CN111967946B - Commodity recommendation method and system based on user-oriented multi-relation information network - Google Patents
Commodity recommendation method and system based on user-oriented multi-relation information network Download PDFInfo
- Publication number
- CN111967946B CN111967946B CN202010919815.5A CN202010919815A CN111967946B CN 111967946 B CN111967946 B CN 111967946B CN 202010919815 A CN202010919815 A CN 202010919815A CN 111967946 B CN111967946 B CN 111967946B
- Authority
- CN
- China
- Prior art keywords
- user
- commodity
- matrix
- predicted
- recommended
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 230000007246 mechanism Effects 0.000 claims abstract description 14
- 239000011159 matrix material Substances 0.000 claims description 196
- 238000006243 chemical reaction Methods 0.000 claims description 54
- 230000009467 reduction Effects 0.000 claims description 38
- 239000013598 vector Substances 0.000 claims description 38
- 230000003993 interaction Effects 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000001627 detrimental effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Engineering & Computer Science (AREA)
- General Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a commodity recommendation method and system based on a user-oriented multi-relation information network. The method comprises the following steps: acquiring a recommended scene; acquiring a user to be predicted; according to the recommended scene, obtaining user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism; acquiring commodity characteristics of commodities to be recommended; the commodity features are a plurality of feature information of the commodity to be recommended; obtaining the score of the user to be predicted on the commodity to be recommended according to the user characteristics of the user to be predicted and the commodity characteristics of the commodity to be recommended; judging whether the score of the to-be-predicted user on the commodity to be recommended is larger than a preset score threshold value or not; if yes, recommending the commodity to be recommended to the user to be predicted; if not, not recommending the commodity to be recommended to the user to be predicted. The invention can realize accurate recommendation of commodities.
Description
Technical Field
The invention relates to the field of recommendation systems, in particular to a commodity recommendation method and system based on a user-oriented multi-relation information network.
Background
The recommendation system predicts interest preferences of the user based on the user's historical consumption record, and thus recommends goods or services to the user. Most recommendation systems consider users as isolated individuals when making recommendations, and ignore relationships between users, resulting in low accuracy of commodity recommendation or service recommendation. In fact, there are many different types of relationships between users, and considering these relationships, it may be more accurate to recommend services.
Disclosure of Invention
The invention aims to provide a commodity recommendation method and system based on a user-oriented multi-relation information network so as to realize accurate recommendation of commodities.
In order to achieve the above object, the present invention provides the following solutions:
a commodity recommendation method based on a user-oriented multi-relation information network comprises the following steps:
acquiring a recommended scene; the recommendation scene comprises a user set, a user characteristic information matrix, a user relation diagram, a commodity set, a commodity characteristic information matrix and an interaction matrix of a user and commodities; the user characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each user; the commodity characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each commodity; the user relation graph is a multi-relation information network among users;
acquiring a user to be predicted;
according to the recommended scene, obtaining user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism; the user characteristics of the users to be predicted are fused with the characteristic information of the neighbor users of the users to be predicted, and the neighbor users of the users to be predicted are a subset of the user set;
acquiring commodity characteristics of commodities to be recommended; the commodity features are a plurality of feature information of the commodity to be recommended;
obtaining the score of the user to be predicted on the commodity to be recommended according to the user characteristics of the user to be predicted and the commodity characteristics of the commodity to be recommended;
judging whether the score of the to-be-predicted user on the commodity to be recommended is larger than a preset score threshold value or not;
recommending the commodity to be recommended to the user to be predicted when the score of the user to be predicted to the commodity to be recommended is greater than a preset score threshold value;
and when the score of the to-be-predicted user on the to-be-recommended commodity is not larger than a preset score threshold value, not recommending the to-be-recommended commodity to the to-be-predicted user.
Optionally, the obtaining the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism according to the recommended scene, and further includes
Acquiring a first conversion matrix;
performing dimension reduction on the user characteristic information matrix by using a first conversion matrix to obtain a dimension-reduced user characteristic information matrix;
constructing a relationship type matrix according to the relationship type set; a one-hot vector corresponding to an ith relation type of an ith behavior of the relation type matrix; the value of the ith dimension in the one-hot vector corresponding to the ith relation type is 1, and the values of the other dimensions are 0;
acquiring a second conversion matrix;
performing dimension reduction on the relationship type matrix by using the second conversion matrix to obtain a relationship type matrix after dimension reduction; the column number of the relation type matrix after the dimension reduction is equal to the column number of the user characteristic information matrix after the dimension reduction;
acquiring a third conversion matrix;
and reducing the dimension of the commodity characteristic information matrix by using the third conversion matrix to obtain the dimension-reduced commodity characteristic information matrix.
Optionally, the obtaining, according to the recommended scenario, the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism specifically includes:
for the ith iteration, the formula is usedAcquiring the user characteristics of the current iteration, i.e. the ith iteration, of each user; wherein F is (i) (j) For the user feature of the current iteration of the jth user, A is the activation function, W i For the weight matrix of the current iteration, F (i-1) (j) For the user characteristics of the jth user in the previous iteration, i.e., the ith-1 th iteration, nei (j) is the neighbor user set of the jth user, att (j, a) j ) For the j-th user and neighbor user a j Attention weight between->For the j-th user and neighbor user a j Type of relationship between->For the j-th user and neighbor user a j Feature vector of the relation type between the feature vector +.>Obtained from a relationship type matrix E, F (i-1) (a j ) For neighbor user a j User characteristic information of the previous iteration;
judging whether the current iteration number reaches the maximum iteration number or not;
if the current iteration number reaches the maximum iteration number, determining the user characteristics of the user to be predicted in the current iteration as the final user characteristics of the user to be predicted;
if the current iteration number does not reach the maximum iteration number, updating the current iteration number, and returning to the utilization formulaAnd acquiring the user characteristics of each user current iteration, and entering the next iteration.
Optionally, the obtaining the score of the user to be predicted on the commodity to be recommended according to the user feature of the user to be predicted and the commodity feature of the commodity to be recommended specifically includes:
using the formulaObtaining the score of the to-be-predicted user on the commodity to be recommended; wherein (1)>Scoring the commodity to be recommended for the user to be predicted; sigma is a sigmoid function; w (w) p Is a conversion matrix; the I is vector splicing operation; f (F) (t) (u * ) For user u to be predicted * Is a user characteristic of (a); m (v) * ) For commodity v to be recommended * Commodity characteristics of (3).
Optionally, the predetermined score threshold is 0.5.
The invention also provides a commodity recommendation system based on the user-oriented multi-relation information network, which comprises:
the recommendation scene acquisition module is used for acquiring recommendation scenes; the recommendation scene comprises a user set, a user characteristic information matrix, a user relation diagram, a commodity set, a commodity characteristic information matrix and an interaction matrix of a user and commodities; the user characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each user; the commodity characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each commodity; the user relation graph is a multi-relation information network among users;
the user to be predicted obtaining module is used for obtaining the user to be predicted;
the user characteristic extraction module is used for acquiring the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism according to the recommended scene; the user characteristics of the users to be predicted are fused with the characteristic information of the neighbor users of the users to be predicted, and the neighbor users of the users to be predicted are a subset of the user set;
the commodity characteristic acquisition module is used for acquiring commodity characteristics of the commodity to be recommended; the commodity features are a plurality of feature information of the commodity to be recommended;
the score acquisition module is used for acquiring the score of the user to be predicted on the commodity to be recommended according to the user characteristics of the user to be predicted and the commodity characteristics of the commodity to be recommended;
the judging module is used for judging whether the score of the to-be-predicted user on the commodity to be recommended is larger than a preset score threshold value or not;
the recommending module is used for recommending the commodity to be recommended to the user to be predicted when the score of the user to be predicted to the commodity to be recommended is larger than a preset score threshold value; and when the score of the to-be-predicted user on the to-be-recommended commodity is not larger than a preset score threshold value, not recommending the to-be-recommended commodity to the to-be-predicted user.
Optionally, the method further comprises:
the first conversion matrix acquisition module is used for acquiring a first conversion matrix before acquiring the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism according to the recommended scene;
the first dimension reduction module is used for reducing the dimension of the user characteristic information matrix by utilizing a first conversion matrix to obtain a dimension-reduced user characteristic information matrix;
the relationship type matrix construction module is used for constructing a relationship type matrix according to the relationship type set; a one-hot vector corresponding to an ith relation type of an ith behavior of the relation type matrix; the value of the ith dimension in the one-hot vector corresponding to the ith relation type is 1, and the values of the other dimensions are 0;
the second conversion matrix acquisition module is used for acquiring a second conversion matrix;
the second dimension reduction module is used for reducing the dimension of the relation type matrix by using the second conversion matrix to obtain a relation type matrix after dimension reduction; the column number of the relation type matrix after the dimension reduction is equal to the column number of the user characteristic information matrix after the dimension reduction;
the third conversion matrix acquisition module is used for acquiring a third conversion matrix;
and the third dimension reduction module is used for reducing the dimension of the commodity characteristic information matrix by utilizing the third conversion matrix to obtain the dimension-reduced commodity characteristic information matrix.
Optionally, the user feature extraction module specifically includes:
a current iteration user characteristic calculation unit for utilizing the formulaAcquiring the user characteristics of the current iteration, i.e. the ith iteration, of each user; wherein F is (i) (j) The user characteristics of the current iteration of the jth user; a is an activation function; w (W) i A weight matrix for the current iteration; f (F) (i-1) (j) The user characteristics of the ith iteration which is the previous iteration of the jth user are the user characteristics of the ith-1 th iteration; nei (j) is the neighbor user set of the jth user; att (j, a) j ) For the j-th user and neighbor user a j Attention weight in between; />For the j-th user and neighbor user a j The type of relationship between them; />For the j-th user and neighbor user a j Feature vector of the relation type between the feature vector +.>Acquiring a relation type matrix E; f (F) (i-1) (a j ) For neighbor user a j User characteristic information of the previous iteration;
the judging unit is used for judging whether the current iteration number reaches the maximum iteration number or not;
the user characteristic determining unit is used for determining the user characteristic of the user to be predicted in the current iteration as the final user characteristic of the user to be predicted when the current iteration number reaches the maximum iteration number;
and the return unit is used for updating the current iteration times when the current iteration times do not reach the maximum iteration times, returning to the current iteration user characteristic calculation unit and entering the next iteration.
Optionally, the score obtaining module specifically includes:
a score calculating unit for using the formulaObtaining the score of the to-be-predicted user on the commodity to be recommended; wherein (1)>Scoring the commodity to be recommended for the user to be predicted; sigma is a sigmoid function; w (w) p Is a conversion matrix; the I is vector splicing operation; f (F) (t) (u * ) For user u to be predicted * Is a user characteristic of (a); m (v) * ) For commodity v to be recommended * Commodity characteristics of (3).
Optionally, the predetermined score threshold is 0.5.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention considers various relations among users, and uses the graph convolution technology to more accurately extract the user characteristics, thereby carrying out high-quality recommendation and improving the accuracy of a recommendation system.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a commodity recommendation method based on a user-oriented multi-relationship information network according to the present invention;
fig. 2 is a schematic structural diagram of a commodity recommendation system based on a user-oriented multi-relationship information network according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In making a recommendation of goods, the relationship between users can be regarded as a heterogram because there are a plurality of different types of relationships (friendship, relatives, classmates, etc.) between users. The invention considers various relations among users and uses the technique of graph convolution to extract more accurate user characteristics so as to recommend good quality commodities.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Fig. 1 is a flow chart of a commodity recommendation method based on a user-oriented multi-relation information network. As shown in fig. 1, the commodity recommendation method based on the user-oriented multi-relation information network of the present invention comprises the following steps:
step 100: and acquiring a recommended scene. The recommendation scene comprises a user set, a user characteristic information matrix, a user relation diagram, a commodity set, a commodity characteristic information matrix and an interaction matrix of a user and commodities; the user characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each user; the commodity characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each commodity. The recommended scenario may be represented as rec= { U, F raw ,V,M raw ,Y,G}。
Wherein U is a set of users, u= { U 1 ,u 2 ,…,u K K users are included, each element in the set representing an individual user.
For a matrix of user profile information, each row represents an initial profile of a user (e.g., age, hobbies, gender, etc. of the user), which may be represented as a vector in the D dimension. For example, three aspects of feature information of a user are known: user ID, user gender, user age. These features are represented as one-hot vectors and/or multi-hot vectors and then stitched together. Suppose there are only three users (k=3), three age groups. Each feature information is represented as follows:
then for a user (id=2, female, 15 years old) the feature information vector for the D dimension of the user is { [0,1,0] } = [0,1,0,0,1,0,1,0], d=8.
Thus, the user characteristic information matrix F can be obtained by splicing the characteristic information vectors corresponding to all users raw Each row of the matrix represents information of a user, the dimension of the matrix is K x D, theThe ith row of the matrix represents user u i Is a feature information of (a).
V is the collection of goods, v= { V 1 ,v 2 ,…,v S And S commodities. Each element in the set represents an individual commodity.
M raw In order to be a commodity characteristic information matrix,each row represents an initial characteristic of an item (e.g., ID, category, price, etc. of the item), which characteristic information may be represented as a vector of D dimensions. Assuming that there are only three items (s=3), two item categories, and three price segments, each feature information can be expressed as:
then for a D-dimensional vector of a commodity (id=2, food, 800-element), where d=8, the feature vector of the commodity is { [0,1,0] [0,1,0] } = [0,1,0,1,0,0,1,0].
Is the interaction matrix of the user and the commodity if user u a Purchased commodity v i The row a and column i of the interaction matrix Y are 1, otherwise 0.
For example, there are 2 users { u } a ，u b Sum of three commodities { v } i ，v j ，v k Presence of interaction behavior u a Purchased v i There is an interaction behaviour u b Purchased v k . The interaction matrix Y is:
[1,0,0]
[0,0,1]
G＝{U,F raw ,R,E raw and the user relation graph is a multi-relation information network among users. Since there are various relationships between users, R represents a set of J relationships＝{r 1 ,r 2 ,…r J -wherein each element of R represents a type of relationship. The various relationships may be understood as various types of relationships between users. Taking 4 relationships as an example (j=4), 1.friend, 2.relative, 3.classmate, 4.teacher and student, then the relationship set r= { friend, relative, classmate, teacher and student }, each relationship is represented by one-hot:
friends: [1,0,0,0]
Relatives: [0,1,0,0]
Classmates: [0,0,1,0]
Teachers and students: [0,0,0,1]
The 4 relation vectors are longitudinally spliced to form a relation type matrixEach row of the relationship type matrix represents a relationship in the set of relationships R. There are only 4 relationships illustrated here, and in reality more relationships may be involved, so such an E raw The dimensions of the original relationship matrix of (a) will be large and such a high-dimensional matrix will be reduced later for subsequent operations.
Step 200: and obtaining the user to be predicted.
After the user to be predicted is obtained, the commodity recommendation prediction is carried out by adopting a graph convolution neural network model. For parameters in the graph roll-up neural network model, the invention trains through back propagation, defining cross entropy as a loss function. And inputting the user to be predicted, the commodity to be recommended and the recommended scene into the graph convolution neural network model, so that a commodity recommendation result can be output.
First, before prediction, the input data arrives at the data embedding layer, and the first conversion matrix W is used n And a third conversion matrix W m For the user characteristic information matrix F respectively raw And a characteristic information matrix M of the commodity raw Dimension reduction, W n And W is m Is D x D s Wherein D is s Is far smaller than D, thus achieving the effect of reducing dimension. The formula is: f=f raw ·W n ；M＝M raw ·W m 。
And obtaining a user dense matrix dimension F and a commodity dense matrix M after dimension reduction. F has a dimension of K s The dimension of M is S.times.D s 。
Furthermore, a relationship type matrix E raw Is J x J. If J is large, then the length of the one-hot vector is large, which is detrimental to subsequent operations. For this purpose, a second conversion matrix W is used r Dimension reduction is carried out on the relation type matrix, W r Dimension J.times.D s Obtaining a relation type dense matrix E after dimension reduction, wherein the dimension is J.times.D s . The formula is: e=e raw ·W r 。
Wherein the first transformation matrix W n A second conversion matrix W r A third conversion matrix W m Is a trainable parameter in the graph roll-up neural network model.
For each user node u i There will be a different type of relationship with each neighbor, with the set rel (i) = { r i～a ,r i～b … … } where r i～a Representing user node u i With neighbor user u a Relationship between them.
Step 300: and according to the recommended scene, acquiring the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism. The user characteristics of the users to be predicted are fused with the characteristic information of the neighbor users of the users to be predicted, and the neighbor users of the users to be predicted are a subset of the user set.
The operation is performed in the convolution layer, and as a plurality of relation types exist among users, the information of the neighbor nodes and the relation information between the neighbor nodes are considered simultaneously when the information of the user nodes is updated. Specifically, for the jth user u j Its neighbors are represented by a set of users, nei (j). To update the jth user u j The information of the neighbor relation and the neighbor user needs to be considered at the same time. For this purpose, an iterative process is performed using a nonlinear transformation based on the attention mechanism. Specifically, the jth user u is updated j The characteristic information process of (a) is as follows:
wherein F is (0) =f, F is the reduced-dimension user feature information matrix; f (F) (1) (j) User characteristics of the first iteration for the jth user; a is an activation function, here ReLU is selected as the activation function (ReLU (x) =max (0, x)); w (W) 1 A weight matrix for the first iteration; f (F) (0) (j) The user characteristic information of the jth user in the user characteristic information matrix F after the dimension reduction is obtained; nei (j) is the neighbor user set of the jth user;for the j-th user and neighbor user a j The type of relationship between them;for the j-th user and neighbor user a j Feature vector of the relation type between the feature vector +.>Acquiring a relation type matrix E; f (F) (0) (a j ) For user a in the user characteristic information matrix F after dimension reduction j Is a user characteristic information of (a); att (j, a) j ) For the j-th user and neighbor user a j The attention weight between is a scalar.
att is a softmax function that,exp(x)＝e x 。
and updating the neighbor information of the jth user node once by utilizing the formula. The updating of the network is performed simultaneously for all nodes, i.e. after the first updating, the user characteristic matrix of the network is updated by the initial F (0) Become F (1) . Updating the new feature matrix F obtained once (1) Only the feature information of the surrounding 1 st order neighbors is considered. To make use of multi-hop information, we adoptThe feature information of the 2,3, … t-order neighbors around the device can be fused by the same method, wherein t is a pre-defined super parameter. That is, for each user node, the feature information of each node is updated in an iterative manner, and for the ith iteration, the formula for updating the feature information of the node is as follows:
wherein F is (i) (j) The user characteristics of the current iteration of the jth user, namely the ith iteration; a is an activation function; w (W) i A weight matrix for the current iteration; f (F) (i-1) (j) The user characteristics of the ith iteration which is the previous iteration of the jth user are the user characteristics of the ith-1 th iteration; nei (j) is the neighbor user set of the jth user; att (j, a) j ) For the j-th user and neighbor user a j Attention weight in between;for the j-th user and neighbor user a j The type of relationship between them; />For the j-th user and neighbor user a j Feature vector of the relation type between the feature vector +.>Acquiring a relation type matrix E; f (F) (i-1) (a j ) For neighbor user a j User characteristic information of the i-1 th iteration.
And then judging whether the current iteration number reaches the maximum iteration number t. If the current iteration number reaches the maximum iteration number, the user characteristic F of the user j to be predicted in the current iteration is obtained (t) (j) And determining the user characteristics of the user to be predicted finally. If the current iteration number does not reach the maximum iteration number, updating the current iteration number i=i+1, updating the user characteristics of each user by using the formula, and entering the next iteration.
Step 400: and acquiring commodity characteristics of the commodity to be recommended. The commodity features are a plurality of feature information of the commodity to be recommended, namely commodity features corresponding to the commodity to be recommended in a commodity feature information matrix.
Step 500: and obtaining the score of the commodity to be recommended of the user to be predicted according to the user characteristics of the user to be predicted and the commodity characteristics of the commodity to be recommended. Specifically, the formula is usedObtaining the score of the to-be-predicted user on the commodity to be recommended; wherein (1)>Scoring the commodity to be recommended for the user to be predicted; sigma is a sigmoid function, sigma (x) =1/(1+e) -x )；w p Is a conversion matrix; the I is vector splicing operation; f (F) (t) (u * ) For user u to be predicted * Is a user characteristic of (a); m (v) * ) For commodity v to be recommended * Commodity characteristics of (3).
Step 600: and judging whether the score of the commodity to be recommended of the user to be predicted is larger than a preset score threshold value. If so, go to step 700; if not, step 800 is performed. The score threshold value predetermined in the present invention may be set to 0.5.
Step 700: and recommending the commodity to be recommended to the user to be predicted.
Step 800: and not recommending the commodity to be recommended to the user to be predicted.
Fig. 2 is a schematic structural diagram of a commodity recommendation system based on a user-oriented multi-relationship information network according to the present invention. As shown in fig. 2, the commodity recommendation system based on the user-oriented multi-relation information network of the present invention comprises the following structures:
a recommended scene acquisition module 201, configured to acquire a recommended scene; the recommendation scene comprises a user set, a user characteristic information matrix, a user relation diagram, a commodity set, a commodity characteristic information matrix and an interaction matrix of a user and commodities; the user characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each user; the commodity characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each commodity; the user relationship graph is a multi-relationship information network between users.
And a to-be-predicted user acquisition module 202, configured to acquire the to-be-predicted user.
The user feature extraction module 203 is configured to obtain, according to the recommended scenario, a user feature of the user to be predicted by using a graph convolution method and an attention mechanism; the user characteristics of the users to be predicted are fused with the characteristic information of the neighbor users of the users to be predicted, and the neighbor users of the users to be predicted are a subset of the user set.
The commodity feature acquisition module 204 is configured to acquire commodity features of a commodity to be recommended; the commodity features are a plurality of feature information of the commodity to be recommended.
The score obtaining module 205 is configured to obtain a score of the user to be predicted on the commodity to be recommended according to the user characteristic of the user to be predicted and the commodity characteristic of the commodity to be recommended.
And the judging module 206 is configured to judge whether the score of the to-be-predicted user on the to-be-recommended commodity is greater than a preset score threshold.
A recommending module 207, configured to recommend the commodity to be recommended to the user to be predicted when the score of the user to be predicted on the commodity to be recommended is greater than a predetermined score threshold; and when the score of the to-be-predicted user on the to-be-recommended commodity is not larger than a preset score threshold value, not recommending the to-be-recommended commodity to the to-be-predicted user.
As another embodiment, the commodity recommendation system based on the user-oriented multi-relation information network of the present invention further includes:
the first conversion matrix acquisition module is used for acquiring a first conversion matrix before acquiring the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism according to the recommended scene.
And the first dimension reduction module is used for reducing the dimension of the user characteristic information matrix by utilizing the first conversion matrix to obtain the dimension-reduced user characteristic information matrix.
The relationship type matrix construction module is used for constructing a relationship type matrix according to the relationship type set; a one-hot vector corresponding to an ith relation type of an ith behavior of the relation type matrix; the value of the ith dimension in the one-hot vector corresponding to the ith relation type is 1, and the values of the other dimensions are 0.
And the second conversion matrix acquisition module is used for acquiring the second conversion matrix.
The second dimension reduction module is used for reducing the dimension of the relation type matrix by using the second conversion matrix to obtain a relation type matrix after dimension reduction; the column number of the relation type matrix after the dimension reduction is equal to the column number of the user characteristic information matrix after the dimension reduction;
the third conversion matrix acquisition module is used for acquiring a third conversion matrix;
and the third dimension reduction module is used for reducing the dimension of the commodity characteristic information matrix by utilizing the third conversion matrix to obtain the dimension-reduced commodity characteristic information matrix.
As another embodiment, in the commodity recommendation system based on the user-oriented multi-relationship information network of the present invention, the user feature extraction module 203 specifically includes:
a current iteration user characteristic calculation unit for utilizing the formulaAcquiring the user characteristics of the current iteration, i.e. the ith iteration, of each user; wherein F is (i) (j) The user characteristics of the current iteration of the jth user; a is an activation function; w (W) i A weight matrix for the current iteration; f (F) (i-1) (j) The user characteristics of the ith iteration which is the previous iteration of the jth user are the user characteristics of the ith-1 th iteration; nei (j) is the neighbor user set of the jth user; att (j, a) j ) For the j-th user and neighbor user a j Attention weight in between; />For the jth user and neighborUser a j The type of relationship between them; />For the j-th user and neighbor user a j Feature vector of the relation type between the feature vector +.>Acquiring a relation type matrix E; f (F) (i-1) (a j ) For neighbor user a j User characteristic information of the i-1 th iteration.
And the judging unit is used for judging whether the current iteration number reaches the maximum iteration number.
And the user characteristic determining unit is used for determining the user characteristic of the user to be predicted in the current iteration as the final user characteristic of the user to be predicted when the current iteration number reaches the maximum iteration number.
And the return unit is used for updating the current iteration times when the current iteration times do not reach the maximum iteration times, returning to the current iteration user characteristic calculation unit and entering the next iteration.
As another embodiment, the commodity recommendation system of the present invention based on the user-oriented multi-relationship information network, the score obtaining module 205 specifically includes:
a score calculating unit for using the formulaObtaining the score of the to-be-predicted user on the commodity to be recommended; wherein (1)>Scoring the commodity to be recommended for the user to be predicted; sigma is a sigmoid function; w (w) p Is a conversion matrix; the I is vector splicing operation; f (F) (t) (u * ) For user u to be predicted * Is a user characteristic of (a); m (v) * ) For commodity v to be recommended * Commodity characteristics of (3).
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (8)
1. A commodity recommendation method based on a user-oriented multi-relation information network is characterized by comprising the following steps:
acquiring a recommended scene; the recommendation scene comprises a user set, a user characteristic information matrix, a user relation diagram, a commodity set, a commodity characteristic information matrix and an interaction matrix of a user and commodities; the user characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each user; the commodity characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each commodity; the user relation graph is a multi-relation information network among users;
acquiring a first conversion matrix;
performing dimension reduction on the user characteristic information matrix by using the first conversion matrix to obtain a dimension-reduced user characteristic information matrix;
constructing a relationship type matrix according to the relationship type set; a one-hot vector corresponding to an ith relation type of an ith behavior of the relation type matrix; the value of the ith dimension in the one-hot vector corresponding to the ith relation type is 1, and the values of the other dimensions are 0;
acquiring a second conversion matrix;
performing dimension reduction on the relationship type matrix by using the second conversion matrix to obtain a relationship type matrix after dimension reduction; the column number of the relation type matrix after the dimension reduction is equal to the column number of the user characteristic information matrix after the dimension reduction;
acquiring a third conversion matrix;
performing dimension reduction on the commodity characteristic information matrix by using the third conversion matrix to obtain a dimension-reduced commodity characteristic information matrix;
using the first conversion matrix W n And the third conversion matrix W m Respectively to the user characteristic information matrix F raw And a characteristic information matrix M of the commodity raw Performing dimension reduction; the dimensionality reduction formula is: f=f raw ·W n ；M＝M raw ·W m The method comprises the steps of carrying out a first treatment on the surface of the Wherein F is a user dense matrix, and M is a commodity dense matrix;
acquiring a user to be predicted;
according to the recommended scene, obtaining user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism; the user characteristics of the user to be predicted are fused with the characteristic information of the neighbor users of the user to be predicted;
acquiring commodity characteristics of commodities to be recommended; the commodity features are a plurality of feature information of the commodity to be recommended;
obtaining the score of the user to be predicted on the commodity to be recommended according to the user characteristics of the user to be predicted and the commodity characteristics of the commodity to be recommended;
judging whether the score of the to-be-predicted user on the commodity to be recommended is larger than a preset score threshold value or not;
recommending the commodity to be recommended to the user to be predicted when the score of the user to be predicted to the commodity to be recommended is greater than a preset score threshold value;
and when the score of the to-be-predicted user on the to-be-recommended commodity is not larger than a preset score threshold value, not recommending the to-be-recommended commodity to the to-be-predicted user.
2. The commodity recommending method based on the user-oriented multi-relation information network according to claim 1, wherein the acquiring the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism according to the recommended scene specifically comprises:
for the ith iteration, the formula is usedAcquiring the user characteristics of the current iteration, i.e. the ith iteration, of each user; wherein F is (i) (j) The user characteristics of the current iteration of the jth user; a is an activation function; w (W) i A weight matrix for the current iteration; f (F) (i-1) (j) The user characteristics of the ith iteration which is the previous iteration of the jth user are the user characteristics of the ith-1 th iteration; nei (j) is the neighbor user set of the jth user; att (j, a) j ) For the j-th user and neighbor user a j Attention weight in between; />For the j-th user and neighbor user a j The type of relationship between them; />For the j-th user and neighbor user a j A feature vector of a relationship type between the two; f (F) (i-1) (a j ) For neighbor user a j User characteristic information of the previous iteration;
judging whether the current iteration number reaches the maximum iteration number or not;
if the current iteration number reaches the maximum iteration number, determining the user characteristics of the user to be predicted in the current iteration as the final user characteristics of the user to be predicted;
if the current iteration number does not reach the maximum iteration number, updating the current iteration number, and returning to the utilization formulaAnd acquiring the user characteristics of each user current iteration, and entering the next iteration.
3. The commodity recommendation method based on the user-oriented multi-relation information network according to claim 1, wherein the obtaining the score of the user to be predicted for the commodity to be recommended according to the user characteristics of the user to be predicted and the commodity characteristics of the commodity to be recommended specifically includes:
using the formulaObtaining the score of the to-be-predicted user on the commodity to be recommended; wherein (1)>Scoring the commodity to be recommended for the user to be predicted; sigma is a sigmoid function; w (w) p Is a conversion matrix; the I is vector splicing operation; f (F) (t) (u * ) For user u to be predicted * Is a user characteristic of (a); m (v) * ) For commodity v to be recommended * Commodity characteristics of (3).
4. The method for recommending items based on a user-oriented multi-relationship information network according to claim 1, wherein the predetermined score threshold is 0.5.
5. A merchandise recommendation system based on a user-oriented multi-relationship information network, comprising:
the recommendation scene acquisition module is used for acquiring recommendation scenes; the recommendation scene comprises a user set, a user characteristic information matrix, a user relation diagram, a commodity set, a commodity characteristic information matrix and an interaction matrix of a user and commodities; the user characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each user; the commodity characteristic information matrix is a matrix formed by a plurality of characteristic information corresponding to each commodity; the user relation graph is a multi-relation information network among users;
the first conversion matrix acquisition module is used for acquiring a first conversion matrix before acquiring the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism according to the recommended scene;
the first dimension reduction module is used for reducing the dimension of the user characteristic information matrix by utilizing the first conversion matrix to obtain a dimension-reduced user characteristic information matrix;
the relationship type matrix construction module is used for constructing a relationship type matrix according to the relationship type set; a one-hot vector corresponding to an ith relation type of an ith behavior of the relation type matrix; the value of the ith dimension in the one-hot vector corresponding to the ith relation type is 1, and the values of the other dimensions are 0;
the second conversion matrix acquisition module is used for acquiring a second conversion matrix;
the second dimension reduction module is used for reducing the dimension of the relation type matrix by using the second conversion matrix to obtain a relation type matrix after dimension reduction; the column number of the relation type matrix after the dimension reduction is equal to the column number of the user characteristic information matrix after the dimension reduction;
the third conversion matrix acquisition module is used for acquiring a third conversion matrix;
the third dimension reduction module is used for reducing the dimension of the commodity characteristic information matrix by utilizing the third conversion matrix to obtain a dimension-reduced commodity characteristic information matrix;
the first conversion matrix Wn and the third conversion matrix Wm are utilized to respectively reduce the dimension of the user characteristic information matrix Fraw and the dimension of the characteristic information matrix Mraw of the commodity; the dimensionality reduction formula is: f=fraw·wn; m=mraw·wm; wherein F is a user dense matrix, and M is a commodity dense matrix;
the user to be predicted obtaining module is used for obtaining the user to be predicted;
the user characteristic extraction module is used for acquiring the user characteristics of the user to be predicted by using a graph convolution method and an attention mechanism according to the recommended scene; the user characteristics of the user to be predicted are fused with the characteristic information of the neighbor users of the user to be predicted;
the commodity characteristic acquisition module is used for acquiring commodity characteristics of the commodity to be recommended; the commodity features are a plurality of feature information of the commodity to be recommended;
the score acquisition module is used for acquiring the score of the user to be predicted on the commodity to be recommended according to the user characteristics of the user to be predicted and the commodity characteristics of the commodity to be recommended;
the judging module is used for judging whether the score of the to-be-predicted user on the commodity to be recommended is larger than a preset score threshold value or not;
the recommending module is used for recommending the commodity to be recommended to the user to be predicted when the score of the user to be predicted to the commodity to be recommended is larger than a preset score threshold value; and when the score of the to-be-predicted user on the to-be-recommended commodity is not larger than a preset score threshold value, not recommending the to-be-recommended commodity to the to-be-predicted user.
6. The commodity recommendation system based on a user-oriented multi-relationship information network according to claim 5, wherein the user feature extraction module specifically comprises:
a current iteration user characteristic calculation unit for utilizing the formulaAcquiring the user characteristics of the current iteration, i.e. the ith iteration, of each user; wherein F is (i) (j) The user characteristics of the current iteration of the jth user; a is an activation function; w (W) i A weight matrix for the current iteration; f (F) (i-1) (j) The user characteristics of the ith iteration which is the previous iteration of the jth user are the user characteristics of the ith-1 th iteration; nei (j) is the neighbor user set of the jth user; att (j, a) j ) For the j-th user and neighbor user a j Attention weight in between; />For the j-th user and neighbor user a j The type of relationship between them; />For the j-th user and neighbor user a j A feature vector of a relationship type between the two; f (F) (i-1) (a j ) For neighbor user a j User characteristic information of the previous iteration;
the judging unit is used for judging whether the current iteration number reaches the maximum iteration number or not;
the user characteristic determining unit is used for determining the user characteristic of the user to be predicted in the current iteration as the final user characteristic of the user to be predicted when the current iteration number reaches the maximum iteration number;
and the return unit is used for updating the current iteration times when the current iteration times do not reach the maximum iteration times, returning to the current iteration user characteristic calculation unit and entering the next iteration.
7. The merchandise recommendation system based on a user oriented multi-relationship information network according to claim 5, wherein said score acquisition module specifically comprises:
a score calculating unit for using the formulaObtaining the score of the to-be-predicted user on the commodity to be recommended; wherein (1)>Scoring the commodity to be recommended for the user to be predicted; sigma is a sigmoid function; w (w) p Is a conversion matrix; the I is vector splicing operation; f (F) (t) (u * ) For user u to be predicted * Is a user characteristic of (a); m (v) * ) For commodity v to be recommended * Commodity characteristics of (3).
8. The user-oriented multi-relationship information network-based commodity recommendation system according to claim 5, wherein said predetermined score threshold is 0.5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010919815.5A CN111967946B (en) | 2020-09-04 | 2020-09-04 | Commodity recommendation method and system based on user-oriented multi-relation information network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010919815.5A CN111967946B (en) | 2020-09-04 | 2020-09-04 | Commodity recommendation method and system based on user-oriented multi-relation information network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111967946A CN111967946A (en) | 2020-11-20 |
| CN111967946B true CN111967946B (en) | 2024-04-12 |
Family
ID=73392095
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010919815.5A Active CN111967946B (en) | 2020-09-04 | 2020-09-04 | Commodity recommendation method and system based on user-oriented multi-relation information network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111967946B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114496165A (en) * | 2022-01-27 | 2022-05-13 | 国家食品安全风险评估中心 | Food diet recommendation method based on heteromorphic graph |
| CN114757745A (en) * | 2022-04-25 | 2022-07-15 | 北京航空航天大学 | E-market scene mining method and system based on joint matrix decomposition |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110737778A (en) * | 2019-09-04 | 2020-01-31 | 北京邮电大学 | Knowledge graph and Transformer based patent recommendation method |
| CN111160954A (en) * | 2019-12-16 | 2020-05-15 | 南京理工大学 | Recommendation method facing group object based on graph convolution network model |
| CN111428147A (en) * | 2020-03-25 | 2020-07-17 | 合肥工业大学 | Social recommendation method of heterogeneous graph volume network combining social and interest information |
| CN111538868A (en) * | 2020-04-28 | 2020-08-14 | 中国科学技术大学 | Knowledge tracking method and exercise recommendation method |
-
2020
- 2020-09-04 CN CN202010919815.5A patent/CN111967946B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110737778A (en) * | 2019-09-04 | 2020-01-31 | 北京邮电大学 | Knowledge graph and Transformer based patent recommendation method |
| CN111160954A (en) * | 2019-12-16 | 2020-05-15 | 南京理工大学 | Recommendation method facing group object based on graph convolution network model |
| CN111428147A (en) * | 2020-03-25 | 2020-07-17 | 合肥工业大学 | Social recommendation method of heterogeneous graph volume network combining social and interest information |
| CN111538868A (en) * | 2020-04-28 | 2020-08-14 | 中国科学技术大学 | Knowledge tracking method and exercise recommendation method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111967946A (en) | 2020-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110275964B (en) | Recommendation model based on knowledge graph and cyclic neural network | |
| CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
| CN108460619B (en) | Method for providing collaborative recommendation model fusing explicit and implicit feedback | |
| CN111859166B (en) | Article scoring prediction method based on improved graph convolution neural network | |
| CN111949865A (en) | Interest point recommendation method based on graph neural network and user long-term and short-term preference | |
| CN106779867B (en) | Support vector regression recommendation method and system based on context awareness | |
| CN110020022B (en) | Data processing method, device, equipment and readable storage medium | |
| CN111967946B (en) | Commodity recommendation method and system based on user-oriented multi-relation information network | |
| CN116204729B (en) | Cross-domain group intelligent recommendation method based on hypergraph neural network | |
| CN115270001B (en) | Privacy protection recommendation method and system based on cloud collaborative learning | |
| US20230267317A1 (en) | Sign-aware recommendation apparatus and method using graph neural network | |
| Kang | Combining grey relationship analysis and neural network to develop attractive automobile booth design | |
| CN113255798A (en) | Classification model training method, device, equipment and medium | |
| CN112508256A (en) | User demand active prediction method and system based on crowdsourcing | |
| CN114971784B (en) | Session recommendation method and system based on graph neural network by fusing self-attention mechanism | |
| CN114065048A (en) | Article recommendation method based on multi-different-pattern neural network | |
| CN114386513A (en) | Interactive grading prediction method and system integrating comment and grading | |
| Ahamed et al. | A recommender system based on deep neural network and matrix factorization for collaborative filtering | |
| CN114298783A (en) | Commodity recommendation method and system based on matrix decomposition and fusion of user social information | |
| US20240037133A1 (en) | Method and apparatus for recommending cold start object, computer device, and storage medium | |
| CN112784177B (en) | Spatial distance adaptive next interest point recommendation method | |
| CN111815410B (en) | Commodity recommendation method based on selective neighborhood information | |
| CN113610610A (en) | Session recommendation method and system based on graph neural network and comment similarity | |
| Crabbe et al. | Improving the efficiency of individualized designs for the mixed logit choice model by including covariates | |
| CN116204723A (en) | Social recommendation method based on dynamic hypergraph representation learning |
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 |