CN109299370B - Multi-pair level personalized recommendation method - Google Patents

Multi-pair level personalized recommendation method Download PDF

Info

Publication number
CN109299370B
CN109299370B CN201811172906.6A CN201811172906A CN109299370B CN 109299370 B CN109299370 B CN 109299370B CN 201811172906 A CN201811172906 A CN 201811172906A CN 109299370 B CN109299370 B CN 109299370B
Authority
CN
China
Prior art keywords
user
commodities
commodity
users
preference
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
CN201811172906.6A
Other languages
Chinese (zh)
Other versions
CN109299370A (en
Inventor
陈恩红
刘淇
于润龙
叶雨扬
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.)
University of Science and Technology of China USTC
Original Assignee
University of Science and Technology of China USTC
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 University of Science and Technology of China USTC filed Critical University of Science and Technology of China USTC
Priority to CN201811172906.6A priority Critical patent/CN109299370B/en
Publication of CN109299370A publication Critical patent/CN109299370A/en
Application granted granted Critical
Publication of CN109299370B publication Critical patent/CN109299370B/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

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)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a multi-pair level personalized recommendation method, which comprises the following steps: extracting implicit feedback information of a user through an internet platform; determining the preference degree of the user to each commodity according to the extracted implicit feedback information of the user, so that the implicit feedback information of the user is divided into a positive feedback set, a negative feedback set and an unknown set, and the three sets are used as training data; optimizing the preference degree of the user to the commodities by adopting a random gradient descent algorithm and combining training data so as to obtain the optimized preference degree of the user to each commodity; and sorting the commodities according to the preference degrees of the commodities from high to low, and recommending a plurality of commodities which are ranked at the top as the commodities liked by the user to the user. The method can be used for mining the information of the commodities which are potentially interesting to the user, and recommending the commodities which are liked by the user to each user in a list form by adopting a personalized recommendation method.

Description

Multi-pair level personalized recommendation method
Technical Field
The invention relates to the field of machine learning and recommendation systems, in particular to a multi-pair level personalized recommendation method.
Background
Collaborative filtering algorithms are one of the most common algorithms in recommendation systems. The previous research focuses more on a collaborative filtering algorithm based on user scoring data, but in more application scenes in life, the user scoring on commodities is difficult to obtain. For example, a user purchasing record is owned on a "cat network", a user attention record is owned on a "microblog" platform, and a user browsing record is owned on an "love art", and the user history records do not contain explicit scoring information, which is called as implicit feedback of the user. Unlike the user's rating data, where implicit feedback only contains positive feedback information about the goods by the user, the large amount of feedback information that is not observed by the commerce platform cannot be easily understood as negative feedback by the user about the goods, since the user is likely not to find the goods, rather than the user disliking the goods.
Since the recommendation system based on implicit feedback lacks a lot of negative feedback information, especially in the case of sparse data, various researchers have conducted a lot of research work around the subject, wherein the point-level regression model and the pair-level ranking model achieve the best recommendation effect.
The point-level regression model takes implicit feedback as the absolute preference value of the user to the commodity, and adopts the point-level quadratic loss function minimization to approximate the absolute preference value of the user to the commodity. However, the training efficiency of the point-level regression model is low, when large-scale users feed back data, a good model cannot be provided in effective duration, and experimental results show that the prediction effect of the point-level regression model on volatile data is poor, and the recommendation effect is greatly influenced by the initialized weight.
The rank-ordering models take the preference relationship between the user and each pair of commodities as basic units, model commodity feedback information observed and commodity feedback information not observed on the platform, and try to maximize the likelihood function of preference hypothesis between the commodity pairs, so as to give a commodity list in which the user is interested. The Bayes personalized recommendation algorithm is the most common algorithm adopting a pair-level ranking model, and a lot of research works in recent years are developed around the Bayes personalized recommendation algorithm and have good application value. However, these rank ordering models suggest that the user prefers those items that have been given positive feedback over those that have not been consumed, thereby ignoring potential user-liked ones of the large number of unobserved items, limiting the understanding and use of the recommendation system to the user's preferences.
Disclosure of Invention
The invention aims to provide a multi-pair level personalized recommendation method, which can be used for mining the commodity information which is potentially interesting to a user and recommending favorite commodities of the user to each user in a list form by adopting a personalized recommendation method.
The purpose of the invention is realized by the following technical scheme:
a multi-pair level personalized recommendation method comprises the following steps:
extracting implicit feedback information of a user through an internet platform;
determining the preference degree of the user to each commodity according to the extracted implicit feedback information of the user, so that the implicit feedback information of the user is divided into a positive feedback set, a negative feedback set and an unknown set, and the three sets are used as training data;
optimizing the preference degree of the user to the commodities by adopting a random gradient descent algorithm and combining training data so as to obtain the optimized preference degree of the user to each commodity;
and sorting the commodities according to the preference degrees of the commodities from high to low, and recommending a plurality of commodities which are ranked at the top as the commodities liked by the user to the user.
According to the technical scheme provided by the invention, commodities which do not generate purchasing behaviors of each user are treated differently, the preference of the user for the commodities is divided into two parts according to the condition that the user does not see or dislike the commodities, the potential consumption demand that the user loses the purchasing behaviors because the user does not know commodity information is deeply mined, and finally, a commodity recommendation list which reflects the preference degree of the user is provided.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a flowchart of a multi-pair level personalized recommendation method according to an embodiment of the present invention;
fig. 2 is a schematic diagram illustrating commodity data partitioning according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a preference relationship of a multi-pair personalized ranking according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention provides a multi-pair level personalized recommendation method, as shown in fig. 1, which mainly comprises the following steps:
step 1, extracting implicit feedback information of a user through an internet platform.
The preferred embodiment of this step is as follows:
firstly, crawling click records and purchase records of users on commodities from an online shopping platform, crawling attention records of users on other users from a social platform and crawling browsing records of users on online media from an online media platform by adopting a crawler technology; regarding the social platform, regarding other users as commodities, and regarding the attention operation as a purchase operation of the commodities; for the online media platform, online media is regarded as commodities, and browsing operation of a user on the online media is regarded as purchasing operation.
Then, preprocessing the data to construct a user-commodity matrix; assuming that a user is u and a commodity is i, and an element (u, i) in a user-commodity matrix records the operation history of the user u on the commodity i; using ruiIndicating the preference of the user u for the item i, and if the user u purchases the item i, i.e., (u, i) ═ 1, the user u is considered to express positive feedback for the item i, denoted as rui＝1。
And 2, determining the preference degree of the user for each commodity according to the extracted implicit feedback information of the user, so that the implicit feedback information of the user is divided into a positive feedback set, a negative feedback set and an unknown set, and the three sets are used as training data.
In the embodiment of the invention, all commodity sets I are divided into a positive feedback set, a negative feedback set and an unknown set. Specifically, the method comprises the following steps: 1) taking the set of commodities purchased by the user u as a positive feedback set
Figure GDA0003354712560000031
I.e. the user's preference for these goods is 1. 2) The set of commodities that the user has observed but not purchased is taken as a negative feedback set
Figure GDA0003354712560000032
3) The commodity set with the positive feedback set and the negative feedback set removed is taken as an unknown set
Figure GDA0003354712560000033
It is not clear whether the user u likes these items because the user u is likely to not see information about the part of the items. It should be noted that the negative feedback set
Figure GDA0003354712560000034
And unknown set
Figure GDA0003354712560000035
Are collections of items for which no feedback is observed on the platform.
The preferable dividing mode of the step is as follows:
1) for the online shopping platform, commodities clicked and purchased by the user u are brought into a positive feedback set
Figure GDA0003354712560000036
Bringing goods not purchased in the click record of the user u into a negative feedback set
Figure GDA0003354712560000037
That is, the user has seen the portion of the items but does not choose to purchase them, and thus tends not to recommend the portion of the items to the user; finally, bringing the commodities which are not clicked by the user u into the unknown set
Figure GDA0003354712560000038
2) For the social platform and the online media platform, the times of the users concerned by other users and the times of the online media browsed by all the users are respectively sorted from at least one to select a part of users in the later orderAnd online media as non-streaming set Ie(ii) a For user u, the other users concerned about and the browsed online media are brought into a positive feedback set
Figure GDA0003354712560000041
Non-flow set IeAnd positive feedback set
Figure GDA0003354712560000042
Difference set of
Figure GDA0003354712560000043
(non-rowset I)e-set of positive feedbacks
Figure GDA0003354712560000044
) Incorporating a negative feedback set
Figure GDA0003354712560000045
Other users of interest to user u and browsed online media and difference set
Figure GDA0003354712560000046
The union of (1) and the difference between all users and all online media are included in the unknown set
Figure GDA0003354712560000047
That is, "(all users and all online media) - [ (difference set)
Figure GDA0003354712560000048
) U (other users and browsing online media of user u)]"the results are included in the unknown set
Figure GDA0003354712560000049
As shown in fig. 2, a diagram of commodity data division is shown. User John likes The movies "The Dark Knight" and "Alien" and will thus incorporate a positive feedback set
Figure GDA00033547125600000410
Some movies that are rarely watched representNon-flow set Ie(shaded), then for user John, shaded non-streaming set IeThe difference set from "The Dark Knight" and "Alien" will be incorporated into The negative feedback set
Figure GDA00033547125600000411
(i.e., not running set I)eWith the data other than Alien) that the user John is not interested in the portion of the movie. For other movies, it cannot be determined whether the user John likes, so that all the movies are included in the unknown set
Figure GDA00033547125600000412
In (1).
And 3, optimizing the preference degree of the user to the commodities by adopting a random gradient descent algorithm and combining training data, thereby obtaining the optimized preference degree of the user to each commodity.
The positive feedback is collected
Figure GDA00033547125600000413
The user's preference degree for each commodity is recorded as 1, and a negative feedback set is used
Figure GDA00033547125600000414
The user's preference degree for each commodity is recorded as 0, and the unknown set is recorded
Figure GDA00033547125600000415
Is the user a question mark for the degree of preference for each item? To indicate.
As shown in FIG. 3, for user u, 6 different commodities are sampled in commodity set I, and are respectively marked as I, j, p, p ', q, q ', wherein I, p, p ' belong to a positive feedback set
Figure GDA00033547125600000416
j, q' belong to a negative feedback set
Figure GDA00033547125600000417
q belongs to the unknown set
Figure GDA00033547125600000418
Then for user u, the preference degree for the commodity is: r isui＝rup＝rup′＝1，ruj≈ruq′≈0，ruq? 0 ≦? Less than or equal to 1; where r represents the degree of preference, and the two subscripts correspond to the user and the goods in that order.
According to the preference of the user u for the commodities, a preference relation assumption of a multi-pair personalized ranking algorithm can be given: because user u has expressed positive feedback on item i and user u is likely to dislike items j and q', there will be a preference relationship r whether item q is an item of interest to user u or notui-ruj≥ruq-ruq′. Similarly, the user u likes the items p and p' at the same time, so the difference of the user for their taste should be small, that is, there is a preference relationship r as followsuq-ruq′≥rup-rup′. Using a mark ruij＝rui-rujThe difference value of the preference degree of the user u for the commodity i and the commodity j can be summarized as r according to the preference assumption of a multi-pair level personalized recommendation algorithm (MPR)uij≥ruqq′≥rupp′Then for all users U the following likelihood function is present:
Figure GDA00033547125600000419
the likelihood function contains 3 different commodity pairs, and the preference relationship between purchased commodities and unpurchased commodities of each user is deeply understood. In general, the difference in preference between two items purchased by user u is less than the difference in preference between two items not purchased by user u, which is less than the difference in preference between one item purchased by user u and one item not purchased by user u. Through the comparison of the user on different preference relationship differences among a plurality of commodity pairs, commodities which are possibly interested in the commodity set which is not purchased by the user are mined.
Will r isuij≥ruqq′，ruqq′≥rupp′Expressed as:
λ(ruij-ruqq′)+(1-λ)(ruqq′-rupp′)；
where λ is a balancing factor for balancing two preference assumption targets, the above equation is abbreviated
Figure GDA0003354712560000056
And approximates the probability value Pr (-) using:
Figure GDA0003354712560000051
for the user u, the preference assumption of the multi-pair level personalized recommendation algorithm is abbreviated as:
Figure GDA0003354712560000057
based on the above rules, the likelihood function for optimizing the multi-pair level personalized recommendation algorithm is expressed as:
Figure GDA0003354712560000052
wherein Θ ═ Uu.∈R1×d，Vi.∈R1×d，biE is R, U is U, I is I, is a parameter to be learned by the model, Uu.Is a feature vector, V, describing user ui.Is a feature vector describing the item i, biThe deviation of the characteristic vector of the commodity I is recorded as R, and d is the dimension of the characteristic vector; r (Θ) is a regularization term set to avoid overfitting during training, and is Σu∈U∑t∈S[αu||Uu.||2+αv||Vt.||2+βv||bt||2]S ═ { i, j, p, p ', q, q' } is the sample sampled for each round of training; lnThe MPR is a log-likelihood function of a multi-pair level personalized recommendation algorithm and is expressed as follows:
Figure GDA0003354712560000053
optimizing the likelihood function by adopting a random gradient descent algorithm (SGD), selecting one record in each iteration process, wherein the record comprises a user u and 6 different commodities i, j, p, p ', q and q', updating parameters of a model according to gradient information to achieve the purpose of optimization, and the final optimization function is expressed as:
Figure GDA0003354712560000054
after obtaining the gradient signal, the model parameters are updated by:
Figure GDA0003354712560000055
in the above formula, γ > 0 represents the learning rate, and Θ 'represents the updated model parameter, and the type of the parameter included in the updated model parameter is the same as Θ, and here, in order to distinguish the model parameter before and after updating, a left-handed factor is added to Θ, and in actual operation, the model parameter Θ' after updating in the current round is assigned to Θ for the next round of training.
And 4, sorting the commodities according to the preference degrees of the commodities from high to low, and recommending a plurality of commodities which are ranked at the top as favorite commodities of the user to the user.
Through the previous learning process, the preference degrees of the users to all commodities, including the commodities of the shopping platform, other users of the social platform and the online media of the online media platform can be obtained. Selecting a plurality of favorite commodities of the target user to recommend to the target user according to the favorite degree sequencing result of the user on the commodities; all users are processed according to the steps 1 to 4, so that the purpose of personalized recommendation is achieved for each user.
According to the scheme of the embodiment of the invention, the commodity set is divided according to the historical interactive information of the user and the commodities, the potential shopping requirements of the user for the commodities which are not purchased are deeply mined by comparing the relationship of preference difference values between the user and the commodity pairs, and finally the preference assumption of the multi-pair personalized recommendation algorithm is optimized based on the random gradient descent algorithm, so that the commodities are recommended to the user in a commodity list which is more in line with the requirements of the user.
Through the above description of the embodiments, it is clear to those skilled in the art that the above embodiments can be implemented by software, and can also be implemented by software plus a necessary general hardware platform. With this understanding, the technical solutions of the embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a usb disk, a removable hard disk, etc.), and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods according to the embodiments of the present invention.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (3)

1. A multi-pair level personalized recommendation method is characterized by comprising the following steps:
extracting implicit feedback information of a user through an internet platform;
determining the preference degree of the user to each commodity according to the extracted implicit feedback information of the user, so that the implicit feedback information of the user is divided into a positive feedback set, a negative feedback set and an unknown set, and the three sets are used as training data;
optimizing the preference degree of the user to the commodities by adopting a random gradient descent algorithm and combining training data so as to obtain the optimized preference degree of the user to each commodity;
sorting the commodities according to the preference degrees of the commodities from high to low, and recommending a plurality of commodities which are ranked at the top as favorite commodities of the user to the user;
the method for optimizing the preference degree of the user to the commodity by adopting the random gradient descent algorithm and combining with the training data comprises the following steps:
the positive feedback is collected
Figure FDA0003354712550000011
The user's preference degree for each commodity is recorded as 1, and a negative feedback set is used
Figure FDA0003354712550000012
The user's preference degree for each commodity is recorded as 0, and the unknown set is recorded
Figure FDA0003354712550000013
Is the user a question mark for the degree of preference for each item? To represent;
for a user u, 6 different commodities are sampled in the commodity set I and are respectively marked as I, j, p, p ', q, q ', wherein the I, p, p ' belong to a positive feedback set
Figure FDA0003354712550000014
j, q' belong to a negative feedback set
Figure FDA0003354712550000015
q belongs to the unknown set
Figure FDA0003354712550000016
Then for user u, the preference degree for the commodity is: r isui＝rup＝rup′＝1，ruj≈ruq′≈0，ruq? 0 ≦? Less than or equal to 1; wherein r represents the preference degree, and the two subscripts correspond to the user and the commodity in sequence; using a mark ruij＝rui-rujRepresenting the difference of the preference degree of the user u to the commodity i and the commodity j, and calculating according to the multi-pair level personalized recommendationPreference assumption for FarpR, has ruij≥ruqq′≥rupp′Then for all users U the following likelihood function is present:
Figure FDA0003354712550000017
will r isuij≥ruqq′,ruqq′≥rupp′Expressed as:
λ(ruij-ruqq′)+(1-λ)(ruqq′-rupp′)；
where λ is a balancing factor for balancing two preference assumption targets, the above equation is abbreviated
Figure FDA0003354712550000018
And approximates the probability value Pr (-) using:
Figure FDA0003354712550000019
for the user u, the preference assumption of the multi-pair level personalized recommendation algorithm is abbreviated as:
Figure FDA00033547125500000110
the likelihood function for optimizing the multi-pair level personalized recommendation algorithm is thus represented as:
Figure FDA0003354712550000021
wherein Θ ═ Uu·∈R1×d,Vi·∈R1×d,biE is R, U is U, I is I, is a parameter to be learned by the model, Uu·Is a feature vector, V, describing user ui·Is a feature vector describing the item i, biIs a commodityI, the offset of the characteristic vector is recorded as R, d is the dimension of the characteristic vector; r (Θ) is a regularization term; lnMPR is the log-likelihood function of a multi-pair personalized recommendation algorithm, expressed as:
Figure FDA0003354712550000022
optimizing the likelihood function by adopting a random gradient descent algorithm, selecting a record in each iteration process, wherein the record comprises a user u and 6 different commodities i, j, p, p ', q and q', updating parameters of a model according to gradient information to achieve the purpose of optimization, and the final optimization function is represented as:
Figure FDA0003354712550000023
wherein S ═ { i, j, p, p ', q, q' };
after obtaining the gradient signal, the model parameters are updated by:
Figure FDA0003354712550000024
in the above equation, γ > 0 represents the learning rate, and Θ' represents the updated model parameter.
2. The method of claim 1, wherein the extracting implicit feedback information of the user through the internet platform comprises:
firstly, crawling click records and purchase records of users on commodities from an online shopping platform, crawling attention records of users on other users from a social platform and crawling browsing records of users on online media from an online media platform by adopting a crawler technology; regarding the social platform, regarding other users as commodities, and regarding the attention operation as a purchase operation of the commodities; regarding an online media platform, regarding online media as commodities, and regarding browsing operation of a user on the online media as purchasing operation;
then, preprocessing the data to construct a user-commodity matrix; assuming that a user is u and a commodity is i, and an element (u, i) in a user-commodity matrix records the operation history of the user u on the commodity i; using ruiIndicating the preference of the user u for the item i, and if the user u purchases the item i, i.e., (u, i) ═ 1, the user u is considered to express positive feedback for the item i, denoted as rui＝1。
3. The method according to claim 2, wherein the determining the preference degree of the user for each commodity according to the extracted implicit feedback information of the user, so as to divide the implicit feedback information of the user into a positive feedback set, a negative feedback set and an unknown set comprises:
taking the set of commodities purchased by the user u as a positive feedback set
Figure FDA0003354712550000031
That is, the preference degree of the user for the commodities is 1; the set of commodities that the user has observed but not purchased is taken as a negative feedback set
Figure FDA0003354712550000032
The commodity set with the positive feedback set and the negative feedback set removed is taken as an unknown set
Figure FDA0003354712550000033
For the online shopping platform, commodities clicked and purchased by the user u are brought into a positive feedback set
Figure FDA0003354712550000034
Bringing goods not purchased in the click record of the user u into a negative feedback set
Figure FDA0003354712550000035
Un-clicked by user uIs brought into the unknown set
Figure FDA0003354712550000036
For the social platform and the online media platform, respectively sequencing the times of the users concerned by other users and the times of the online media browsed by all the users from at least to a minimum, and selecting a part of the users and the online media which are sequenced later as a non-streaming set Ie(ii) a For user u, the other users concerned about and the browsed online media are brought into a positive feedback set
Figure FDA0003354712550000037
Non-flow set IeAnd positive feedback set
Figure FDA0003354712550000038
Difference set of
Figure FDA0003354712550000039
Incorporating a negative feedback set
Figure FDA00033547125500000310
Other users of interest to user u and browsed online media and difference set
Figure FDA00033547125500000311
The union of (1) and the difference between all users and all online media are included in the unknown set
Figure FDA00033547125500000312
CN201811172906.6A 2018-10-09 2018-10-09 Multi-pair level personalized recommendation method Active CN109299370B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811172906.6A CN109299370B (en) 2018-10-09 2018-10-09 Multi-pair level personalized recommendation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811172906.6A CN109299370B (en) 2018-10-09 2018-10-09 Multi-pair level personalized recommendation method

Publications (2)

Publication Number Publication Date
CN109299370A CN109299370A (en) 2019-02-01
CN109299370B true CN109299370B (en) 2022-03-01

Family

ID=65161832

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811172906.6A Active CN109299370B (en) 2018-10-09 2018-10-09 Multi-pair level personalized recommendation method

Country Status (1)

Country Link
CN (1) CN109299370B (en)

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109934681B (en) * 2019-03-13 2022-01-11 中国科学技术大学 Recommendation method for user interested goods
CN110362753B (en) * 2019-04-10 2021-12-17 深思考人工智能机器人科技（北京）有限公司 Personalized neural network recommendation method and system based on user implicit feedback
CN110084670B (en) * 2019-04-15 2022-03-25 东北大学 Shelf commodity combination recommendation method based on LDA-MLP
CN110059251B (en) * 2019-04-22 2022-10-28 郑州大学 Collaborative filtering recommendation method based on multi-relation implicit feedback confidence
CN110377783B (en) * 2019-06-05 2023-02-28 深圳大学 Audio and video recommendation method and device and computer equipment
CN110348895A (en) * 2019-06-29 2019-10-18 北京淇瑀信息科技有限公司 A kind of personalized recommendation method based on user tag, device and electronic equipment
CN110489645A (en) * 2019-08-15 2019-11-22 深圳市云积分科技有限公司 A kind of brand association consumer orients the information processing method and device of marketing
CN110503531B (en) * 2019-08-30 2022-04-19 中国科学技术大学 Dynamic social scene recommendation method based on time sequence perception
CN110544129A (en) * 2019-09-05 2019-12-06 创新奇智(青岛)科技有限公司 Personalized recommendation method for social e-commerce users
CN110473543B (en) * 2019-09-25 2020-11-27 北京蓦然认知科技有限公司 Voice recognition method and device
CN110727872A (en) * 2019-10-21 2020-01-24 深圳微品致远信息科技有限公司 Method and device for mining ambiguous selection behavior based on implicit feedback
CN111339327B (en) * 2020-02-20 2024-05-07 北京达佳互联信息技术有限公司 Work recommendation method and device, server and storage medium
CN111598644B (en) * 2020-04-01 2023-05-02 华瑞新智科技（北京）有限公司 Article recommendation method, device and medium
CN111523940B (en) * 2020-04-23 2023-04-18 华中科技大学 Deep reinforcement learning-based recommendation method and system with negative feedback
CN114637922B (en) * 2022-05-17 2022-08-23 浙江大学滨江研究院 User preference recommendation method and device based on neural network set operation
CN114997956B (en) * 2022-06-14 2023-01-10 杭州洋驼网络科技有限公司 Mother and infant product intelligent recommendation system based on big data
CN114780862B (en) * 2022-06-21 2022-09-23 达而观数据(成都)有限公司 User interest vector extraction method, extraction model and computer system

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101383942A (en) * 2008-08-01 2009-03-11 深圳市天威视讯股份有限公司 Hidden customer characteristic extracting method and television program recommendation method and system
CN102982107A (en) * 2012-11-08 2013-03-20 北京航空航天大学 Recommendation system optimization method with information of user and item and context attribute integrated
CN103617540A (en) * 2013-10-17 2014-03-05 浙江大学 E-commerce recommendation method of tracking user interest changes
CN107766439A (en) * 2017-09-21 2018-03-06 汉鼎宇佑互联网股份有限公司 A kind of personalized recommendation method of fusion structure feature and implicit feedback
CN108460619A (en) * 2018-01-16 2018-08-28 杭州电子科技大学 A kind of fusion shows the Collaborative Recommendation model of implicit feedback

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140025490A1 (en) * 2012-07-17 2014-01-23 Bharathi Shekar Automated recommendations based on historic location-preference information

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101383942A (en) * 2008-08-01 2009-03-11 深圳市天威视讯股份有限公司 Hidden customer characteristic extracting method and television program recommendation method and system
CN102982107A (en) * 2012-11-08 2013-03-20 北京航空航天大学 Recommendation system optimization method with information of user and item and context attribute integrated
CN103617540A (en) * 2013-10-17 2014-03-05 浙江大学 E-commerce recommendation method of tracking user interest changes
CN107766439A (en) * 2017-09-21 2018-03-06 汉鼎宇佑互联网股份有限公司 A kind of personalized recommendation method of fusion structure feature and implicit feedback
CN108460619A (en) * 2018-01-16 2018-08-28 杭州电子科技大学 A kind of fusion shows the Collaborative Recommendation model of implicit feedback

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于大规模隐式反馈的个性化推荐;印鉴等;《软件学报》;20140930;第25卷(第9期);第1953-1966页 *

Also Published As

Publication number Publication date
CN109299370A (en) 2019-02-01

Similar Documents

Publication Publication Date Title
CN109299370B (en) Multi-pair level personalized recommendation method
CN108629665B (en) Personalized commodity recommendation method and system
CN107330115B (en) Information recommendation method and device
Karatzoglou et al. Multiverse recommendation: n-dimensional tensor factorization for context-aware collaborative filtering
Kant et al. Merging user and item based collaborative filtering to alleviate data sparsity
CN107563841B (en) Recommendation system based on user score decomposition
Yang et al. Collaborative competitive filtering: learning recommender using context of user choice
CN111461841B (en) Article recommendation method, device, server and storage medium
CN111310063B (en) Neural network-based article recommendation method for memory perception gated factorization machine
CN104317835A (en) New user recommendation method for video terminal
CN107545471B (en) Big data intelligent recommendation method based on Gaussian mixture
CN105809474B (en) Hierarchical commodity information filtering recommendation method
US11386301B2 (en) Cluster and image-based feedback system
CN113468227A (en) Information recommendation method, system, device and storage medium based on graph neural network
WO2019038736A1 (en) Apparatus and method for training a similarity model used to predict similarity between items
Katarya et al. Effectivecollaborative movie recommender system using asymmetric user similarity and matrix factorization
CN112085525A (en) User network purchasing behavior prediction research method based on hybrid model
CN113918832A (en) Graph convolution collaborative filtering recommendation system based on social relationship
Sun-Hosoya et al. Activmetal: Algorithm recommendation with active meta learning
CN110910207A (en) Method and system for improving commodity recommendation diversity
CN113918834A (en) Graph convolution collaborative filtering recommendation method fusing social relations
CN109934681B (en) Recommendation method for user interested goods
CN110059251B (en) Collaborative filtering recommendation method based on multi-relation implicit feedback confidence
Han et al. Optimizing ranking algorithm in recommender system via deep reinforcement learning
CN111310029B (en) Mixed recommendation method based on user commodity portrait and potential factor feature extraction

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