CN107292695B - Collaborative filtering recommendation method based on SRV function - Google Patents

Collaborative filtering recommendation method based on SRV function Download PDF

Info

Publication number
CN107292695B
CN107292695B CN201710257526.1A CN201710257526A CN107292695B CN 107292695 B CN107292695 B CN 107292695B CN 201710257526 A CN201710257526 A CN 201710257526A CN 107292695 B CN107292695 B CN 107292695B
Authority
CN
China
Prior art keywords
users
user
distance
classifying
new user
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.)
Active
Application number
CN201710257526.1A
Other languages
Chinese (zh)
Other versions
CN107292695A (en
Inventor
张超
张亮
李俊清
霍明
柳平增
张蕾
滕琳
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.)
Shandong Agricultural University
Original Assignee
Shandong Agricultural 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 Shandong Agricultural University filed Critical Shandong Agricultural University
Priority to CN201710257526.1A priority Critical patent/CN107292695B/en
Publication of CN107292695A publication Critical patent/CN107292695A/en
Application granted granted Critical
Publication of CN107292695B publication Critical patent/CN107292695B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Accounting & Taxation (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Finance (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a collaborative filtering recommendation algorithm based on an SRV function, which comprises the following steps: classifying all perception attributes of the users according to data of all the users, and specifically classifying the users by using a K-means clustering method; extracting the characteristics of the category to obtain the representative elements of the category, and then obtaining a corresponding spider-web graph; firstly, calculating the distance of each representative element of the added new user to see whether the new user is the user of the type, classifying the new user if the new user is the user of the type, and turning to the next step if the new user is not the user of the type; and rotating the new user, wherein the rotation angle can be determined according to the number of the attributes each time until the optimal rotation angle is found. The invention has the characteristics of high efficiency and high accuracy.

Description

Collaborative filtering recommendation method based on SRV function
Technical Field
The invention belongs to the technical field of internet, and particularly relates to a collaborative filtering recommendation method based on an SRV function.
Background
A new shopping mode is started in a sales mode mainly based on an e-commerce in the Internet age, people can freely select favorite articles on the Internet, but accompanying with abundant commodities, various promotion advertisements, red point recommendation whether webpage popup or WeChat or recommendation in mails are sufficient, normal online chatting cannot be performed basically, articles can be selected on websites, the goods are interrupted for many times, the main reason is overload of information brought by online shopping, but as a user, the user cannot browse and check information one by one to judge advantages and disadvantages, and purchases after comparison, because the general consumer has a not clear consumption tendency and is easily influenced by other factors, but as the e-commerce, the online emotion and the shopping emotion of the consumer are damaged once and again, so that the soy sauce money finally appears to buy vinegar, or buying a lot of useless goods instead of buying a lot of things originally, which wastes time and costs more money, and in turn only drinks 40489to the e-commerce, and slaking thirst, which is not good for both parties because the buying emotion of hurting the consumer repeatedly will have great adverse effect, and the idea of the consumer is to buy the good-minded commodity, while the e-commerce is to circulate the commodity faster, so that the system intervention is recommended to relieve the conflict between the two parties.
Current recommendation systems fall into two main categories:
the first type is a recommendation system based on users, which is called as a collaborative filtering method, the main method is to recommend the user goods to the user according to the purchasing habit of the close vicinity of the user or the friend circle of the user, the method is characterized in that the method has a certain shopping similarity to the user or is a method which is very good for the user to recommend the use condition of the goods to the user by the friends of the user, but along with the continuous expansion of the user range, particularly the WeChat friend circle is continuously occupied by merchants such as WeChat merchants and the like, so that the satisfaction degree of the user is gradually reduced, particularly when the WeChat merchants exist in the friend circle of the user, the trust degree of the user to the goods is seriously injured, and along with the gradual growth of the nine-zero later generation, which is called as the consumer force army, the user is a real digital generation, so to say that no mobile phone can live, but the person is a generation which is different from the former generation with very obvious personality, the method is characterized in that various curiosity and oddness are liked, particularly so-called non-mainstream, so that the unique website is liked and the situation is unwilling to be popular in any aspect, and therefore the phenomenon called long tail phenomenon appears in the sales situation of a large number of large websites, namely the sales volume of some commodities is not as large as that of hot-sold commodities except the hot-sold commodities, but the sales volumes of the commodities are added together but can be as large as that of the hot-sold commodities, sometimes even exceed the hot-sold commodities, so that the sales recommendation of the long tail commodities is the key point of the websites, but the accurate recommendation in the existing situation can be realized, and the collaborative filtering method is obviously not suitable because the personal characteristics of each user are ignored, and therefore, the refining operation is required on the basis of the original method.
The second type is an item-based recommendation system, which is mainly characterized in that related commodities or the same commodities of different brands or surrounding items related to searched commodities are recommended to a user according to past search records of the user, the main concern of the item is that the commodity is not associated with the use habits of the user, most websites adopt a recommendation algorithm which collects the internet browsing history of the user through a background, certainly, the recommendation algorithm is available for the user, can help the user to find the needed commodities or the prices of the commodities or the surrounding products which are not considered by the user in time, and is useful for real-time search of the user, but the search of the user is random, the search of the website can also be carried out on other websites, so the search cannot be fixed for a long time, and the website can grasp the consumption habits of the user and can provide the user with very good shopping experience The user's good recommendation system is retained.
Based on the two recommendation systems, it can be seen that both the collaborative filtering method and the article-based recommendation system have disadvantages, the former ignores the personality characteristics of the user and is particularly obvious in the current user group, and the latter cannot leave out that the user only has randomness and contingency for a long time, so that a new algorithm must be re-designed to improve the recommendation system to adapt to new changes and situations, while the SRV function-based recommendation system algorithm is designed by re-fitting the advantages of the two while eliminating some disadvantages of the two.
Disclosure of Invention
The invention aims to provide a collaborative filtering recommendation method based on an SRV function, which improves the traditional collaborative filtering method, and classifies the characteristics of users more deeply instead of simply comparing the distance or the relationship degree of friends, so that the efficiency is higher, and simultaneously, new classification is carried out on articles to ensure the accuracy of recommendation.
The specific technical scheme is as follows:
a collaborative filtering recommendation method based on an SRV function includes the following steps according to three different conditions:
in the first case: the attributes and scores not corresponding one-to-one
Step1, classifying all perception attributes of the users according to the data of all the users, dividing the perception attributes of the users into nine sensory experiences such as brands, advertisements, quality, express delivery speed and the like, and obtaining specific scores according to historical data;
step2, classifying users by using a K-means clustering method according to the scoring of the first Step, and obtaining representative elements of each class, namely K1, K2, Kn;
step3 gives specific scores of the representative elements in the second Step as K1(K11, K12, K13.. K19) and then obtains a corresponding spider-web map;
step4 marks the new user as N (N1, N2.. N9), first applies the distance formula
Figure GDA0002734122410000031
Wherein
Figure GDA0002734122410000032
Calculating the distance of each representative element, calculating the distance between each representative element and each representative element in the second step to obtain the minimum distance value, wherein the user belongs to the category, and if not, the next step is carried out;
step5, each time a new user rotates, the rotation angle can be determined according to the attributeUntil the optimum rotation angle is found.
Figure GDA0002734122410000033
Wherein
Figure GDA0002734122410000034
And O is a rotation matrix, and the minimum value of the distance at the moment can be finally determined by an iteration method.
In the second case: each attribute matching the score
Step1, classifying all perception attributes of the users according to the data of all the users, dividing the perception attributes of the users into nine sensory experiences such as brands, advertisements, quality, express delivery speed and the like, and obtaining specific scores according to historical data;
step2, classifying users by using a K-means clustering method according to the scoring of the first Step, and obtaining representative elements of each class, namely K1, K2, Kn;
and Step3 pushing according to the corresponding attributes and the scores.
In the third case: a certain attribute missing score
Step1, classifying all perception attributes of the users according to the data of all the users, dividing the perception attributes of the users into nine sensory experiences such as brands, advertisements, quality, express delivery speed and the like, and obtaining specific scores according to historical data;
step2, classifying users by using a K-means clustering method according to the scoring of the first Step, and obtaining representative elements of each class, namely K1, K2, Kn;
step3 adds the missing attribute. Assuming that X is a fully informative variable and Y is a variable with missing values, K or a subset thereof is first clustered and then the mean of the different classes is interpolated for the class to which the missing individual belongs.
Compared with the prior art, the invention has the beneficial effects that:
the invention improves the traditional collaborative filtering method, and deeper classification is carried out according to the characteristics of the user instead of simple comparison according to the distance or the relationship degree of friends, so that the efficiency is higher, and meanwhile, new classification is carried out on articles to ensure the accuracy of recommendation.
The method is more accurate in classification due to the fact that a more accurate vector method is adopted in technical implementation, and meanwhile, a better solution can be provided for the classification of unknown users through a rotation error correction method.
The method mainly adopts MATLAB software to unitize vectors, can adopt a database to collect and process, and can use a new processing method similar to a graph when solving and processing unknown users, so that the method is more intuitive and novel, and has better processing precision.
Drawings
FIG. 1 is a flowchart of a collaborative filtering recommendation method based on SRV function according to the present invention, wherein FIG. 1a is a first scenario and FIG. 1b is a second scenario;
FIG. 2 is a spider-web diagram of a user, wherein FIG. 2a is a spider-web diagram of a user 1, FIG. 2b is a spider-web diagram of a user 2, and FIG. 2c is a spider-web diagram of a user; fig. 2d is a user 4 spider-web diagram.
Detailed Description
The technical solution of the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
As shown in FIG. 1, the method comprises the following steps:
step1 classifies all the perception attributes of the users according to the data of all the users, and specifically, the users can be classified firstly by using a K-means clustering method.
Step2 extracts the characteristics of the belonged category to obtain the representative elements of the belonged category, and then the corresponding spider-web graph can be obtained.
Step3 firstly carries out distance calculation of each representative element on the added new user to see whether the new user is the user of the class, if so, the new user is classified, and if not, the next Step is carried out.
Step4 rotates the new user, and each rotation angle can be decided according to the attribute until finding the optimal rotation angle.
Wherein FIG. 1a is the first case
Two closed curves, properties
Figure GDA0002734122410000051
And score
Figure GDA0002734122410000052
1、k＝0；
2. The for cycle is repeated when i 1, 2.., w,
3. change of
Figure GDA0002734122410000061
And obtain
Figure GDA0002734122410000062
4. According to K and
Figure GDA0002734122410000063
calculating the rotation angle O(i，k)
5. Suppose that
Figure GDA0002734122410000064
6. Calculating K and by DP
Figure GDA0002734122410000065
Gamma of (2)(i，k)
7. Calculating a function H(i，k)
8. End for cycle
9. Looking for H(min，k)＝min1≤i≤w{H(i，k)And give a consistent O(min，k)、γ(min，k)And
Figure GDA0002734122410000066
10. adding by function F
Figure GDA0002734122410000067
And assume that
Figure GDA0002734122410000068
11. If the stop criterion is met, then stop, otherwise let k +1 and continue with step 2.
FIG. 1b shows the second case
1、k＝0；
2. Finding the rotation angle O by SVDk+1＝argminOHo(O，γk)；
3. Finding gamma using dynamic proceduresk+1≈argminγHo(Ok+1，γ)；
4. If the stop criterion is met, then stop, otherwise let k +1 and continue with step 2.
The method mainly divides the commodity recommending process into two steps, wherein in the first step, the users are mainly classified and clustered according to different characteristics of the users, and in the second step, browsing initiated by the users in the purchasing process is identified and recommended according to the characteristics of the users, so that the whole commodity recommending process is completed.
In the first step, the users are firstly classified according to the occupation of the users, then the information of the users is collected and then all the users are classified, the users are classified into a brand recognition type, an appearance recognition type, a quality type, a price sensitive type, a function recognition type, an express speed type, a region recognition type, a style recognition type, an advertising effect type and a middle-inferior type according to the standard, or the users are overlapped according to the characteristics of some types, for example, the express speed type can be overlapped with the style recognition type and can also be overlapped with the advertising effect type, and the detailed classification is shown in an attached table.
In the second step, reasonable recommendation is performed for browsing initiated by the user, not only simple recommendation on brands, such as that the user wants to buy one pair of shoes and browse one nike, but also adidas is given during recommendation, information of the user needs to be evaluated in advance and then recommended, and clustering is performed according to some information of the user instead of seeing the brands.
c) The principle is explained as follows:
the principle of the recommendation algorithm is mainly that according to the distance between closed graphs, firstly, information of each user is changed into a closed graph by utilizing spider-web graphs in EXCEL, then, the users are clustered and classified by utilizing the distance constructed by an SRV function, the distance between the graph of the new user and the graph of the known class can be calculated for the new user, and the minimum value is the class which can be regrouped in a rotation error correction mode if the class has errors.
The algorithm is as follows:
phi SRV function
Let β (t): d → RnIs a parameterized function with a domain D of closed intervals [0,1 ]]Value range of Rn
Setting F: rn→RnAnd is
Figure GDA0002734122410000071
Where | · | | is generally taken as a two-norm in euclidean space, then the SRV function can be defined as the following form q: d → RnWherein
Figure GDA0002734122410000072
And satisfies the following conditions
Figure GDA0002734122410000073
The visual understanding of the SRV function is defined in the closed interval [0,1 ]]Has a length of 1 and is L2A proper subset of the space.
② definition of distance between two closed figures
For clustering analysis, all curves are classified first, and the classification criterion is given according to the distance between closed curves, so the distance between curves is defined, and because there is a certain similarity between two closed images, the following distances can be defined:
Figure GDA0002734122410000081
wherein
Figure GDA0002734122410000082
The discrete case is represented by the inner product of the two.
③ user addition after clustering
Assuming that the number of standard users in the clustered database is N, the representative sequence of the users is v1(t)，v2(t)，...，vN(t), for the newly added user as α (t), the minimum distance between α (t) and the first N criteria is first obtained as the category, and if the system prompts an error or an obvious deviation occurs during recommendation, the following rotation error correction can be performed.
Rotation error correction
If the newly added user's product selection is not in a serious match with the previous representative user,
Figure GDA0002734122410000085
wherein
Figure GDA0002734122410000083
And O is a rotation matrix, and the minimum value of the distance at the moment can be finally determined by an iteration method.
And (3) operation experiment:
1. importing of data
As shown in table 1.
TABLE 1
Figure GDA0002734122410000084
2. Index interpretation
As shown in table 2.
TABLE 2
Figure GDA0002734122410000091
3. Calculating a score
The scoring relationships were calculated for only the four users in fig. 2 above, as shown in table 3:
TABLE 3
1 2 3 4
1 1 0.2645 0.1652 0.1914
2 0.2645 1 0.1713 0.1789
3 0.1652 0.1713 1 0.1519
4 0.1914 0.1789 0.1519 1
From the calculation results, the user 4 and the user 3 are closest to each other, so that the commodities purchased by the two users can be considered to be consistent under certain conditions, the user 1 can be recommended the same type of commodity of the user 2, the user 1 and the user 2 are far away from each other, the two commodities are classified into different categories, the table shows that the requirement of the user 1 on the commodity is high, the requirement on brand, advertisement and particularly express speed is high, the style is biased to the manifold style, and the user 2 is biased to the quality type, so that the two commodities are greatly different.
For example, when the user 1 selects the milk powder, the user 1 is not aware of the purchasing intention of the user 2, but the user 1 can be recommended to purchase the product of the user 2 more accurately because the purchasing intention of the user 2 is known.
From the above calculation results, it can be seen that the reliability of the method for user classification is very good.
If the second kind of situation occurs, that is, the classification of the unknown user is determined, the rotation method is used, and the specific algorithm is as follows:
taking the above four data as an example, the first three are assumed to be known categories, but the type of the user 4 is unknown, that is, the score of the user 4 is not a score matching each attribute but an estimated score, so that the distance cannot be calculated by the original inner product, and only the rotation error correction can be performed.
For example, the relationship between the user 4 and the user 1 is calculated, an inner product of the user 4 and the user 1 is calculated first, then an image of the user 4 is rotated by an angle θ, then a new inner product is calculated, then the angle 2 θ is rotated, the.
The above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited thereto, and any simple modifications or equivalent substitutions of the technical solutions that can be obviously obtained by those skilled in the art within the technical scope of the present invention are within the scope of the present invention.

Claims (1)

1. A collaborative filtering recommendation method based on SRV function is characterized in that a recommendation process of commodities is divided into two steps, in the first step, users are mainly classified and clustered, reasonable division is carried out according to different characteristics of the users, and in the second step, browsing initiated by the users in a purchasing process is identified and recommended according to characteristics of the users, so that the whole recommendation process of the commodities is completed;
the method comprises the following steps:
in the first case: the attributes and scores not corresponding one-to-one
Step1, classifying all perception attributes of the users according to the data of all the users, dividing the perception attributes of the users into nine sensory experiences of brands, advertisements, appearances, qualities, prices, functions, express speeds, regions and styles, and respectively obtaining specific scores according to historical data;
step2, classifying the users by using a K-means clustering method according to the scoring of the first Step, and obtaining the representative elements of each class as K1, K2, … and Kn;
step3 gives specific scores of the representative elements in the second Step as K1(K11, K12, K13, … K19) and then obtains a corresponding spider-web map;
step4 marks the new user as N (N1, N2.. N9), first applies the distance formula
Figure FDA0002734122400000014
Wherein
Figure FDA0002734122400000011
Calculating the distance of each representative element, calculating the distance between each representative element and each representative element in the second step to obtain the minimum distance value, wherein the user belongs to the category, and if not, the next step is carried out;
step5, determining the rotation angle of the new user according to the attribute of the new user each time until finding the optimal rotation angle;
Figure FDA0002734122400000012
wherein
Figure FDA0002734122400000013
Wherein O is a rotation matrix, and the minimum value of the distance at the moment is finally determined by an iteration method;
in the second case: each attribute matching the score
Step1, classifying all perception attributes of the users according to the data of all the users, dividing the perception attributes of the users into nine sensory experiences of brands, advertisements, appearances, qualities, prices, functions, express speeds, regions and styles, and respectively obtaining specific scores according to historical data;
step2, classifying the users by using a K-means clustering method according to the scoring of the first Step, and obtaining the representative elements of each class as K1, K2, … and Kn;
step3 pushing according to the corresponding attribute and the score;
in the third case: a certain attribute missing score
Step1, classifying all perception attributes of the users according to the data of all the users, dividing the perception attributes of the users into nine sensory experiences of brands, advertisements, appearances, qualities, prices, functions, express speeds, regions and styles, and respectively obtaining specific scores according to historical data;
step2, classifying the users by using a K-means clustering method according to the scoring of the first Step, and obtaining the representative elements of each class as K1, K2, … and Kn;
step3 adding the missing attribute; assuming that X is a complete-information variable and Y is a variable with a missing value (K1, K2, …, K (n-1)), K or a subset thereof is first clustered, and then the mean values of different classes are interpolated according to the class to which the missing individual belongs;
according to the distance between the closed graphs, firstly, the information of each user is changed into a closed graph by utilizing a spider-web graph in EXCEL, then, the users are clustered and classified by utilizing the distance constructed by an SRV function, the distance between the graph of the new user and the graph of the known class can be calculated for the new user, the minimum value is determined classification, and if the class has errors, the new user can be regrouped in a rotary error correction mode;
description of the drawings:
phi SRV function
Let β (t) D → RnIs a parameterized function with a domain D of closed intervals [0,1 ]]Value range of Rn
If F is Rn→RnAnd is
Figure FDA0002734122400000021
Where | is generally taken as the two-norm of Euclidean space, then the SRV function can be defined as q in the form D → RnWherein
Figure FDA0002734122400000031
And satisfies the following conditions
Figure FDA0002734122400000032
The visual understanding of the SRV function is defined in the closed interval [0,1 ]]Has a length of 1 and is L2A proper subset of the space;
② definition of distance between two closed figures
For clustering analysis, all curves are classified first, and the classification criterion is given according to the distance between closed curves, so the distance between curves is defined, and because there is a certain similarity between two closed images, the following distances can be defined:
Figure FDA0002734122400000036
wherein
Figure FDA0002734122400000033
The discrete case is expressed by the inner product of the two;
③ user addition after clustering
Assuming that the number of standard users in the clustered database is N, the representative sequence of the users is v1(t),v2(t),…,vN(t), for the newly added user marked as α (t), firstly solving the minimum distance between α (t) and the first N standards as the class, and if the system prompts errors or obvious deviation occurs during recommendation, performing the following rotation error correction;
rotation error correction
If the newly added user's product selection is not in a serious match with the previous representative user,
Figure FDA0002734122400000034
wherein
Figure FDA0002734122400000035
And O is a rotation matrix, and the minimum value of the distance at the moment can be finally determined by an iteration method.
CN201710257526.1A 2017-04-19 2017-04-19 Collaborative filtering recommendation method based on SRV function Active CN107292695B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710257526.1A CN107292695B (en) 2017-04-19 2017-04-19 Collaborative filtering recommendation method based on SRV function

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710257526.1A CN107292695B (en) 2017-04-19 2017-04-19 Collaborative filtering recommendation method based on SRV function

Publications (2)

Publication Number Publication Date
CN107292695A CN107292695A (en) 2017-10-24
CN107292695B true CN107292695B (en) 2020-12-04

Family

ID=60093865

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710257526.1A Active CN107292695B (en) 2017-04-19 2017-04-19 Collaborative filtering recommendation method based on SRV function

Country Status (1)

Country Link
CN (1) CN107292695B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108765174A (en) * 2018-05-29 2018-11-06 泰康保险集团股份有限公司 Insure rebating method and device
CN111079639B (en) * 2019-12-13 2023-09-19 中国平安财产保险股份有限公司 Method, device, equipment and storage medium for constructing garbage image classification model

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101859313A (en) * 2009-04-08 2010-10-13 索尼公司 Messaging device and method and program thereof
CN103886001A (en) * 2013-09-22 2014-06-25 天津思博科科技发展有限公司 Personalized commodity recommendation system
CN105956074A (en) * 2016-04-28 2016-09-21 北京航空航天大学 Single image scene six-degree-of-freedom positioning method of adjacent pose fusion guidance

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8661034B2 (en) * 2010-02-03 2014-02-25 Gartner, Inc. Bimodal recommendation engine for recommending items and peers

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101859313A (en) * 2009-04-08 2010-10-13 索尼公司 Messaging device and method and program thereof
CN103886001A (en) * 2013-09-22 2014-06-25 天津思博科科技发展有限公司 Personalized commodity recommendation system
CN105956074A (en) * 2016-04-28 2016-09-21 北京航空航天大学 Single image scene six-degree-of-freedom positioning method of adjacent pose fusion guidance

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
《Coordinate-descent for learning orthogonal matrices through Givens rotations》;Uri Shalit等;《The 31st International Conference for Machine Learning (ICML 2014)》;20131213;全文 *
《基于SVD的协同过滤推荐算法研究》;陈清浩;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160115;全文 *
《基于改进的K_means和奇异值分解的协同过滤研究》;张磊;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160315;全文 *

Also Published As

Publication number Publication date
CN107292695A (en) 2017-10-24

Similar Documents

Publication Publication Date Title
CN103246980B (en) Information output method and server
US10846775B1 (en) Identifying item recommendations through recognized navigational patterns
CN106651546B (en) Electronic commerce information recommendation method oriented to smart community
Sivapalan et al. Recommender systems in e-commerce
CN108664564B (en) Improved collaborative filtering recommendation method based on article content characteristics
TW201501059A (en) Method and system for recommending information
US20220138831A1 (en) Method of Providing Fashion Item Recommendation Service Using User's Body Type and Purchase History
CN102456203A (en) Method for determining candidate product linked list as well as related device
US20150199772A1 (en) Interacting with electronic commerce users using social media
CN104317835A (en) New user recommendation method for video terminal
CN107563857B (en) Customer purchasing behavior analysis method based on logic reasoning negative association rule pruning technology
Kommineni et al. Machine learning based efficient recommendation system for book selection using user based collaborative filtering algorithm
US20090037292A1 (en) Intelligent shopping search system
CN107194769A (en) A kind of Method of Commodity Recommendation that content is searched for based on user
CN110619559A (en) Method for accurately recommending commodities in electronic commerce based on big data information
CN107292695B (en) Collaborative filtering recommendation method based on SRV function
CN104933595A (en) Collaborative filtering recommendation method based on Markov prediction model
CN112651805A (en) Commodity recommendation method and system for online shopping mall
CN110020918B (en) Recommendation information generation method and system
Liao et al. A rough set-based association rule approach implemented on a brand trust evaluation model
KR102378072B1 (en) Method, apparatus and computer program for style recommendation
CN111507804B (en) Emotion perception commodity recommendation method based on mixed information fusion
CN111192112A (en) Multi-platform interaction method and device
CN112116421A (en) Method and system for optimizing and recommending commodities in self-built welfare mall
Mishra et al. Solving cold start problem using MBA

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant