CN111695024A - Object evaluation value prediction method and system, and recommendation method and system - Google Patents

Object evaluation value prediction method and system, and recommendation method and system Download PDF

Info

Publication number
CN111695024A
CN111695024A CN201910181966.2A CN201910181966A CN111695024A CN 111695024 A CN111695024 A CN 111695024A CN 201910181966 A CN201910181966 A CN 201910181966A CN 111695024 A CN111695024 A CN 111695024A
Authority
CN
China
Prior art keywords
user
model
vector
asdae
evaluation value
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.)
Pending
Application number
CN201910181966.2A
Other languages
Chinese (zh)
Inventor
徐邵稀
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.)
Beijing Jingdong Century Trading Co Ltd
Beijing Jingdong Shangke Information Technology Co Ltd
Original Assignee
Beijing Jingdong Century Trading Co Ltd
Beijing Jingdong Shangke Information Technology Co Ltd
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 Beijing Jingdong Century Trading Co Ltd, Beijing Jingdong Shangke Information Technology Co Ltd filed Critical Beijing Jingdong Century Trading Co Ltd
Priority to CN201910181966.2A priority Critical patent/CN111695024A/en
Publication of CN111695024A publication Critical patent/CN111695024A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a prediction method and a prediction system of an object evaluation value, a recommendation method and a recommendation system, electronic equipment and a storage medium. The prediction method comprises the following steps: establishing a database; inquiring target user parameters, target object parameters and object evaluation values from a database, and constructing user characteristic vectors, object characteristic vectors and object scoring matrixes; training an aSDAE model by taking the user characteristic vector subjected to noise processing, the object characteristic vector and the object scoring matrix as training samples to obtain a collaborative filtering model; the output parameters of the collaborative filtering model comprise a user hidden factor vector and an object hidden factor vector; the product of the two is used to predict the evaluation value of the user for the object. The method and the device can accurately predict the evaluation value of the user on the unknown object, namely accurately predict the preference degree of the user on the unknown object.

Description

Object evaluation value prediction method and system, and recommendation method and system
Technical Field
The present invention relates to the field of object recommendation technologies, and in particular, to a method and a system for predicting an object evaluation value, a recommendation method and a recommendation system, an electronic device, and a storage medium.
Background
With the explosive increase of information quantity brought by the rapid development of networks, personalized recommendation is brought about in order to avoid that users browse a large amount of irrelevant information and products and are buried in the problem of information overload. The personalized recommendation is to recommend information and objects which are interested by the user to the user according to the interest characteristics and purchasing behaviors of the user. Collaborative filtering based recommendation methods are of great interest in both academic and industrial fields due to their good performance. Different from the traditional recommendation based on content filtering and direct content analysis, the method is characterized in that the interest of the user is analyzed through collaborative filtering, similar (interested) users of the specified user are found in the user group, and the evaluation of the similar users on a certain object is integrated to form the preference degree prediction of the specified user on the object by the system. In addition, deep learning is applied to a recommendation system, and is fused with a traditional collaborative filtering algorithm, so that the requirements of users, the characteristics of projects and historical interaction among the users and the projects can be better understood.
However, the robustness and the anti-interference performance of the conventional collaborative filtering model cannot be guaranteed, that is, any slight change to the model parameters may be used as a normal sample value by the deep learning model for learning, and an over-fitting phenomenon is easily generated to influence the system performance, thereby influencing the accuracy of personalized recommendation.
Disclosure of Invention
The invention provides a method and a system for predicting an object evaluation value, a method and a system for recommending, an electronic device and a storage medium, and aims to overcome the defect that the robustness and the anti-interference performance of a model cannot be guaranteed to cause low accuracy of personalized recommendation when a collaborative filtering model in the prior art is adopted for personalized recommendation.
The invention solves the technical problems through the following technical scheme:
a prediction method of an object evaluation value, the prediction method comprising:
establishing a database; the database is used for storing user data and object data;
querying a target user parameter from the user data, querying a target object parameter and an object evaluation value from the object data, and constructing a user characteristic vector, an object characteristic vector and an object evaluation matrix according to the target user parameter, the object evaluation value and the target object parameter;
carrying out noise processing on the user characteristic vector, the object characteristic vector and the object scoring matrix;
training an aSDAE model by taking the user characteristic vector subjected to noise processing, the object characteristic vector and the object scoring matrix as training samples to obtain a collaborative filtering model;
the output parameters of the collaborative filtering model comprise a user hidden factor vector and an object hidden factor vector;
the product of the user hidden factor vector and the object hidden factor vector is used for predicting the evaluation value of the user on the object.
Preferably, in the process of training the aSDAE model, model parameters satisfy gaussian distribution; and/or the output result of each layer of the aSDAE model satisfies a Gaussian distribution or a Dirac delta distribution.
Preferably, an objective function for training the aSDAE model is constructed based on bayesian maximum likelihood theory.
Preferably, the aSDAE model includes a user aSDAE model and an object aSDAE model;
the user implicit factor vector is the sum of an output result of an intermediate layer of the user aSDAE model and a first error;
the object implicit factor vector is the sum of an output result of an intermediate layer of the object aSDAE model and a second error;
the first error and the second error both obey a gaussian distribution.
Preferably, before the step of performing the noise processing on the user feature vector, the object feature vector and the object score matrix, the method further includes:
and preprocessing the user characteristic vector, the object characteristic vector and the object scoring matrix to enable the user characteristic vector, the object characteristic vector and the object scoring matrix to meet Gaussian distribution.
An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the object evaluation value prediction method of any one of the above when executing the computer program.
A computer-readable storage medium on which a computer program is stored, the computer program, when executed by a processor, implementing the steps of the object evaluation value prediction method of any one of the above.
A recommendation method, the recommendation method comprising:
predicting a user characteristic vector, an object characteristic vector and a user implicit factor vector and an object implicit factor vector of an object scoring matrix which are subjected to noise processing by using the prediction method of the object evaluation value;
calculating the product of the user hidden factor vector and the object hidden factor vector, and sequencing the objects according to the sequence of the products from high to low;
recommending a plurality of objects ranked at the top to the user.
An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the above-mentioned recommendation method when executing the computer program.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the recommendation method described above.
A prediction system of an object evaluation value, the prediction system comprising:
a database for storing user data and object data;
the data acquisition module is used for inquiring target user parameters from the user data, inquiring target object parameters and object evaluation values from the object data, and constructing user characteristic vectors, object characteristic vectors and object evaluation matrixes according to the target user parameters, the object evaluation values and the target object parameters;
the noise adding module is used for adding noise to the user characteristic vector, the object characteristic vector and the object scoring matrix;
the model training module is used for training an aSDAE model by taking the user characteristic vector subjected to noise processing, the object characteristic vector and the object scoring matrix as training samples to obtain a collaborative filtering model;
the output parameters of the collaborative filtering model comprise a user hidden factor vector and an object hidden factor vector;
the product of the user hidden factor vector and the object hidden factor vector is used for predicting the evaluation value of the user on the object.
Preferably, in the process of training the aSDAE model, model parameters satisfy gaussian distribution; and/or the output result of each layer of the aSDAE model satisfies a Gaussian distribution or a Dirac delta distribution.
Preferably, the prediction system further comprises:
and the function construction module is used for constructing and training a target function of the aSDAE model based on the Bayesian maximum likelihood theory.
Preferably, the aSDAE model includes a user aSDAE model and an object aSDAE model;
the user implicit factor vector is the sum of an output result of an intermediate layer of the user aSDAE model and a first error;
the object implicit factor vector is the sum of an output result of an intermediate layer of the object aSDAE model and a second error;
the first error and the second error both obey a gaussian distribution.
Preferably, the prediction system further comprises:
and the data processing module is used for preprocessing the user characteristic vector, the object characteristic vector and the object scoring matrix to enable the user characteristic vector, the object characteristic vector and the object scoring matrix to meet Gaussian distribution.
A recommendation system, the recommendation system comprising: a calculation module, a ranking module, a recommendation module, and a prediction system using the object assessment value as described in any of the above;
the calculation module is used for calling the prediction system to predict the user hidden factor vector and the object hidden factor vector of the user characteristic vector, the object characteristic vector and the object scoring matrix which are subjected to noise processing, and calculating the product of the user hidden factor vector and the object hidden factor vector;
the sorting module is used for sorting the objects according to the sequence of the products from high to low;
the recommending module is used for recommending a plurality of objects ranked in the front to the user.
The positive progress effects of the invention are as follows: the method and the device can accurately predict the evaluation value of the user on the unknown object, namely accurately predict the preference degree of the user on the unknown object, and provide reference for personalized object recommendation. And during model training, Gaussian modeling is carried out on the noise of various weight parameters of the aSDAE model, so that the phenomenon of overfitting is effectively avoided to a certain extent, and the robustness and the anti-interference capability of the model are greatly enhanced.
Drawings
Fig. 1 is a flowchart of a method of predicting an object evaluation value according to embodiment 1 of the present invention.
Fig. 2 is a model configuration diagram used in the method for predicting an object evaluation value according to embodiment 1 of the present invention.
Fig. 3 is a schematic structural diagram of an electronic device according to embodiment 2 of the present invention.
Fig. 4 is a flowchart of a recommendation method according to embodiment 4 of the present invention.
Fig. 5 is a block diagram of a system for predicting an object evaluation value according to embodiment 7 of the present invention.
Fig. 6 is a schematic block diagram of a recommendation system according to embodiment 8 of the present invention.
Detailed Description
The invention is further illustrated by the following examples, which are not intended to limit the scope of the invention.
Example 1
The present embodiment provides a method for predicting an object evaluation value, which is used for predicting a defect value of a scoring matrix of an object (for example, a commodity, an article, a coupon, and the like), that is, calculating a predicted evaluation value of an unknown object by a user, so as to accurately predict the user's preference degree of the unknown object.
As shown in fig. 1, the object evaluation value prediction method of the present embodiment includes the steps of:
step 100, establishing a database.
The database is used for storing user data and object data.
Step 101, querying a target user parameter from the user data, querying a target object parameter and an object evaluation value from the object data, and constructing a user characteristic vector, an object characteristic vector and an object evaluation matrix according to the target user parameter, the object evaluation value and the target object parameter.
Wherein, the user feature vector comprises the following parameters: user name, age, gender, address, total purchase times, average browsing time, average amount of money consumed, and the like; see table 1 for an example of 5 users (U1, U2, U3, U4, and U5).
TABLE 1
Figure BDA0001991588360000061
The object feature vector includes the following parameters: object name, category, price, origin, delivery location, total number of purchases and total number of browsed items, etc.; see table 2, for example, with 20 subjects (I1, I2, … …, I19, and I20).
TABLE 2
Figure BDA0001991588360000062
Figure BDA0001991588360000071
The object evaluation values represent the scores of the objects by different users. The object scoring matrix is constructed from the object evaluation values and is a sparse matrix. Assuming a rating of 1-5, a higher score indicates a higher user preference for the object, see table 3, where the missing value indicates that the user preference for the object is unknown, and is calculated as 0. Generally, the rating value in the object rating matrix R can be obtained by comprehensive calculation through implicit feedback (such as the stay time of the user on a certain object page, the browsing times, etc.) or display feedback (such as the rating score of the user on a certain object, etc.) of the user.
TABLE 3
Figure BDA0001991588360000072
And 102, carrying out noise adding treatment on the user characteristic vector, the object characteristic vector and the object scoring matrix.
For example, white noise, gaussian noise, etc. are added to the user eigenvector, the object eigenvector, and the object scoring matrix to improve the robustness of the model.
The method for processing the noise of the object scoring matrix specifically comprises the following steps:
step 102-1, converting the object scoring matrix into a first evaluation value matrix and a second evaluation value matrix.
The first evaluation value matrix represents evaluation values of each user i to all n kinds of objects; the second evaluation value matrix represents evaluation values of all m users on the object j, specifically:
will evaluate the matrix
Figure BDA0001991588360000073
Converted into a first evaluation value matrix
Figure BDA0001991588360000074
Figure BDA0001991588360000075
For each user i ∈ {1, …, m },
Figure BDA0001991588360000076
for the evaluation value vectors of all the n kinds of objects for the user i, that is, the row vectors of the evaluation matrix R, the scoring matrix R shown in table 3 is taken as an example,
Figure BDA0001991588360000077
Figure BDA0001991588360000078
converting the evaluation matrix R into a second evaluation value matrix
Figure BDA0001991588360000079
Figure BDA0001991588360000081
For each object j ∈ {1, …, n },
Figure BDA0001991588360000082
the evaluation value vectors for all m users for the object j, i.e., the column vectors of the evaluation matrix R, take the scoring matrix R shown in table 3 as an example,
Figure BDA0001991588360000083
Figure BDA0001991588360000084
step 102-2, pair
Figure BDA0001991588360000085
And
Figure BDA0001991588360000086
respectively carrying out noise adding processing to obtain
Figure BDA0001991588360000087
And
Figure BDA0001991588360000088
Figure BDA0001991588360000089
is a pair of
Figure BDA00019915883600000810
The evaluation value vector after the noise processing is added,
Figure BDA00019915883600000811
is a pair of
Figure BDA00019915883600000812
And adding the evaluation value vector after the noise processing.
The user feature vector and the object feature vector are used separately
Figure BDA00019915883600000813
And
Figure BDA00019915883600000814
to represent, the corresponding noisy representation is
Figure BDA00019915883600000815
And
Figure BDA00019915883600000816
Figure BDA00019915883600000817
in this embodiment, before step 102-2, the steps can be performed
Figure BDA00019915883600000818
And xi，yjPerforming a pre-treatment to make it noiseless
Figure BDA00019915883600000819
And xi，yjObey the following gaussian distribution:
Figure BDA00019915883600000820
Figure BDA00019915883600000821
Figure BDA00019915883600000822
Figure BDA00019915883600000823
so that
Figure BDA00019915883600000824
And xi，yjAnd is close to each layer of output result of the model in the training process in an infinite way.
And 103, inputting the user characteristic vector subjected to noise processing, the object characteristic vector and the object scoring matrix as training samples into an aSDAE model, and performing model training to obtain a collaborative filtering model.
The output parameters of the collaborative filtering model comprise an object hidden factor vector and a user hidden factor vector. The product of the user hidden factor vector and the object hidden factor vector represents the prediction evaluation value of the user on the object. The aSDAE model comprises: a user aSDAE model, an object aSDAE model and a matrix decomposition model.
In step 103, training the learning object hidden factor vector and the user hidden factor vector based on a stochastic gradient algorithm (SGD), specifically:
step 103-1, mixing
Figure BDA0001991588360000091
And
Figure BDA0001991588360000092
input the user aSDAE model
Figure BDA0001991588360000093
And
Figure BDA0001991588360000094
inputting an object aSDAE model, and training a learning object hidden factor vector and a user hidden factor vector.
The output parameters of the user aSDAE model are user hidden factor vectors, and the output parameters of the object aSDAE model are object hidden factor vectors.
Referring to fig. 2, in each iteration, the update rule is trained as follows:
Figure BDA0001991588360000095
Figure BDA0001991588360000096
wherein, U and V are two low-rank (low-rank) matrixes obtained based on a matrix decomposition model, and R is approximately equal to UVT；
Figure BDA0001991588360000097
Representing the objective function of each iteration, η the learning rate of the stochastic gradient descent algorithm, ui' and vj' denotes the result after the current iteration as the basis for the next iteration.
In fig. 2, L is the number of layers of the user aSDAE model and the object aSDAE model, L ∈ {1, …, L };
Figure BDA0001991588360000098
for the user aSDAE model
Figure BDA0001991588360000099
Layer I results as input;
Figure BDA00019915883600000910
for the user aSDAE model
Figure BDA00019915883600000911
As an input
Figure BDA00019915883600000922
Layer (intermediate layer) results.
Figure BDA00019915883600000912
As a subject aSDAE model to
Figure BDA00019915883600000913
Layer I results as input;
Figure BDA00019915883600000914
as a subject aSDAE model to
Figure BDA00019915883600000915
As an input
Figure BDA00019915883600000923
Layer (intermediate layer) results.
Figure BDA00019915883600000916
For the user aSDAE model
Figure BDA00019915883600000917
Layer I results as input;
Figure BDA00019915883600000918
as a subject aSDAE model to
Figure BDA00019915883600000919
As input layer i results. WlWeight parameter of user aSDAE model at l layer, blFor corresponding weight parameter, TlIs composed of
Figure BDA00019915883600000920
Adding weight parameters of the l layer of the user aSDAE model; w'lIs a weight parameter, b 'of the object aSDAE model in the l'lIs a corresponding offset vector, T'lIs composed of
Figure BDA00019915883600000921
And adding weight parameters of the l layer of the object aSDAE model.
In this embodiment, when model training is performed, the model parameter Wl、Tl、bl、W′l、T′lAnd b'lAnd the output result of each layer of the model satisfies the following distribution:
(a) user weight matrix
Figure BDA0001991588360000101
Each column n ∈ {1, … Kl-1Is Wl,*n，KlNumber of nodes on layer I of table user aSDAE model, K0＝KLN; set object weight matrix
Figure BDA0001991588360000102
Figure BDA0001991588360000103
Each column of n '∈ {1, … K'l-1Is W'l，*n′，K′lRepresenting the number of nodes, K ', of the l-layer of the user aSDAE'0＝K′L＝m；
Let Wl,*nAnd W'l，*n′The values of (c) are subject to the following distribution:
Figure BDA0001991588360000104
Figure BDA0001991588360000105
wherein,
Figure BDA0001991588360000106
is Kl*KlA unit matrix of size;
Figure BDA0001991588360000107
is K'l*K′lA unit matrix of size;
(b) weight matrix of user feature vectors
Figure BDA0001991588360000108
Each column n ∈ {1, … p } of is Tl,*nSetting a weight matrix of the object feature vector
Figure BDA0001991588360000109
Each column n '∈ {1, … q } of (a) is T'l，*n′；
Let Tl,*nAnd T'l，*n′The values of (c) are subject to the following distribution:
Figure BDA00019915883600001010
Figure BDA00019915883600001011
(c) output results of each layer of user aSDAE model and object aSDAE model
Figure BDA00019915883600001012
And
Figure BDA00019915883600001013
the following distribution is obeyed:
Figure BDA00019915883600001014
Figure BDA00019915883600001015
wherein,
Figure BDA00019915883600001016
when lambda issWhen the temperature approaches to + ∞, the temperature of the reaction kettle is increased,
Figure BDA00019915883600001017
obeying a Gaussian distribution to
Figure BDA00019915883600001018
Figure BDA00019915883600001019
A centered Dirac delta distribution;
Figure BDA00019915883600001020
obeying a Gaussian distribution to
Figure BDA00019915883600001021
Figure BDA00019915883600001022
A centered Dirac delta distribution;
for u is pairediAnd vjDerivation
Figure BDA00019915883600001023
Can be further refined as:
Figure BDA00019915883600001024
Figure BDA0001991588360000111
Figure BDA0001991588360000112
Cijas confidence parameters, i.e.:
Figure BDA0001991588360000113
(d) let the corresponding offset vector blAnd b'lThe following distribution is obeyed:
Figure BDA0001991588360000114
Figure BDA0001991588360000115
blis a constant parameter, b 'of each layer of a user aSDAE model'lConstant parameters of each layer of the object aSDAE model;
wherein λ isw,λt,λs,λn,λu,λvIs a hyper-parameter of the model.
And 103-2, judging whether the current aSDAE model meets the constraint condition of the optimization target.
In the step 103-2, if the judgment is yes, the step 103-3 is executed; if not, returning to the step 103-1, reselecting the training sample, and debugging the model parameters until the objective function is obtained
Figure BDA0001991588360000117
Is the minimum value.
In this embodiment, according to the bayesian maximum likelihood theory, the optimal objective function of the model training is obtained as follows:
Figure BDA0001991588360000116
Figure BDA0001991588360000121
when lambda issWhen approaching + ∞, minimize the objective function
Figure BDA00019915883600001211
The following steps are changed:
Figure BDA0001991588360000122
and 103-3, determining the current aSDAE model as a final collaborative filtering model.
Adding a first error to an intermediate layer of a user aSDAE model in a current aSDAE modeliI.e. by
Figure BDA0001991588360000123
And adding a second error to the intermediate layer of the object aSDAE modeljI.e. by
Figure BDA0001991588360000124
Figure BDA0001991588360000125
As output parameters of the model, i.e. object implicit factor vectors vjAnd a user implicit factor vector uiAt this time vjAnd uiProduct of (2)
Figure BDA0001991588360000126
The following distribution is obeyed:
Figure BDA0001991588360000127
Cijas confidence parameters, i.e.:
Figure BDA0001991588360000128
wherein the first erroriAnd a second errorjThe following distribution is obeyed:
Figure BDA0001991588360000129
Figure BDA00019915883600001210
where k is the dimension of the implicit factor vector.
Therefore, the collaborative filtering model can predict the evaluation value of the current user on the unknown object, namely accurately predict the preference degree of the user on the unknown object, and provide reference for personalized object recommendation. Specifically, when the collaborative filtering model is used for prediction, if learning is performed, a user implicit factor matrix is obtained
Figure BDA0001991588360000131
Figure BDA0001991588360000132
And object implicit factor matrix sum
Figure BDA0001991588360000133
The prediction evaluation matrix can be approximately obtained
Figure BDA0001991588360000134
Namely, it is
Figure BDA0001991588360000135
And then, for each user, a list of evaluation sequencing of each object by the user can be obtained based on the product of the user implicit factor matrix and the object implicit factor matrix. And during model training, Gaussian modeling is carried out on the noise of various weight parameters of the aSDAE model, so that the phenomenon of overfitting is effectively avoided to a certain extent, and the robustness and the anti-interference capability of the model are greatly enhanced.
Example 2
Fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, which shows a block diagram of an exemplary electronic device 90 suitable for implementing an embodiment of the present invention. The electronic device 90 shown in fig. 3 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiment of the present invention.
As shown in fig. 3, the electronic device 90 may take the form of a general purpose computing device, which may be a server device, for example. The components of the electronic device 90 may include, but are not limited to: the at least one processor 91, the at least one memory 92, and a bus 93 that connects the various system components (including the memory 92 and the processor 91).
The bus 93 includes a data bus, an address bus, and a control bus.
Memory 92 may include volatile memory, such as Random Access Memory (RAM)921 and/or cache memory 922, and may further include Read Only Memory (ROM) 923.
Memory 92 may also include a program tool 925 (or utility) having a set (at least one) of program modules 924, such program modules 924 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.
The processor 91 executes various functional applications and data processing, such as a prediction method of an object evaluation value provided in embodiment 1 of the present invention, by executing the computer program stored in the memory 92.
The electronic device 90 may also communicate with one or more external devices 94 (e.g., keyboard, pointing device, etc.). Such communication may be through an input/output (I/O) interface 95. Also, the model-generated electronic device 90 may also communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet) via a network adapter 96. As shown, the network adapter 96 communicates with the other modules of the model-generated electronic device 90 via a bus 93. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the model-generating electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.
It should be noted that although in the above detailed description several units/modules or sub-units/modules of the electronic device are mentioned, such a division is merely exemplary and not mandatory. Indeed, the features and functionality of two or more of the units/modules described above may be embodied in one unit/module according to embodiments of the invention. Conversely, the features and functions of one unit/module described above may be further divided into embodiments by a plurality of units/modules.
Example 3
The present embodiment provides a computer-readable storage medium on which a computer program is stored, which when executed by a processor, implements the steps of the object evaluation value prediction method provided in embodiment 1.
More specific examples, among others, that the readable storage medium may employ may include, but are not limited to: a portable disk, a hard disk, random access memory, read only memory, erasable programmable read only memory, optical storage device, magnetic storage device, or any suitable combination of the foregoing.
In a possible embodiment, the present invention may also be embodied in the form of a program product including program code for causing a terminal device to execute the steps in the prediction method of an object evaluation value described in embodiment 1 when the program product is run on the terminal device.
Where program code for carrying out the invention is written in any combination of one or more programming languages, the program code may be executed entirely on the user device, partly on the user device, as a stand-alone software package, partly on the user device and partly on a remote device or entirely on the remote device.
Example 4
As shown in fig. 4, the recommendation method of the present embodiment includes the following steps:
and step 201, predicting the user hidden factor vector and the object hidden factor vector of the user characteristic vector, the object characteristic vector and the object scoring matrix which are subjected to noise processing.
Specifically, in step 201, the user hidden factor vector and the object hidden factor vector are predicted using the object evaluation values in embodiment 1.
Step 202, calculating the product of the user implicit factor vector and the object implicit factor vector, and sequencing the objects according to the sequence of the products from high to low.
The product of the user implicit factor vector and the object implicit factor vector is also the predicted evaluation value of the user on the object.
And step 203, recommending a plurality of objects ranked in the front to the user.
Thus, recommending appropriate objects to users of a specific group is realized to improve the click conversion rate of the objects, such as: the coupon of the maternal and infant products is recommended to the pregnant women, the coupon of the digital products is recommended to the digital darts, and the coupon of cosmetics or clothes is recommended to fashion girls and the like.
Taking the data shown in tables 1 to 3 in embodiment 1 as an example, the data is input into the collaborative filtering model as an input parameter and is obtained through calculation
Figure BDA0001991588360000151
As shown in table 4:
TABLE 4
Figure BDA0001991588360000152
Referring to table 4, the evaluation values missing from the original evaluation matrix R are all estimated. Thus, according to
Figure BDA0001991588360000153
A recommendation list corresponding to each user can be obtained, see table 5:
TABLE 5
Figure BDA0001991588360000154
Figure BDA0001991588360000161
Wherein the recommendation list is based on the evaluation matrix
Figure BDA0001991588360000162
The evaluation values of the 20 kinds of commodities by the user are sorted from large to small, that is, the commodity with the high evaluation value is arranged at the top of the list, that is, the corresponding object can be recommended to the user according to the size of the evaluation value.
Example 5
The embodiment provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the recommendation method in embodiment 4 when executing the computer program.
Example 6
The present embodiment provides a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the recommendation method in embodiment 4.
Example 7
The present embodiment provides a system for predicting an object evaluation value, which is used for calculating a defect value existing in a scoring matrix of an object (for example, a commodity, an article, a coupon, and the like), that is, calculating a predicted evaluation value of an unknown object by a user, so as to accurately predict a user's preference degree for the unknown object, as shown in fig. 5, the establishing system includes: a database 10, a data acquisition module 11, a noise adding module 12, a model training module 13, a function building module 14 and a data processing module 15.
The database 10 is used to store user data and object data.
The data acquisition module 11 is configured to query a target user parameter from the user data, query a target object parameter and an object evaluation value from the object data, and construct a user feature vector, an object feature vector, and an object evaluation matrix according to the target user parameter, the object evaluation value, and the target object parameter.
Wherein, the user feature vector comprises the following parameters: user name, age, gender, address, total number of purchases, average browsing time, average amount of money spent, etc. The object feature vector includes the following parameters: object name, category, price, origin, shipping location, total number of purchases and total number of browses, etc. The object evaluation values represent the scores of the objects by different users. The object scoring matrix is constructed from the object evaluation values and is a sparse matrix. Assuming that the grade is 1-5, the larger the score is, the higher the user's preference for the object is, and the missing value in the sparse matrix indicates that the preference of a certain user for a certain object is unknown, and is represented by 0 in calculation. Generally, the rating value in the object rating matrix R can be obtained by comprehensive calculation through implicit feedback (such as the stay time of the user on a certain object page, the browsing times, etc.) or display feedback (such as the rating score of the user on a certain object, etc.) of the user.
The noise adding module 12 is used for adding noise to the user feature vector, the object feature vector and the object scoring matrix.
For example, white noise, gaussian noise, etc. are added to the user eigenvector, the object eigenvector, and the object scoring matrix to improve the robustness of the model.
When the object scoring matrix is subjected to the noise adding treatment:
the noise addition module 12 converts the object score matrix into a first evaluation value matrix and a second evaluation value matrix.
The first evaluation value matrix represents evaluation values of each user i to all n kinds of objects; the second evaluation value matrix represents evaluation values of all m users on the object j, specifically:
will evaluate the matrix
Figure BDA0001991588360000171
Converted into a first evaluation value matrix
Figure BDA0001991588360000172
Figure BDA0001991588360000173
For each user i ∈ {1, …, m },
Figure BDA0001991588360000174
evaluating values for all n kinds of objects for the user i, namely row vectors of an evaluation matrix R;
converting the evaluation matrix R into a second evaluation value matrix
Figure BDA0001991588360000175
Figure BDA0001991588360000176
For each object j ∈ {1, …, n },
Figure BDA0001991588360000177
the evaluation values for the object j for all m users, i.e. the column vectors of the evaluation matrix R.
The noise adding modules 12 are respectively paired
Figure BDA0001991588360000178
And
Figure BDA0001991588360000179
respectively carrying out noise adding processing to obtain
Figure BDA00019915883600001710
And
Figure BDA00019915883600001711
Figure BDA0001991588360000181
is a pair of
Figure BDA0001991588360000182
The evaluation value vector after the noise processing is added,
Figure BDA0001991588360000183
is a pair of
Figure BDA0001991588360000184
And adding the evaluation value vector after the noise processing.
The user feature vector and the object feature vector are used separately
Figure BDA0001991588360000185
And
Figure BDA0001991588360000186
to represent, the corresponding noisy representation is
Figure BDA0001991588360000187
And
Figure BDA0001991588360000188
Figure BDA0001991588360000189
in this embodiment, before performing the noise processing on the user feature vector, the object feature vector, and the object scoring matrix, the data processing module 15 may also be used to perform the noise processing on the user feature vector, the object feature vector, and the object scoring matrix
Figure BDA00019915883600001810
And xi，yjPerforming a pretreatment toWithout noise
Figure BDA00019915883600001811
And xi，yjObey the following gaussian distribution:
Figure BDA00019915883600001812
Figure BDA00019915883600001813
Figure BDA00019915883600001814
Figure BDA00019915883600001815
so that
Figure BDA00019915883600001816
And xi，yjAnd is close to each layer of output result of the model in the training process in an infinite way.
The data processing module 15 inputs the preprocessed user feature vectors, object feature vectors and object scoring matrices to the noise module 12.
The model training module 13 is configured to input the user feature vector, the object feature vector, and the object scoring matrix, which are subjected to the noise processing, as training samples into the aSDAE model, and train the training samples to obtain the collaborative filtering model.
The output parameters of the collaborative filtering model comprise a user hidden factor vector and an object hidden factor vector. The product of the user hidden factor vector and the object hidden factor vector represents the prediction evaluation value of the user on the object. The aSDAE model comprises: a user aSDAE model, an object aSDAE model and a matrix decomposition model.
In this embodiment, the learning object hidden factor vector and the user hidden factor vector are trained based on a stochastic gradient algorithm (SGD), specifically:
will be provided with
Figure BDA00019915883600001817
And
Figure BDA00019915883600001818
input the user aSDAE model
Figure BDA00019915883600001819
And
Figure BDA00019915883600001820
inputting an object aSDAE model, and training a learning object hidden factor vector and a user hidden factor vector.
The output parameters of the user aSDAE model are user hidden factor vectors, and the output parameters of the object aSDAE model are object hidden factor vectors.
Referring to fig. 2, in each iteration, the update rule is trained as follows:
Figure BDA0001991588360000191
Figure BDA0001991588360000192
wherein, U and V are two low-rank (low-rank) matrixes obtained based on a matrix decomposition model, and R is approximately equal to UVT；
Figure BDA0001991588360000193
Representing the objective function of each iteration, η the learning rate of the stochastic gradient descent algorithm, ui' and vj' denotes the result after the current iteration as the basis for the next iteration.
In fig. 2, L is the number of layers of the user aSDAE model and the object aSDAE model, L ∈ {1, …, L };
Figure BDA0001991588360000194
for the user aSDAE model
Figure BDA0001991588360000195
Layer I results as input;
Figure BDA0001991588360000196
for the user aSDAE model
Figure BDA0001991588360000197
As an input
Figure BDA00019915883600001921
Layer (intermediate layer) results.
Figure BDA0001991588360000198
As a subject aSDAE model to
Figure BDA0001991588360000199
Layer I results as input;
Figure BDA00019915883600001910
as a subject aSDAE model to
Figure BDA00019915883600001911
As an input
Figure BDA00019915883600001922
Layer (intermediate layer) results.
Figure BDA00019915883600001912
For the user aSDAE model
Figure BDA00019915883600001913
Layer I results as input;
Figure BDA00019915883600001914
as a subject aSDAE model to
Figure BDA00019915883600001915
As input layer i results. WlTo useWeight parameter of user aSDAE model on l layer, blFor corresponding weight parameter, TlIs composed of
Figure BDA00019915883600001916
Adding weight parameters of the l layer of the user aSDAE model; w'lIs a weight parameter, b 'of the object aSDAE model in the l'lIs a corresponding offset vector, T'lIs composed of
Figure BDA00019915883600001917
And adding weight parameters of the l layer of the object aSDAE model.
In this embodiment, when model training is performed, the model parameter Wl、Tl、bl、W′l、T′lAnd b'lAnd the output result of each layer of the model satisfies the following distribution:
(a) user weight matrix
Figure BDA00019915883600001918
Each column n ∈ {1, … Kl-1Is Wl,*n，KlNumber of nodes on layer I of table user aSDAE model, K0＝KLN; set object weight matrix
Figure BDA00019915883600001919
Figure BDA00019915883600001920
N' ∈ {1, … Kl′-1Is Wl′,*n′，K′lRepresenting the number of nodes, K ', of the l-layer of the user aSDAE'0＝K′L＝m；
Let Wl,*nAnd W'l,*n′The values of (c) are subject to the following distribution:
Figure BDA0001991588360000201
Figure BDA0001991588360000202
wherein,
Figure BDA0001991588360000203
is Kl*KlA unit matrix of size;
Figure BDA0001991588360000204
is K'l*K′lA unit matrix of size;
(b) weight matrix of user feature vectors
Figure BDA0001991588360000205
Each column n ∈ {1, … p } of is Tl,*nSetting a weight matrix of the object feature vector
Figure BDA0001991588360000206
Each column n '∈ {1, … q } of (a) is T'l,*n′；
Let Tl,*nAnd T'l,*n′The values of (c) are subject to the following distribution:
Figure BDA0001991588360000207
Figure BDA0001991588360000208
(c) output results of each layer of user aSDAE model and object aSDAE model
Figure BDA0001991588360000209
And
Figure BDA00019915883600002010
the following distribution is obeyed:
Figure BDA00019915883600002011
Figure BDA00019915883600002012
wherein,
Figure BDA00019915883600002013
when lambda issWhen the temperature approaches to + ∞, the temperature of the reaction kettle is increased,
Figure BDA00019915883600002014
obeying a Gaussian distribution to
Figure BDA00019915883600002015
Figure BDA00019915883600002016
A centered Dirac delta distribution;
Figure BDA00019915883600002017
obeying a Gaussian distribution to
Figure BDA00019915883600002018
Figure BDA00019915883600002019
A centered Dirac delta distribution;
for u is pairediAnd vjDerivation
Figure BDA00019915883600002020
Can be further refined as:
Figure BDA00019915883600002021
Figure BDA00019915883600002022
Figure BDA0001991588360000211
Cijas confidence parameters, i.e.:
Figure BDA0001991588360000212
(d) let the corresponding offset vector blAnd b'lThe following distribution is obeyed:
Figure BDA0001991588360000213
Figure BDA0001991588360000214
blis a constant parameter, b 'of each layer of a user aSDAE model'lConstant parameters of each layer of the object aSDAE model;
wherein λ isw,λt,λs,λn,λu,λvIs a hyper-parameter of the model.
And after each iteration, judging whether the current aSDAE model meets the constraint condition of the optimization target.
If so, determining the current aSDAE model as a final collaborative filtering model; if not, reselecting the training sample, and debugging the model parameters until the target function is reached
Figure BDA0001991588360000217
Is the minimum value.
In this embodiment, the function building module 14 builds a target function of the training aSDAE model based on the bayesian maximum likelihood theory, which is specifically as follows:
Figure BDA0001991588360000215
when lambda issWhen approaching + ∞, minimize the objective function
Figure BDA0001991588360000216
The following steps are changed:
Figure BDA0001991588360000221
when the objective function
Figure BDA00019915883600002212
When the difference is the minimum value, determining the current aSDAE model as a final collaborative filtering model, and adding a first error to an intermediate layer of a user aSDAE model in the current aSDAE modeliI.e. by
Figure BDA0001991588360000222
And adding a second error to the intermediate layer of the object aSDAE modeljI.e. by
Figure BDA0001991588360000223
As output parameters of the model, i.e. object implicit factor vectors vjAnd a user implicit factor vector uiAt this time vjAnd uiProduct of (A) RijThe following distribution is obeyed:
Figure BDA0001991588360000224
Cijas confidence parameters, i.e.:
Figure BDA0001991588360000225
wherein the first erroriAnd a second errorjThe following distribution is obeyed:
Figure BDA0001991588360000226
Figure BDA0001991588360000227
where k is the dimension of the implicit factor vector.
The collaborative filtering model can accurately predict the evaluation value of the current user to the unknown object, namely the accurate predictionAnd measuring the preference degree of the user to the unknown object, and providing reference for personalized object recommendation. Specifically, when the collaborative filtering model is used for prediction, if learning is performed, a user implicit factor matrix is obtained
Figure BDA0001991588360000228
And object implicit factor matrix sum
Figure BDA0001991588360000229
The prediction evaluation matrix can be approximately obtained
Figure BDA00019915883600002210
Namely, it is
Figure BDA00019915883600002211
Therefore, for each user, a list of the evaluation ranking of each object by the user can be obtained based on the product of the user implicit factor matrix and the object implicit factor matrix.
Example 8
The present embodiment provides a recommendation system, including: a calculation module 21, a ranking module 22, a recommendation module 23, and a prediction system 24 using the object evaluation value in embodiment 7.
The calculation module 21 is configured to invoke a prediction system to predict the user hidden factor vector and the object hidden factor vector of the user feature vector, the object feature vector and the object score matrix that are subjected to the noise processing, and calculate a product of the user hidden factor vector and the object hidden factor vector.
The product of the user implicit factor vector and the object implicit factor vector is also the predicted evaluation value of the user on the object.
The sorting module 22 is used for sorting the objects according to the order of products from high to low.
The recommending module 23 is used for recommending a plurality of objects ranked at the top to the user.
Thus, recommending appropriate objects to users of a specific group is realized to improve the click conversion rate of the objects, such as: the coupon of the maternal and infant products is recommended to the pregnant women, the coupon of the digital products is recommended to the digital darts, and the coupon of cosmetics or clothes is recommended to fashion girls and the like.
While specific embodiments of the invention have been described above, it will be appreciated by those skilled in the art that this is by way of example only, and that the scope of the invention is defined by the appended claims. Various changes and modifications to these embodiments may be made by those skilled in the art without departing from the spirit and scope of the invention, and these changes and modifications are within the scope of the invention.

Claims (16)

1. A prediction method of an object evaluation value, characterized by comprising:
establishing a database; the database is used for storing user data and object data;
querying a target user parameter from the user data, querying a target object parameter and an object evaluation value from the object data, and constructing a user characteristic vector, an object characteristic vector and an object evaluation matrix according to the target user parameter, the object evaluation value and the target object parameter;
carrying out noise processing on the user characteristic vector, the object characteristic vector and the object scoring matrix;
training an aSDAE model by taking the user characteristic vector subjected to noise processing, the object characteristic vector and the object scoring matrix as training samples to obtain a collaborative filtering model;
the output parameters of the collaborative filtering model comprise a user hidden factor vector and an object hidden factor vector;
the product of the user hidden factor vector and the object hidden factor vector is used for predicting the evaluation value of the user on the object.
2. The method for predicting an object evaluation value according to claim 1, wherein in training the aSDAE model, model parameters satisfy gaussian distribution; and/or the output result of each layer of the aSDAE model satisfies a Gaussian distribution or a Dirac delta distribution.
3. The method for predicting an object assessment value according to claim 1, wherein an objective function for training the aSDAE model is constructed based on a bayesian maximum likelihood theory.
4. The method for predicting an object evaluation value according to claim 1, wherein the aSDAE model includes a user aSDAE model and an object aSDAE model;
the user implicit factor vector is the sum of an output result of an intermediate layer of the user aSDAE model and a first error;
the object implicit factor vector is the sum of an output result of an intermediate layer of the object aSDAE model and a second error;
the first error and the second error both obey a gaussian distribution.
5. The method of predicting an object evaluation value according to any one of claims 1 to 4, wherein the step of subjecting the user feature vector, the object feature vector, and the object score matrix to a noise processing is preceded by:
and preprocessing the user characteristic vector, the object characteristic vector and the object scoring matrix to enable the user characteristic vector, the object characteristic vector and the object scoring matrix to meet Gaussian distribution.
6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of predicting an object evaluation value according to any one of claims 1 to 5 when executing the computer program.
7. A computer-readable storage medium on which a computer program is stored, the computer program, when being executed by a processor, implementing the steps of the prediction method of an object evaluation value according to any one of claims 1 to 5.
8. A recommendation method, characterized in that the recommendation method comprises:
predicting a user hidden factor vector and an object hidden factor vector of the noisy user feature vector, the object feature vector, and the object score matrix using the prediction method of the object evaluation value according to any one of claims 1 to 5;
calculating the product of the user hidden factor vector and the object hidden factor vector, and sequencing the objects according to the sequence of the products from high to low;
recommending a plurality of objects ranked at the top to the user.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the recommendation method as claimed in claim 8 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the recommendation method as claimed in claim 8.
11. A prediction system of an object evaluation value, characterized by comprising:
a database for storing user data and object data;
the data acquisition module is used for inquiring target user parameters from the user data, inquiring target object parameters and object evaluation values from the object data, and constructing user characteristic vectors, object characteristic vectors and object evaluation matrixes according to the target user parameters, the object evaluation values and the target object parameters;
the noise adding module is used for adding noise to the user characteristic vector, the object characteristic vector and the object scoring matrix;
the model training module is used for training an aSDAE model by taking the user characteristic vector subjected to noise processing, the object characteristic vector and the object scoring matrix as training samples to obtain a collaborative filtering model;
the output parameters of the collaborative filtering model comprise a user hidden factor vector and an object hidden factor vector;
the product of the user hidden factor vector and the object hidden factor vector is used for predicting the evaluation value of the user on the object.
12. The system for predicting an object evaluation value according to claim 11, wherein in training the aSDAE model, model parameters satisfy gaussian distribution; and/or the output result of each layer of the aSDAE model satisfies a Gaussian distribution or a Dirac delta distribution.
13. The prediction system of an object evaluation value according to claim 11, characterized in that the prediction system further comprises:
and the function construction module is used for constructing and training a target function of the aSDAE model based on the Bayesian maximum likelihood theory.
14. The system for predicting an object assessment value as set forth in claim 11, wherein said aSDAE model comprises a user aSDAE model and an object aSDAE model;
the user implicit factor vector is the sum of an output result of an intermediate layer of the user aSDAE model and a first error;
the object implicit factor vector is the sum of an output result of an intermediate layer of the object aSDAE model and a second error;
the first error and the second error both obey a gaussian distribution.
15. The prediction system of an object evaluation value according to any one of claims 11 to 14, characterized by further comprising:
and the data processing module is used for preprocessing the user characteristic vector, the object characteristic vector and the object scoring matrix to enable the user characteristic vector, the object characteristic vector and the object scoring matrix to meet Gaussian distribution.
16. A recommendation system, characterized in that the recommendation system comprises: a calculation module, a ranking module, a recommendation module, and a prediction system using the object evaluation value of any one of claims 11-15;
the calculation module is used for calling the prediction system to predict the user hidden factor vector and the object hidden factor vector of the user characteristic vector, the object characteristic vector and the object scoring matrix which are subjected to noise processing, and calculating the product of the user hidden factor vector and the object hidden factor vector;
the sorting module is used for sorting the objects according to the sequence of the products from high to low;
the recommending module is used for recommending a plurality of objects ranked in the front to the user.
CN201910181966.2A 2019-03-11 2019-03-11 Object evaluation value prediction method and system, and recommendation method and system Pending CN111695024A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910181966.2A CN111695024A (en) 2019-03-11 2019-03-11 Object evaluation value prediction method and system, and recommendation method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910181966.2A CN111695024A (en) 2019-03-11 2019-03-11 Object evaluation value prediction method and system, and recommendation method and system

Publications (1)

Publication Number Publication Date
CN111695024A true CN111695024A (en) 2020-09-22

Family

ID=72474746

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910181966.2A Pending CN111695024A (en) 2019-03-11 2019-03-11 Object evaluation value prediction method and system, and recommendation method and system

Country Status (1)

Country Link
CN (1) CN111695024A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112233810A (en) * 2020-10-20 2021-01-15 武汉华大基因科技有限公司 Treatment scheme comprehensive curative effect evaluation method based on real world clinical data
CN112966148A (en) * 2021-03-05 2021-06-15 安徽师范大学 Video recommendation method and system based on deep learning and feature fusion
CN113076475A (en) * 2021-03-26 2021-07-06 马上消费金融股份有限公司 Information recommendation method, model training method and related equipment
CN113656681A (en) * 2021-07-08 2021-11-16 北京奇艺世纪科技有限公司 Object evaluation method, device, equipment and storage medium
CN114881689A (en) * 2022-04-26 2022-08-09 驰众信息技术（上海）有限公司 Building recommendation method and system based on matrix decomposition

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112233810A (en) * 2020-10-20 2021-01-15 武汉华大基因科技有限公司 Treatment scheme comprehensive curative effect evaluation method based on real world clinical data
CN112233810B (en) * 2020-10-20 2021-11-23 武汉华大基因科技有限公司 Treatment scheme comprehensive curative effect evaluation method based on real world clinical data
CN112966148A (en) * 2021-03-05 2021-06-15 安徽师范大学 Video recommendation method and system based on deep learning and feature fusion
CN113076475A (en) * 2021-03-26 2021-07-06 马上消费金融股份有限公司 Information recommendation method, model training method and related equipment
CN113656681A (en) * 2021-07-08 2021-11-16 北京奇艺世纪科技有限公司 Object evaluation method, device, equipment and storage medium
CN113656681B (en) * 2021-07-08 2023-08-11 北京奇艺世纪科技有限公司 Object evaluation method, device, equipment and storage medium
CN114881689A (en) * 2022-04-26 2022-08-09 驰众信息技术（上海）有限公司 Building recommendation method and system based on matrix decomposition

Similar Documents

Publication Publication Date Title
CN111222332B (en) Commodity recommendation method combining attention network and user emotion
CN109241424B (en) A kind of recommended method
CN108230058B (en) Product recommendation method and system
CN111695024A (en) Object evaluation value prediction method and system, and recommendation method and system
CN108509573B (en) Book recommendation method and system based on matrix decomposition collaborative filtering algorithm
CN110647696B (en) Business object sorting method and device
CN112308650B (en) Recommendation reason generation method, device, equipment and storage medium
CN106251174A (en) Information recommendation method and device
CN112487199B (en) User characteristic prediction method based on user purchasing behavior
EP4322031A1 (en) Recommendation method, recommendation model training method, and related product
CN114358657B (en) Post recommendation method and device based on model fusion
CN114240555A (en) Click rate prediction model training method and device and click rate prediction method and device
CN114861050A (en) Feature fusion recommendation method and system based on neural network
CN118043802A (en) Recommendation model training method and device
Wei et al. DLGNN: A Double-layer Graph Neural Network Model Incorporating Shopping Sequence Information for Commodity Recommendation.
CN112991026A (en) Commodity recommendation method, system, equipment and computer readable storage medium
CN110020195B (en) Article recommendation method and device, storage medium and electronic equipment
EP4398128A1 (en) Recommendation method and related device
CN112328918B (en) Commodity sorting method, computing device and computer-readable storage medium
CN110956528B (en) Recommendation method and system for e-commerce platform
Li et al. Research on recommendation algorithm based on e-commerce user behavior sequence
CN114429384A (en) Intelligent product recommendation method and system based on e-commerce platform
CN113641900A (en) Information recommendation method and device
CN112991025A (en) Intelligent insurance recommendation method, system and equipment and computer readable storage medium
CN111651643A (en) Processing method of candidate content and related equipment

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