻用户u c对待推荐商品m a的评分信任度且W ca＝decay itear，decay为衰减率，itear为迭代次数，每一轮新添的分值对应一个信任系数，每一轮乘以衰减率。
本申请还提供了一种商品推荐装置，包括：
获取模块，用于获取不同用户分别针对不同商品的评分数据，所述评分数据用于表征用户对商品的喜欢程度；
计算模块，用于根据不同用户对商品的评分数据，对所述不同用户之间的用户相似度进行计算；
推荐模块，用于根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据，以对商品进行推荐。
可选地，所述获取模块用于：
采用集合U＝{u 1,...,u l}和M＝{m 1,...,m n}分别代表用户集合与商品集合，l和n分别表示用户数量和商品数量，输入商品评分矩阵
Figure PCTCN2020075477-appb-000012
其中ri j∈{0,s}表示用户u i对商品m j的评分；
若r ij取值为0则表示用户对商品未评价，若r ij取非零值则表示用户对商品评价的分值。
可选地，所述计算模块用于：
对任意两个用户u g和u i，采用S(u g,u i)＝A(u g,u i)C(u g,u i)PCC(u g,u i)计算二者之间的用户相似度；
其中，A(u g,u i)表示用户u g和用户u i的一致性程度，计算方式为：
Figure PCTCN2020075477-appb-000013
V g表示基于用户u g的评分向量，
Figure PCTCN2020075477-appb-000014
中的上角标T表示转置，V i表示基于用户u i的评分向量；C(u g,u i)表示折衷因素，计算方式为：
Figure PCTCN2020075477-appb-000015
I g为用户u g评价过的物品集合，I i为用户u i评价过的物品集合；PCC(u g,u i)是计算两个用户之间的Pearson相似度，计算方式为：
Figure PCTCN2020075477-appb-000016
其中，r gj是用户u g对物品m j的评分，
Figure PCTCN2020075477-appb-000017
表示用户u i对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000018
代表用户u g对商品非零评分的平均值；r ij是用户u i对物品m j的评分；I g∩I i表示用户u g和用户u i共同评价过的商品集合。
可选地，所述推荐模块用于：
在目标用户为u t∈U，给所述目标用户u t推荐待推荐商品m a∈{m j|m j∈M∧r tj＝0}时，预测目标用户u t对所述待推荐商品m a的评分
Figure PCTCN2020075477-appb-000019
为：
Figure PCTCN2020075477-appb-000020
其中，m a∈{m j|m j∈M∧r tj＝0}代表用户未曾评分的商品，∧是条件连接符号，r ca表示用户u c对商品m a的评分，
Figure PCTCN2020075477-appb-000021
表示目标用户u t对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000022
表示近邻用户u c对商品非零评分的平均值，S(u t,u c)表示目标用户u t和近邻用户u c的用户相似性，NK是目标用户u t的近邻用户数集，包含与目标用户的用户相似度从高到低选取的K个近邻用户，W ca为近邻用户u c对待推荐商品m a的评分信任度且W ca＝decay itear，decay为衰减率，itear为迭代次数，每一轮新添的分值对应一个信任系数，每一轮乘以衰减率。
本申请还提供了一种商品推荐设备，包括：
存储器，用于存储计算机程序；
处理器，用于执行所述计算机程序时实现上述任一种商品推荐方法的步骤。
本申请还提供了一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现上述任一种所述商品推荐方法的步骤。
本发明所提供的商品推荐方法，通过获取不同用户分别针对不同商品的评分数据，评分数据用于表征用户对商品的喜欢程度；根据不同用户对商品的评分数据，对不同用户之间的用户相似度进行计算；根据目标用户的近邻用户对待推荐商品的评分数据得到目标用户对待推荐商品的评分数据，以对商品进行推荐。本申请基于近邻传播的商品推荐方法，采用迭代的思想，将每一轮的评分数据带入到下一轮预测评分当中，使得评分矩阵越来越稠密，也致使预测评分越来越精确。因此，本申请可以更好地提高商品推荐的性能。此外，本申请还提供了一种具有上述技术效果的商品推荐装置、设备以及计算机可读存储介质。
附图说明
为了更清楚的说明本发明实施例或现有技术的技术方案，下面将对实 施例或现有技术描述中所需要使用的附图作简单的介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请所提供的商品推荐方法的一种具体实施方式的流程图；
图2为本发明实施例提供的商品推荐装置的结构框图；
图3为本申请所提供的商品推荐设备结构框图。
具体实施方式
随着用户和商品的数量的不断增加，评分矩阵的稀疏性也越来越明显。按照相似度计算以及矩阵分解已不能更好地提升推荐性能。本发明的核心是提供一种商品推荐方法、装置、设备以及计算机可读存储介质，以解决上述技术问题。
为了使本技术领域的人员更好地理解本发明方案，下面结合附图和具体实施方式对本发明作进一步的详细说明。显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本申请所提供的商品推荐方法的一种具体实施方式的流程图如图1所示，该方法包括：
步骤S101：获取不同用户分别针对不同商品的评分数据，所述评分数据用于表征用户对商品的喜欢程度；
采用集合U＝{u 1,...,u l}和M＝{m 1,...,m n}分别代表用户集合与商品集合，l和n分别表示用户数量和商品数量，输入商品评分矩阵
Figure PCTCN2020075477-appb-000023
其中r ij∈{0,s}表示用户u i对商品m j的评分；若r ij取值为0则表示用户对商品未评价，若r ij取非零值则表示用户对商品评价的分值。分值的大小表示用户对商品的喜欢程度的高低。
步骤S102：根据不同用户对商品的评分数据，对所述不同用户之间的 用户相似度进行计算；
具体地，采用ACPCC(Accordance and Compromise based Pearson Correlation Coefficient)相似度对不同用户之间的用户相似度进行计算。对任意两个用户u g和u i，采用S(u g,u i)＝A(u g,u i)C(u g,u i)PCC(u g,u i)计算二者之间的用户相似度；
其中，A(u g,u i)表示用户u g和用户u i的一致性程度，计算方式为：
Figure PCTCN2020075477-appb-000024
V g表示基于用户u g的评分向量，
Figure PCTCN2020075477-appb-000025
中的上角标T表示转置，V i表示基于用户u i的评分向量；C(u g,u i)表示折衷因素，计算方式为：
Figure PCTCN2020075477-appb-000026
I g为用户u g评价过的物品集合，I i为用户u i评价过的物品集合；PCC(u g,u i)是计算两个用户之间的Pearson相似度，计算方式为：
Figure PCTCN2020075477-appb-000027
其中，r gj是用户u g对物品m j的评分，
Figure PCTCN2020075477-appb-000028
表示用户u i对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000029
代表用户u g对商品非零评分的平均值；r ij是用户u i对物品m j的评分；I g∩I i表示用户u g和用户u i共同评价过的商品集合。例如，用户u g打分是{1,3,4,2,1,1,2},那么
Figure PCTCN2020075477-appb-000030
就等于2。
步骤S103：根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据，以对商品进行推荐。
在目标用户为u t∈U，给所述目标用户u t推荐待推荐商品m a∈{m j|m j∈M∧r tj＝0}时，预测目标用户u t对所述待推荐商品m a的评分
Figure PCTCN2020075477-appb-000031
为：
Figure PCTCN2020075477-appb-000032
其中，m a∈{m j|m j∈M∧r tj＝0}代表用户未曾评分的商品，∧是条件连接符号，表示并且的意思；u c∈NK∧r ca＞0表示两个条件同时满足，NK是目标用户u t的近邻用户数集，包含与目标用户相似度从高到低选取的K个近邻用户，再从这些用户里挑选出对商品a评过分的用户。r ca表示用户u c对商品m a的评分，
Figure PCTCN2020075477-appb-000033
表示目标用户u t对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000034
表示近邻用户u c对商品非零评分的平均值，S(u t,u c)表示目标用户u t和近邻用户u c的用户相似性，NK是目标用户u t的近邻用户数集，包含与目标用户的用户相似度从高到低选取的K个近邻用户，W ca为近邻用户u c对待推荐商品m a的评分信任度且W ca＝decay itear，decay为衰减率，itear为迭代次数，每一轮新添的分值对应一个信任系数，每一轮乘以衰减率。
本发明所提供的商品推荐方法，通过获取不同用户分别针对不同商品的评分数据，评分数据用于表征用户对商品的喜欢程度；根据不同用户对商品的评分数据，对不同用户之间的用户相似度进行计算；根据目标用户的近邻用户对待推荐商品的评分数据得到目标用户对待推荐商品的评分数据，以对商品进行推荐。本申请基于近邻传播的商品推荐方法，采用迭代的思想，将每一轮的评分数据带入到下一轮预测评分当中，使得评分矩阵越来越稠密，也致使预测评分越来越精确。因此，本申请可以更好地提高商品推荐的性能。
下面以一具体实施例对本申请所提供的商品推荐方法进行进一步详细阐述，本申请实施例选取Movielens数据集ML_100k数据进行了测试，该数据集由942个用户，1682个电影数量，一共评分记录达到100000之多。该数据集评分范围为{1,2,3,4,5}，1表示讨厌的电影，5表示很喜欢的电影。该数据集的稀疏程度达到93.7％构成了一个942*1682的评分矩阵，其中对未评价过的电影用0表示。划出80％作为训练集，剩余部分作为测试集。
具体实施步骤如下：
令集合U＝{u 1,...,u l}和M＝{m 1,...,m n}分别代表用户集合与商品集合，l和n分别表示用户数量和商品数量。在本实施例中，l＝942，n＝1682。
输入商品评分矩阵
Figure PCTCN2020075477-appb-000035
其中r ij∈{0,s}表示用户u i对商品m j的 评分。若r ij取值为0表示用户对商品未评价，若r ij取其他值，则表示用户对商品评价过，值即为评分。评分值的大小表示用户对商品的喜欢程度的高低。在本实施例中，s为5。
采用ACPCC(Accordance and Compromise based Pearson Correlation Coefficient)相似度对不同用户之间的用户相似度进行计算。对任意两个用户u g和u i，它们之间的相似度计算方式如下：
S(u g,u i)＝A(u g,u i)C(u g,u i)PCC(u g,u i)
其中A(u g,u i)表示u g和近邻用户u i一致性的程度，计算方式为：
Figure PCTCN2020075477-appb-000036
V g表示基于用户u g的评分向量；C(u g,u i)表示折衷因素，考虑两个用户之间共同评分的百分比，计算方式为：
Figure PCTCN2020075477-appb-000037
其中I g是用户u g评价过的物品集合。PCC(u g,u i)是计算两个用户之间的Pearson相似度，计算方式为：
Figure PCTCN2020075477-appb-000038
其中r gj是用户u g对物品m j的评分，
Figure PCTCN2020075477-appb-000039
表示用户u i对商品非零评分的平均值。
假设目标用户为u t∈U，为目标用户u t推荐某个商品m a∈{m j|m j∈M∧r tj＝0}，也就是预测目标用户u t对商品m a的评分
Figure PCTCN2020075477-appb-000040
Figure PCTCN2020075477-appb-000041
其中
Figure PCTCN2020075477-appb-000042
表示用户u t对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000043
表示近邻用户u c对商品非零评分的平均值，S(u t,u c)表示目标用户u t和近邻用户u c的相似性，NK是目标用户u t的近邻用户数集，包含用户相似度从高到低选取的K个用户， W ca为用户u c对商品m a的评分信任度且
W ca＝decay itear
其中decay为衰减率，itear是迭代次数。每一轮新添的分值要有一个信任系数，每一轮乘以衰减率。在本实施例中，K的取值为40，decay为0.9，itear为15次。
本发明的效果可以通过如下实验验证：对数据集随机划分5次训练集和测试集。在推荐预测中，对比方法包括本发明、基于用户的PCC相似度(简写为UCF-PCC)和基于商品的Pearson相似度的算法(简写为ICF-PCC)以及未迭代预测的相似度(ACPCC)。
采用绝对值误差指标(MAE)和平方误差指标(RMSE)还有召回率(Recall)对推荐效果进行评估：
Figure PCTCN2020075477-appb-000044
Figure PCTCN2020075477-appb-000045
Figure PCTCN2020075477-appb-000046
其中l'和n'是测试集上的用户数和电影数量，IR gt是测试集上用户喜欢的电影集合，IR gp是推荐给用户喜欢的电影集合。结果如表1所示，可以看出本发明的推荐性能明显好于其他对比方法。
表1三种算法的推荐结果对比
推荐 MAE RMSE Recall
UCF-PCC 0.9838 1.1174 0.2607
ICF-PCC 0.8716 1.0196 0.0571
ACPCC 0.7689 0.9771 0.7056
本发明 0.7445 0.9569 0.7295
下面对本发明实施例提供的商品推荐装置进行介绍，下文描述的商品 推荐装置与上文描述的商品推荐方法可相互对应参照。
图2为本发明实施例提供的商品推荐装置的结构框图，参照图2商品推荐装置可以包括：
获取模块100，用于获取不同用户分别针对不同商品的评分数据，所述评分数据用于表征用户对商品的喜欢程度；
计算模块200，用于根据不同用户对商品的评分数据，对所述不同用户之间的用户相似度进行计算；
推荐模块300，用于根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据，以对商品进行推荐。
作为一种具体实施方式，本申请所提供的推荐装置中所述获取模块用于：
采用集合U＝{u 1,...,u l}和M＝{m 1,...,m n}分别代表用户集合与商品集合，l和n分别表示用户数量和商品数量，输入商品评分矩阵
Figure PCTCN2020075477-appb-000047
其中r ij∈{0,s}表示用户u i对商品m j的评分；
若r ij取值为0则表示用户对商品未评价，若r ij取非零值则表示用户对商品评价的分值。
作为一种具体实施方式，本申请所提供的推荐装置中所述计算模块用于：
对任意两个用户u g和u i，采用S(u g,u i)＝A(u g,u i)C(u g,u i)PCC(u g,u i)计算二者之间的用户相似度；
其中，A(u g,u i)表示用户u g和用户u i的一致性程度，计算方式为：
Figure PCTCN2020075477-appb-000048
V g表示基于用户u g的评分向量；C(u g,u i)表示折衷因素，计算方式为：
Figure PCTCN2020075477-appb-000049
I g为用户u g评价过的物品集合，PCC(u g,u i)是计算两个用户之间的Pearson相似度，计算方式为：
Figure PCTCN2020075477-appb-000050
其中，r gj是用户u g对物品m j的评分，
Figure PCTCN2020075477-appb-000051
表示用户u i对商品非零评分的平均值。
作为一种具体实施方式，本申请所提供的推荐装置中所述推荐模块用于：
在目标用户为u t∈U，给所述目标用户u t推荐待推荐商品m a∈{m j|m j∈M∧r tj＝0}时，预测目标用户u t对所述待推荐商品m a的评分
Figure PCTCN2020075477-appb-000052
为：
Figure PCTCN2020075477-appb-000053
其中，
Figure PCTCN2020075477-appb-000054
表示目标用户u t对商品非零评分的平均值，
Figure PCTCN2020075477-appb-000055
表示近邻用户u c对商品非零评分的平均值，S(u t,u c)表示目标用户u t和近邻用户u c的用户相似性，NK是目标用户u t的近邻用户数集，包含与目标用户的用户相似度从高到低选取的K个近邻用户，W ca为近邻用户u c对待推荐商品m a的评分信任度且W ca＝decay itear，decay为衰减率，itear为迭代次数，每一轮新添的分值对应一个信任系数，每一轮乘以衰减率。
本实施例的商品推荐装置用于实现前述的商品推荐方法，因此商品推荐装置中的具体实施方式可见前文中的商品推荐方法的实施例部分，例如，获取模块100，计算模块200，推荐模块300，分别用于实现上述商品推荐方法中步骤S101，S102，S103，所以，其具体实施方式可以参照相应的各个部分实施例的描述，在此不再赘述。
本发明所提供的商品推荐方法，通过获取不同用户分别针对不同商品的评分数据，评分数据用于表征用户对商品的喜欢程度；根据不同用户对商品的评分数据，对不同用户之间的用户相似度进行计算；根据目标用户的近邻用户对待推荐商品的评分数据得到目标用户对待推荐商品的评分数据，以对商品进行推荐。本申请基于近邻传播的商品推荐方法，采用迭代 的思想，将每一轮的评分数据带入到下一轮预测评分当中，使得评分矩阵越来越稠密，也致使预测评分越来越精确。因此，本申请可以更好地提高商品推荐的性能。
此外，本申请还提供了一种商品推荐设备，如图3本申请所提供的商品推荐设备结构框图所示，该设备包括：
存储器11，用于存储计算机程序；
处理器12，用于执行所述计算机程序时实现上述任一种商品推荐方法的步骤。
此外，本申请还提供了一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现上述任一种商品推荐方法的步骤。
综上，本申请通过获取不同用户分别针对不同商品的评分数据，评分数据用于表征用户对商品的喜欢程度；根据不同用户对商品的评分数据，对不同用户之间的用户相似度进行计算；根据目标用户的近邻用户对待推荐商品的评分数据得到目标用户对待推荐商品的评分数据，以对商品进行推荐。本申请基于近邻传播的商品推荐方法，采用迭代的思想，将每一轮的评分数据带入到下一轮预测评分当中，使得评分矩阵越来越稠密，也致使预测评分越来越精确。因此，本申请可以更好地提高商品推荐的性能。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其它实施例的不同之处，各个实施例之间相同或相似部分互相参见即可。对于实施例公开的装置而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
专业人员还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以直接用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
以上对本发明所提供的商品推荐方法、装置、设备以及计算机可读存储介质进行了详细介绍。本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想。应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以对本发明进行若干改进和修饰，这些改进和修饰也落入本发明权利要求的保护范围内。

Claims (10)

  1. 一种商品推荐方法，其特征在于，包括：
    获取不同用户分别针对不同商品的评分数据，所述评分数据用于表征用户对商品的喜欢程度；
    根据不同用户对商品的评分数据，对所述不同用户之间的用户相似度进行计算；
    根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据，以对商品进行推荐。
  2. 如权利要求1所述的商品推荐方法，其特征在于，所述获取不同用户分别针对不同商品的评分数据包括：
    采用集合U＝{u 1,...,u l}和M＝{m 1,...,m n}分别代表用户集合与商品集合，l和n分别表示用户数量和商品数量，输入商品评分矩阵
    Figure PCTCN2020075477-appb-100001
    其中r ij∈{0,s}表示用户u i对商品m j的评分；
    若r ij取值为0则表示用户对商品未评价，若r ij取非零值则表示用户对商品评价的分值。
  3. 如权利要求2所述的商品推荐方法，其特征在于，所述根据不同用户对商品的评分数据，对所述不同用户之间的用户相似度进行计算包括：
    对任意两个用户u g和u i，采用S(u g,u i)＝A(u g,u i)C(u g,u i)PCC(u g,u i)计算二者之间的用户相似度；
    其中，A(u g,u i)表示用户u g和用户u i的一致性程度，计算方式为：
    Figure PCTCN2020075477-appb-100002
    V g表示基于用户u g的评分向量，
    Figure PCTCN2020075477-appb-100003
    中的上角标T表示转置，V i表示基于用户u i的评分向量；C(u g,u i)表示折衷因素，计算方式为：
    Figure PCTCN2020075477-appb-100004
    I g为用户u g评价过的物品集合，I i为用户u i评价过的物品集合；PCC(u g,u i)是计算两个用户之间的Pearson相似度，计算方式为：
    Figure PCTCN2020075477-appb-100005
    其中，r gj是用户u g对物品m j的评分，
    Figure PCTCN2020075477-appb-100006
    表示用户u i对商品非零评分的平均值，
    Figure PCTCN2020075477-appb-100007
    代表用户u g对商品非零评分的平均值；r ij是用户u i对物品m j的评分；I g∩I i表示用户u g和用户u i共同评价过的商品集合。
  4. 如权利要求3所述的商品推荐方法，其特征在于，所述根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据包括：
    在目标用户为u t∈U，给所述目标用户u t推荐待推荐商品m a∈{m j|m j∈M∧r tj＝0}时，预测目标用户u t对所述待推荐商品m a的评分
    Figure PCTCN2020075477-appb-100008
    为：
    Figure PCTCN2020075477-appb-100009
    其中，m a∈{m j|m j∈M∧r tj＝0}代表用户未曾评分的商品，∧是条件连接符号，r ca表示用户u c对商品m a的评分，
    Figure PCTCN2020075477-appb-100010
    表示目标用户u t对商品非零评分的平均值，
    Figure PCTCN2020075477-appb-100011
    表示近邻用户u c对商品非零评分的平均值，S(u t,u c)表示目标用户u t和近邻用户u c的用户相似性，NK是目标用户u t的近邻用户数集，包含与目标用户的用户相似度从高到低选取的K个近邻用户，W ca为近邻用户u c对待推荐商品m a的评分信任度且W ca＝decay itear，decay为衰减率，itear为迭代次数，每一轮新添的分值对应一个信任系数，每一轮乘以衰减率。
  5. 一种商品推荐装置，其特征在于，包括：
    获取模块，用于获取不同用户分别针对不同商品的评分数据，所述评分数据用于表征用户对商品的喜欢程度；
    计算模块，用于根据不同用户对商品的评分数据，对所述不同用户之间的用户相似度进行计算；
    推荐模块，用于根据目标用户的近邻用户对待推荐商品的评分数据得到所述目标用户对所述待推荐商品的评分数据，以对商品进行推荐。
  6. 如权利要求5所述的商品推荐装置，其特征在于，所述获取模块用于：
    采用集合U＝{u 1,...,u l}和M＝{m 1,...,m n}分别代表用户集合与商品集合，l和n分别表示用户数量和商品数量，输入商品评分矩阵
    Figure PCTCN2020075477-appb-100012
    其中r ij∈{0,s}表示用户u i对商品m j的评分；
    若r ij取值为0则表示用户对商品未评价，若r ij取非零值则表示用户对商品评价的分值。
  7. 如权利要求6所述的商品推荐装置，其特征在于，所述计算模块用于：
    对任意两个用户u g和u i，采用S(u g,u i)＝A(u g,u i)C(u g,u i)PCC(u g,u i)计算二者之间的用户相似度；
    其中，A(u g,u i)表示用户u g和用户u i的一致性程度，计算方式为：
    Figure PCTCN2020075477-appb-100013
    V g表示基于用户u g的评分向量，
    Figure PCTCN2020075477-appb-100014
    中的上角标T表示转置，V i表示基于用户u i的评分向量；C(u g,u i)表示折衷因素，计算方式为：
    Figure PCTCN2020075477-appb-100015
    I g为用户u g评价过的物品集合，I i为用户u i评价过的物品集合；PCC(u g,u i)是计算两个用户之间的Pearson相似度，计算方式为：
    Figure PCTCN2020075477-appb-100016
    其中，r gj是用户u g对物品m j的评分，
    Figure PCTCN2020075477-appb-100017
    表示用户u i对商品非零评分的平均值，
    Figure PCTCN2020075477-appb-100018
    代表用户u g对商品非零评分的平均值；r ij是用户u i对物品m j的评分；I g∩I i表示用户u g和用户u i共同评价过的商品集合。
  8. 如权利要求7所述的商品推荐装置，其特征在于，所述推荐模块用于：
    在目标用户为u t∈U，给所述目标用户u t推荐待推荐商品m a∈{m j|m j∈M∧r tj＝0}时，预测目标用户u t对所述待推荐商品m a的评分
    Figure PCTCN2020075477-appb-100019
    为：
    Figure PCTCN2020075477-appb-100020
    其中，m a∈{m j|m j∈M∧r tj＝0}代表用户未曾评分的商品，∧是条件连接符号，r ca表示用户u c对商品m a的评分，
    Figure PCTCN2020075477-appb-100021
    表示目标用户u t对商品非零评分的平均值，
    Figure PCTCN2020075477-appb-100022
    表示近邻用户u c对商品非零评分的平均值，S(u t,u c)表示目标用户u t和近邻用户u c的用户相似性，NK是目标用户u t的近邻用户数集，包含与目标用户的用户相似度从高到低选取的K个近邻用户，W ca为近邻用户u c对待推荐商品m a的评分信任度且W ca＝decay itear，decay为衰减率，itear为迭代次数，每一轮新添的分值对应一个信任系数，每一轮乘以衰减率。
  9. 一种商品推荐设备，其特征在于，包括：
    存储器，用于存储计算机程序；
    处理器，用于执行所述计算机程序时实现如权利要求1至4任一项所述商品推荐方法的步骤。
  10. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1至4任一项所述商品推荐方法的步骤。
PCT/CN2020/075477 2019-02-18 2020-02-17 一种商品推荐方法、装置、设备以及存储介质 Ceased WO2020168992A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910123041.2A CN109658210A (zh) 2019-02-18 2019-02-18 一种商品推荐方法、装置、设备以及存储介质
CN201910123041.2 2019-02-18

Publications (1)

Publication Number Publication Date
WO2020168992A1 true WO2020168992A1 (zh) 2020-08-27

Family

ID=66123097

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/075477 Ceased WO2020168992A1 (zh) 2019-02-18 2020-02-17 一种商品推荐方法、装置、设备以及存储介质

Country Status (2)

Country Link
CN (1) CN109658210A (zh)
WO (1) WO2020168992A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109658210A (zh) * 2019-02-18 2019-04-19 苏州大学 一种商品推荐方法、装置、设备以及存储介质
CN111523590B (zh) * 2020-04-21 2023-04-07 莫毓昌 面向评分矩阵的多数据源融合方法
CN111695040B (zh) * 2020-06-12 2023-03-24 陕西师范大学 一种基于情感标签的时尚品推荐方法、系统及装置
CN112115359B (zh) * 2020-09-16 2023-10-31 苏州大学 一种基于多阶近邻预测的推荐系统及方法
CN113535790A (zh) * 2021-07-19 2021-10-22 成都数之联科技有限公司 协同推荐优化方法、装置、电子设备和可读存储介质
CN113672813B (zh) * 2021-08-26 2024-08-02 武汉卓尔数字传媒科技有限公司 物品推荐方法、装置、终端及存储介质
CN114090900A (zh) * 2021-11-05 2022-02-25 中国联合网络通信集团有限公司 确定用户可信度的方法、装置、设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101271559A (zh) * 2008-05-16 2008-09-24 华东师范大学 一种基于用户局部兴趣挖掘的协作推荐系统
CN101482884A (zh) * 2009-01-21 2009-07-15 华东师范大学 一种基于用户偏好评分分布的协作推荐系统
CN104281956A (zh) * 2014-10-27 2015-01-14 南京信息工程大学 基于时间信息的适应用户兴趣变化的动态推荐方法
CN104462597A (zh) * 2014-12-31 2015-03-25 湖南大学 一种综合用户正负性评分和评分偏好因子的协同过滤算法
US20150339756A1 (en) * 2014-05-23 2015-11-26 Ebay Inc. Personalizing alternative recommendations using search context
CN109658210A (zh) * 2019-02-18 2019-04-19 苏州大学 一种商品推荐方法、装置、设备以及存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102033867B (zh) * 2010-12-14 2012-07-25 西北工业大学 用于xml文档分类的语义相似度度量方法
CN105069122B (zh) * 2015-08-12 2018-08-21 天津大学 一种基于用户行为的个性化推荐方法及其推荐装置
CN105205128B (zh) * 2015-09-14 2018-08-28 清华大学 基于评分特征的时序推荐方法及推荐装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101271559A (zh) * 2008-05-16 2008-09-24 华东师范大学 一种基于用户局部兴趣挖掘的协作推荐系统
CN101482884A (zh) * 2009-01-21 2009-07-15 华东师范大学 一种基于用户偏好评分分布的协作推荐系统
US20150339756A1 (en) * 2014-05-23 2015-11-26 Ebay Inc. Personalizing alternative recommendations using search context
CN104281956A (zh) * 2014-10-27 2015-01-14 南京信息工程大学 基于时间信息的适应用户兴趣变化的动态推荐方法
CN104462597A (zh) * 2014-12-31 2015-03-25 湖南大学 一种综合用户正负性评分和评分偏好因子的协同过滤算法
CN109658210A (zh) * 2019-02-18 2019-04-19 苏州大学 一种商品推荐方法、装置、设备以及存储介质

Also Published As

Publication number Publication date
CN109658210A (zh) 2019-04-19

Similar Documents

Publication Publication Date Title
WO2020168992A1 (zh) 一种商品推荐方法、装置、设备以及存储介质
CN106503022B (zh) 推送推荐信息的方法和装置
Liu et al. Wisdom of the better few: cold start recommendation via representative based rating elicitation
Ning et al. A comprehensive survey of neighborhood-based recommendation methods
CN104317835B (zh) 视频终端的新用户推荐方法
Ocepek et al. Improving matrix factorization recommendations for examples in cold start
WO2020207074A1 (zh) 一种信息推送的方法及设备
CN109840833B (zh) 贝叶斯协同过滤推荐方法
WO2015032353A1 (zh) 视频推荐方法及装置
WO2020238502A1 (zh) 物品推荐方法及装置、电子设备及存储介质
CN111797319B (zh) 推荐方法、装置、设备及存储介质
Valcarce et al. Additive smoothing for relevance-based language modelling of recommender systems
CN110287373A (zh) 基于评分预测和用户特征的协同过滤电影推荐方法及系统
US20180025364A1 (en) Information processing apparatus, information processing method, and program
CN110347935A (zh) 基于用户兴趣变化的个性化影视项目推荐方法及系统
Zhou et al. The power of ground user in recommender systems
US20140365456A1 (en) Item-based recommendation engine for recommending a highly-associated item
Ai et al. Measuring similarity based on user activeness in recommender systems to improve algorithm scalability
CN110309361B (zh) 一种视频评分的确定方法、推荐方法、装置及电子设备
CN113837846B (zh) 商品推荐方法、装置、计算机设备和存储介质
CN104794135B (zh) 一种对搜索结果进行排序的方法和装置
CN110457592A (zh) 一种基于图熵的社交网络推荐方法
CN114201687A (zh) 信息推荐方法、装置、电子设备及存储介质
CN104951956A (zh) 一种电影排名方法和装置
CN112115359A (zh) 一种基于多阶近邻预测的推荐系统及方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20758990

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20758990

Country of ref document: EP

Kind code of ref document: A1