CN112612955A - Product pushing method and system based on deep learning - Google Patents
Product pushing method and system based on deep learning Download PDFInfo
- Publication number
- CN112612955A CN112612955A CN202011508268.8A CN202011508268A CN112612955A CN 112612955 A CN112612955 A CN 112612955A CN 202011508268 A CN202011508268 A CN 202011508268A CN 112612955 A CN112612955 A CN 112612955A
- Authority
- CN
- China
- Prior art keywords
- product
- matrix
- user
- model parameter
- current
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a product pushing method and system based on deep learning, and belongs to the technical field of artificial intelligence. The product pushing method based on deep learning comprises the following steps: constructing a current user information matrix according to the user identity information; creating a current product information matrix according to the order of clicking products by a user; inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability; and determining a pushed product according to the product recommendation probability and pushing the pushed product. The invention can push specific products for different users, improve the precision and accuracy of product pushing and further improve the satisfaction of the users.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a product pushing method and system based on deep learning.
Background
With the rapid development of technologies such as cloud computing, big data, internet of things and the like, the emergence of various applications in the internet space causes explosive growth of data scale. The big data brings revolutionary development to the human society and also brings serious 'information overload' problem, and product pushing is an effective method for solving the problem, and can analyze the preference and interest of the user according to the historical behaviors of the user so as to predict objects which the user may be interested in next, thereby pushing corresponding products to the user. However, most of the existing pushing methods only analyze from the perspective of users, and influence of products on user preferences is not considered, so that the precision and accuracy of product pushing are greatly influenced, and the user satisfaction is reduced.
Disclosure of Invention
The embodiment of the invention mainly aims to provide a product pushing method and system based on deep learning, so as to push specific products for different users, improve the precision and accuracy of product pushing and further improve the satisfaction of the users.
In order to achieve the above object, an embodiment of the present invention provides a product pushing method based on deep learning, including:
constructing a current user information matrix according to the user identity information;
creating a current product information matrix according to the order of clicking products by a user;
inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;
and determining a pushed product according to the product recommendation probability and pushing the pushed product.
The embodiment of the invention also provides a product pushing system based on deep learning, which comprises:
the current user information matrix unit is used for constructing a current user information matrix according to the user identity information;
the current product information matrix unit is used for creating a current product information matrix according to the sequence of products clicked by users;
the product recommendation probability unit is used for inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;
and the product pushing unit is used for determining a pushed product according to the product recommendation probability and pushing the pushed product.
The embodiment of the present invention further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of the deep learning-based product push method are implemented.
The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the deep learning-based product pushing method are implemented.
According to the product pushing method and system based on deep learning, a current user information matrix is constructed according to user identity information, a current product information matrix is created according to the sequence of products clicked by users, then the current user information matrix and the current product information matrix are input into a product recommendation model to obtain the product recommendation probability, finally, a pushed product is determined according to the product recommendation probability and is pushed, specific products can be pushed according to different users, the product pushing precision and accuracy are improved, and the user satisfaction is further improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a flowchart of a deep learning-based product pushing method according to an embodiment of the present invention;
FIG. 2 is a flowchart of a product pushing method based on deep learning according to another embodiment of the present invention;
fig. 3 is a flowchart of S102;
FIG. 4 is a flow diagram of creating a product recommendation model in an embodiment of the present invention;
FIG. 5 is a flowchart of S302 in an embodiment of the present invention;
FIG. 6 is a schematic diagram of constructing a directed graph in the embodiment of the present invention;
FIG. 7 is a schematic diagram of a product information matrix according to an embodiment of the present invention;
FIG. 8 is a block diagram of a product pushing system based on deep learning according to an embodiment of the present invention;
fig. 9 is a block diagram of a computer device in the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As will be appreciated by one skilled in the art, embodiments of the present invention may be embodied as a system, apparatus, device, method, or computer program product. Accordingly, the present disclosure may be embodied in the form of: entirely hardware, entirely software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.
In view of the fact that the influence of a product on user preference is not considered in the prior art, the precision and accuracy of product pushing are greatly influenced, and the user satisfaction is reduced, the embodiment of the invention provides a product pushing method based on deep learning, and the defects that single-angle modeling, insufficient feature utilization and incomplete behavior sequence analysis are overcome in the conventional recommendation algorithm. The invention has the following advantages: first, modeling from the perspective of the user and the product, respectively, may simultaneously consider the user's own long-term preferences and the user's short-term purchasing interests, thereby more accurately capturing the user's current interests. Second, by analyzing the user and the product separately with a variety of features, a more accurate representation of the user and the product is obtained. And thirdly, explicitly modeling time interval information and clicking times in clicking behaviors of the user, and determining the preference degree of the user on a product through the stay time of the user on the product. In summary, the present invention aims to utilize a deep learning method to fuse multiple features and model the behavior records of a user from two aspects of the user and a product, so as to improve the product pushing accuracy, further improve the user satisfaction and increase the product purchase rate. The present invention will be described in detail below with reference to the accompanying drawings.
Fig. 1 is a flowchart of a product pushing method based on deep learning in an embodiment of the present invention. Fig. 2 is a flowchart of a product pushing method based on deep learning according to another embodiment of the present invention. As shown in fig. 1-2, the product push method based on deep learning includes:
s101: and constructing a current user information matrix according to the user identity information.
In specific implementation, the current user information matrix can be constructed by referring to the way of constructing the historical user information matrix.
The steps of constructing the historical user information matrix are as follows:
a. obtaining user representation vector u according to ID number of useri1. Firstly, a user set U ═ U is obtained from a user data set1,u2,…,uK]Where K is the number of users. Obtaining a vector representation of each user by using one-hot vectors, and then representing the vector representation to a d-dimensional representation space to obtain ui1。
b. Vectorizing the user identity (students, office workers, farmers and the like) to obtain a vector ui2；
c. And segmenting the income of the user and vectorizing and representing the income. For example, let the income interval of all users be [2000,5000 ]]Then can be segmented into [2000,3000 ], [3000,4000) and [4000,5000 ]]The three segments are numbered as 1, 2 and 3 respectively, and then the user can be marked as 1, 2 or 3 according to the income condition of the userObtaining the characterization vector u of the income situation of the useri3。
d. Fusing the three vectors obtained in the steps a, b and c to obtain the long-term preference (user information matrix) of the user, namely the historical user information matrix ui＝[ui1；ui2；ui3]Wherein [;]representing vector stitching.
S102: and creating a current product information matrix according to the sequence of clicking the products by the user.
Wherein the current product information matrix may be constructed with reference to a manner of constructing the historical product information matrix. The current product information matrix and the historical product information matrix both comprise a product ID matrix, a product category matrix and a product price matrix.
Fig. 3 is a flowchart of S102. As shown in fig. 3, S102 includes:
s201: a product ID sequence is created according to the order in which the user clicks on the product.
In specific implementation, the products are mapped to a uniform characterization space according to the product ID numbers. Firstly, extracting all non-repeated products interacted by users from interaction data of the users and the products to obtain a product ID sequence X [ < X >1,x2,…,xn]Wherein n is the number of non-repeating products. Obtaining the representation of each product by using one-hot vectors, then representing the representation to a d-dimensional representation space to obtain the representation v of the product IDi。
S202: and determining a product category sequence and a product price sequence according to the product ID sequence.
In specific implementation, the categories of the products can be mapped to a uniform characterization space according to the category IDs. Firstly, non-repeated product categories are extracted from a product data set to obtain a product category sequence C ═ C1,c2,…,cm]Wherein m represents the number of non-repeating categories, and m ≦ n. Obtaining the representation of each category by using one-hot vectors, then representing the representation to a d-dimensional representation space, and obtaining the representation m of the product categoryi。
In the same way, the product price is processed in the same way as the income of the user, namely the product price is mapped toCorresponding interval to obtain product price sequence Ri＝[r1,r2,......rp]Wherein p represents the number of price intervals and p is less than or equal to n. Obtaining the representation of each price interval by using one-hot vectors, then representing the price intervals to a d-dimensional representation space, and obtaining the representation p of the product pricei。
S203: and generating a product ID matrix according to the product ID sequence and the residence time of the user on each product.
FIG. 6 is a schematic diagram of constructing a directed graph in the embodiment of the present invention. As shown in FIG. 6, the historical sequence of interaction behavior of a user (either the current user or the historical user) may be constructed as a directed graph in FIG. 6 by the time of the click event. User in FIG. 61、user2And user3Each representing a different user.
As shown in fig. 6, according to the product ID sequence X ═ X1,x2,…,xn]And constructing a directed graph of the product ID sequence. Ith node x in the directed graphiProduct representing user's ith click, each edge (x)i-1,xi) Representing the user clicking x firsti-1Then click on xi。
Fig. 7 is a schematic diagram of a product information matrix in the embodiment of the present invention. As shown in fig. 7, with the user in fig. 61For example, consider that the user's dwell time on a product reflects the user's preference for each product, i.e., longer dwell times indicate more interest in the product. Therefore, when constructing the product ID sequence directed graph, each edge in the directed graph can be weighted according to the dwell time (i.e. time interval) of the user on each product, and the weight calculation formula is as follows:
wherein, χ1There is a weight for the product ID sequence to the edge in the graph. As shown in FIG. 7, a product ID Out-edge Matrix (Out-edge Matrix) can be constructed according to the weight of each edge in the product ID sequence directed graph
And product ID into edge Matrix (In-edge Matrix)
And then will
And
splicing to generate a product ID matrix Ax。
For example, x in FIG. 71Point of direction x2，x2Point of direction x3，x3Point of direction x4User at product x1、x2、x3The residence times were 20s, 40s and 60s, respectively, so that x1x2The weight of the edge is 1/6, x2x3The weight of the edge is 1/3, x3x1The weight of the edge is 1/2, the product ID edge matrix
The first row, second column, elements are 1/6, the second row, third column, elements are 1/3, the third row, fourth column, elements are 1/2, and the other elements in the product ID edge matrix are 0; product ID edge-entering matrix
The second row and first column of (a) is 1/6, the third row and second column is 1/3, the fourth row and third column is 1/2, and the other elements in the product ID edge matrix are 0.
S204: and generating a product category matrix according to the product category sequence, and generating a product price matrix according to the product price sequence.
As shown in fig. 6, the sequence C ═ C according to the product category1,c2,...,cm]And constructing a product category sequence directed graph. I.e. product x1Is of class c1Product xnIs of class cm. The product is put intoThe category sequence is constructed into a product category sequence directed graph, each node represents a category, and each edge (c)i-1,ci) Representing the user clicking on category ci-1After the product is clicked on the category ciThe product of (1).
As shown in fig. 6, according to the product price sequence Ri＝[r1,r2,……rp]And constructing a product price sequence directed graph. I.e. product x1The price interval is r1Product xnThe price interval is rp. Constructing the product price sequence into a product price sequence directed graph, wherein each node represents a price interval and each edge (r)i-1,ri) Representing the user clicking on price ri-1After the product is clicked, the price is riThe product of (1).
As shown in fig. 7, the weight of the edge in the product category sequence directed graph or the product price sequence directed graph is determined according to the out-degree or in-degree of the node in the product category sequence directed graph or the product price sequence directed graph and the occurrence number of the edge. In specific implementation, the weight of the edge in the product category sequence directed graph or the product price sequence directed graph can be determined through the following formula:
wherein, χ2And weighting the edges in the product category sequence directed graph or the product price sequence directed graph.
As shown in FIG. 7, a product category Out-edge Matrix (Out-edge Matrix) can be constructed according to the weight of each edge in the product category sequence directed graph
And product categories into edge matrices (In-edge Matrix)
And then will
And
splicing to generate a product category matrix Ac. When the denominator in the formula is the out degree of the node, the corresponding weight is an element in the product category out-edge matrix; when the denominator in the formula is the entry of the node, the corresponding weight is an element in the product category entry edge matrix. For example, c in FIG. 71Point of direction c2，c2Point of direction c3，c3Point of direction c1Thus product category edge matrix
The elements of the second row and the second column in the first row, the elements of the third row and the third column in the second row and the elements of the first column in the third row are all 1, and other elements in the product category edge-out matrix are 0; product categories are incorporated into edge matrices (In-edge Matrix)
The elements of the first row and the third column, the elements of the first row and the first column, and the elements of the third row and the second column are all 1, and the other elements in the product category edge-entering matrix are 0.
As shown in FIG. 7, a product price edge Matrix (Out-edge Matrix) can be constructed according to the weight of each edge in the product price sequence directed graph
And product price In edge Matrix (In-edge Matrix)
And then will
And
splicing to generate a product price matrix Ar. When denominator in the formula is sectionWhen the point is out, the corresponding weight is the element in the product price edge matrix; and when the denominator in the formula is the entry of the node, the corresponding weight is an element in the product price entry edge matrix. For example, r in FIG. 71Direction r2，r2Direction r1，r1Direction r3Thus, the product price edge matrix
The elements of the second row and the second column in the first row and the elements of the third column in the first row are 1/2, the elements of the first column in the second row are 1, and the other elements in the product price edge matrix are 0; product categories are incorporated into edge matrices (In-edge Matrix)
The elements of the second row and the second column, the elements of the first column of the second row and the elements of the first column of the third row are all 1, and the other elements in the product category edge-entering matrix are 0.
S103: and inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain the product recommendation probability.
In specific implementation, the product recommendation model outputs a product recommendation probability corresponding to each current product.
S104: and determining a pushed product according to the product recommendation probability and pushing the pushed product.
When the method is specifically implemented, the current product corresponding to the maximum value of the product recommendation probability can be determined to be a push product, and the push product is pushed to the current user.
The execution subject of the deep learning based product push method shown in fig. 1 may be a computer. As can be seen from the flow shown in fig. 1, the product push method based on deep learning according to the embodiment of the present invention first constructs a current user information matrix according to user identity information, creates a current product information matrix according to a sequence in which users click products, then inputs the current user information matrix and the current product information matrix into a product recommendation model to obtain a product recommendation probability, and finally determines a push product according to the product recommendation probability and pushes the push product.
FIG. 4 is a flow diagram of creating a product recommendation model in an embodiment of the present invention. As shown in fig. 4, the step of creating a product recommendation model based on the historical user information matrix and the historical product information matrix includes:
the following iterative process is performed:
s301: and generating a product matrix according to the historical product information matrix and the first model parameter matrix.
Wherein the first model parameter matrix comprises a first parameter matrix H and a second parameter matrix WzA third parameter matrix WrA fourth parameter matrix WoThe fifth parameter matrix UzThe sixth parameter matrix UrAnd a seventh parameter matrix Uo。
In specific implementation, the historical product information matrix comprises a product ID matrix AxProduct category matrix AcAnd a product price matrix Ar. As described above, using the representations of the neural network learning nodes for the product ID sequence, the product category sequence, and the product price sequence, respectively, can yield a representation v of the product IDiCharacterization m of product categoriesiAnd characterization of product price pi. Characterization of neural network learning nodes using gated graphs, with characterization of product ID viFor example, the formula for node update is as follows:
wherein the content of the first and second substances,
for the characterization of the ith node at time t,
for the product ID matrix AxRow i of (a), when the node represents a product category or a product price respectively,
may be replaced with a product category matrix acRow i of (1)
Or product price matrix ArRow i of (1)
Represents the product ID sequence clicked by the user at the time of t-1, b is an offset vector, sigma () is a Sigmoid function,
to represent the characterization of the ith product clicked on by the user at time t-1,
and representing that the user clicks the characterization of the ith product at the time t. The representation that the user clicks the ith product at the moment t can be obtained in the same way
And the representation of the price interval of the ith product clicked by the user at the moment t
Characterization of products learned via graph neural network
And
performing concat fusion to obtain a product matrix q of the ith productiI.e. by
S302: and determining the product prediction recommendation probability according to the product matrix, the second model parameter matrix, the third model parameter matrix and the historical user information matrix.
Fig. 5 is a flowchart of S302 in the embodiment of the present invention. As shown in fig. 5, S302 includes:
s401: and generating a user initial interest characteristic matrix according to the product matrix and the second model parameter matrix.
Wherein the second model parameter matrix comprises an eighth parameter matrix W and a ninth parameter matrix W2And a tenth parameter matrix W3And an eleventh parameter matrix W4。
In specific implementation, the attention mechanism is combined to obtain the global preference s of the userglobal(ii) a Then the product matrix q of the first product clicked by the user1And the product matrix q of the last product clicked on by the usernAs local preference S of a userlocalThe formula is as follows:
αi＝wTσ(W2qn+W3qi+c)；
slocal＝q1+qn；
wherein c is a constant.
Finally, global preference s of the user is determinedglobalAnd local preferences S of the userlocalPerforming concat fusion to obtain a user initial interest characteristic matrix (user purchase interest representation) spThe following are:
sp＝W4[sglobal；slocal]。
s402: and generating a final user interest characteristic matrix according to the third model parameter matrix, the initial user interest characteristic matrix and the historical user information matrix.
Wherein the third model parameter matrix comprises a twelfth parameter matrix W1And a thirteenth parameter matrix W5。
In specific implementation, the long-term preference of the user and the purchase interest expression of the user can be fused through the following formula to obtain the final interest characteristic matrix u of the userfi：
u'i＝W1ui＝W1[ui1；ui2；ui3]；
ufi＝W5[u'i；sp]。
S403: and determining the product prediction recommendation probability according to the final interest characteristic matrix of the user and the product matrix.
In specific implementation, the product prediction recommendation probability can be determined by the following formula:
wherein the content of the first and second substances,
the predicted recommended probability for the ith product.
S303: and determining a loss function according to the product forecast recommendation probability and the actually purchased product.
In particular, the loss function may be determined by the following formula:
wherein the content of the first and second substances,
as a loss function, yiThe one-hot vector of the actual purchased product is shown, wherein lambda is a first preset parameter, and theta is a second preset parameter.
S304: and judging whether the loss function is smaller than a preset value or not.
S305: and when the loss function is smaller than a preset value, creating a product recommendation model according to the first model parameter matrix, the second model parameter matrix and the third model parameter matrix in the current iteration.
S306: and when the loss function is larger than or equal to the preset value, updating the first model parameter matrix, the second model parameter matrix and the third model parameter matrix according to the loss function, and continuously executing iterative processing.
In particular implementation, when the loss function is greater than or equal to the preset value, the first model parameter matrix, the second model parameter matrix and the third model parameter matrix may be updated by using a time back propagation algorithm (BPTT) to train the product recommendation model.
The specific process of the embodiment of the invention is as follows:
1. and generating a product matrix according to the historical product information matrix and the first model parameter matrix.
2. And generating a user initial interest characteristic matrix according to the product matrix and the second model parameter matrix.
3. And generating a final user interest characteristic matrix according to the third model parameter matrix, the initial user interest characteristic matrix and the historical user information matrix.
4. And determining the product prediction recommendation probability according to the final interest characteristic matrix of the user and the product matrix.
5. And determining a loss function according to the product forecast recommendation probability and the actually purchased product.
6. And when the loss function is smaller than a preset value, establishing a product recommendation model according to the first model parameter matrix, the second model parameter matrix and the third model parameter matrix in the current iteration, otherwise, updating the first model parameter matrix, the second model parameter matrix and the third model parameter matrix according to the loss function, and returning to the step 1.
7. And constructing a current user information matrix according to the user identity information.
8. And creating a product ID sequence according to the sequence of products clicked by the user, and determining a product category sequence and a product price sequence according to the product ID sequence.
9. And generating a product ID matrix according to the product ID sequence and the residence time of the user on each product.
10. And generating a product category matrix according to the product category sequence, and generating a product price matrix according to the product price sequence.
11. And inputting the current user information matrix, the product ID matrix, the product category matrix and the product price matrix into a product recommendation model to obtain the product recommendation probability.
12. And determining that the current product corresponding to the maximum value of the product recommendation probability is a pushed product, and pushing the pushed product to the current user.
In summary, the invention discovers the interest of the user by modeling the user identity and the product click sequence of the user, predicts the next product that the user may purchase, and further pushes the product for the user, and has the following advantages:
1. the method and the system respectively perform modeling from the user perspective and the product perspective, discover the long-term preference of the user by modeling the characteristics of the user, and mine the short-term purchasing interest of the user by modeling the product interacted by the user. The method combining the identity information and the purchase history of the user is beneficial to finding out the purchase characteristics of the user better, and the product is pushed for the user under the actual condition of considering the user, so that the user satisfaction can be improved to a great extent, the transaction efficiency is improved, and the product purchase rate is increased.
2. The invention integrates various characteristics when modeling the user and the product, analyzes the user behavior and the product from different dimensions, and can reduce the problems of detail noise caused by excessive user behavior and cold start caused by insufficient user behavior, thereby obtaining more accurate user representation.
3. According to the invention, the product ID sequence, the product category sequence and the product price sequence are constructed into the directed graph, and the representation of the neural network learning node of the graph is used, so that the complex transfer relationship among the products clicked by the user can be captured, and further the transfer of the user interest can be found.
4. Since users tend to spend more time on products of interest, the present invention explicitly models the user's dwell time and number of clicks on the product to find the product the user really likes.
5. The first and last products clicked by the user tend to be closely related to the user's intent, the first product representing the user's direction of view (i.e., what product the user wants to view) and the last product representing the user's direction of purchase (i.e., what product the user will ultimately purchase). Thus, the present invention takes the representation of the first product and the last product as local representations of the user to more accurately capture the user's true intent.
Based on the same inventive concept, the embodiment of the invention also provides a product pushing system based on deep learning, and as the principle of solving the problems of the system is similar to the product pushing method based on deep learning, the implementation of the system can refer to the implementation of the method, and repeated parts are not described again.
Fig. 8 is a block diagram of a product pushing system based on deep learning in the embodiment of the present invention. As shown in fig. 8, the product push system based on deep learning includes:
the current user information matrix unit is used for constructing a current user information matrix according to the user identity information;
the current product information matrix unit is used for creating a current product information matrix according to the sequence of products clicked by users;
the product recommendation probability unit is used for inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;
and the product pushing unit is used for determining a pushed product according to the product recommendation probability and pushing the pushed product.
In one embodiment, the current product information matrix comprises a product ID matrix, a product category matrix and a product price matrix;
the current product information matrix unit is specifically configured to:
creating a product ID sequence according to the order of clicking products by a user;
determining a product category sequence and a product price sequence according to the product ID sequence;
generating a product ID matrix according to the product ID sequence and the residence time of the user on each product;
and generating a product category matrix according to the product category sequence, and generating a product price matrix according to the product price sequence.
In one embodiment, the system further comprises a product recommendation model creating unit, configured to:
the following iterative process is performed:
generating a product matrix according to the historical product information matrix and the first model parameter matrix;
determining a product prediction recommendation probability according to the product matrix, the second model parameter matrix, the third model parameter matrix and the historical user information matrix;
determining a loss function according to the product prediction recommendation probability and the actual purchased product;
and when the loss function is smaller than a preset value, establishing a product recommendation model according to the first model parameter matrix, the second model parameter matrix and the third model parameter matrix in the current iteration, otherwise, updating the first model parameter matrix, the second model parameter matrix and the third model parameter matrix according to the loss function, and continuously executing the iteration processing.
In one embodiment, the product recommendation model creating unit is specifically configured to:
generating a user initial interest characteristic matrix according to the product matrix and the second model parameter matrix;
generating a final user interest characteristic matrix according to the third model parameter matrix, the initial user interest characteristic matrix and the historical user information matrix;
and determining the product prediction recommendation probability according to the final interest characteristic matrix of the user and the product matrix.
To sum up, the product pushing system based on deep learning of the embodiment of the present invention constructs a current user information matrix according to user identity information, creates a current product information matrix according to a sequence in which users click products, inputs the current user information matrix and the current product information matrix into a product recommendation model to obtain a product recommendation probability, determines a pushed product according to the product recommendation probability and pushes the pushed product, and can push specific products for different users, thereby improving the precision and accuracy of product pushing and further improving the user satisfaction.
The embodiment of the invention also provides a specific implementation mode of computer equipment, which can realize all the steps in the product pushing method based on deep learning in the embodiment. Fig. 9 is a block diagram of a computer device in an embodiment of the present invention, and referring to fig. 9, the computer device specifically includes the following:
a processor (processor)901 and a memory (memory) 902.
The processor 901 is configured to call a computer program in the memory 902, and the processor implements all the steps in the deep learning based product pushing method in the above embodiments when executing the computer program, for example, the processor implements the following steps when executing the computer program:
constructing a current user information matrix according to the user identity information;
creating a current product information matrix according to the order of clicking products by a user;
inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;
and determining a pushed product according to the product recommendation probability and pushing the pushed product.
To sum up, the computer device of the embodiment of the present invention constructs a current user information matrix according to the user identity information, creates a current product information matrix according to the order in which the user clicks the products, then inputs the current user information matrix and the current product information matrix into a product recommendation model to obtain a product recommendation probability, and finally determines a pushed product according to the product recommendation probability and pushes the pushed product, so that specific products can be pushed by different users, the precision and accuracy of product pushing are improved, and the user satisfaction is further improved.
An embodiment of the present invention further provides a computer-readable storage medium capable of implementing all the steps in the deep learning based product pushing method in the foregoing embodiment, where the computer-readable storage medium stores a computer program, and the computer program, when executed by a processor, implements all the steps in the deep learning based product pushing method in the foregoing embodiment, for example, when the processor executes the computer program, the processor implements the following steps:
constructing a current user information matrix according to the user identity information;
creating a current product information matrix according to the order of clicking products by a user;
inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;
and determining a pushed product according to the product recommendation probability and pushing the pushed product.
To sum up, the computer-readable storage medium of the embodiment of the present invention constructs a current user information matrix according to user identity information, creates a current product information matrix according to a sequence in which users click products, inputs the current user information matrix and the current product information matrix into a product recommendation model to obtain a product recommendation probability, determines a pushed product according to the product recommendation probability and pushes the pushed product, and can push specific products for different users, thereby improving the precision and accuracy of product pushing and further improving the user satisfaction.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are only exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Those of skill in the art will further appreciate that the various illustrative logical blocks, units, and steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate the interchangeability of hardware and software, various illustrative components, elements, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design requirements of the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present embodiments.
The various illustrative logical blocks, or elements, or devices described in connection with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a digital signal processor, an Application Specific Integrated Circuit (ASIC), a field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a digital signal processor core, or any other similar configuration.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may be stored in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. For example, a storage medium may be coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC, which may be located in a user terminal. In the alternative, the processor and the storage medium may reside in different components in a user terminal.
In one or more exemplary designs, the functions described above in connection with the embodiments of the invention may be implemented in hardware, software, firmware, or any combination of the three. If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media that facilitate transfer of a computer program from one place to another. Storage media may be any available media that can be accessed by a general purpose or special purpose computer. For example, such computer-readable media can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to carry or store program code in the form of instructions or data structures and which can be read by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Additionally, any connection is properly termed a computer-readable medium, and, thus, is included if the software is transmitted from a website, server, or other remote source via a coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL), or wirelessly, e.g., infrared, radio, and microwave. Such discs (disk) and disks (disc) include compact disks, laser disks, optical disks, DVDs, floppy disks and blu-ray disks where disks usually reproduce data magnetically, while disks usually reproduce data optically with lasers. Combinations of the above may also be included in the computer-readable medium.
Claims (10)
1. A product pushing method based on deep learning is characterized by comprising the following steps:
constructing a current user information matrix according to the user identity information;
creating a current product information matrix according to the order of clicking products by a user;
inputting the current user information matrix and the current product information matrix into a product recommendation model established based on a historical user information matrix and a historical product information matrix to obtain a product recommendation probability;
and determining a pushed product according to the product recommendation probability and pushing the pushed product.
2. The deep learning based product pushing method according to claim 1, wherein the current product information matrix comprises a product ID matrix, a product category matrix and a product price matrix;
creating a current product information matrix according to the order in which users click products includes:
creating a product ID sequence according to the order of clicking products by a user;
determining a product category sequence and a product price sequence according to the product ID sequence;
generating a product ID matrix according to the product ID sequence and the residence time of the user on each product;
and generating a product category matrix according to the product category sequence, and generating a product price matrix according to the product price sequence.
3. The deep learning based product pushing method according to claim 1, wherein the step of creating a product recommendation model based on the historical user information matrix and the historical product information matrix comprises:
the following iterative process is performed:
generating a product matrix according to the historical product information matrix and the first model parameter matrix;
determining a product prediction recommendation probability according to the product matrix, the second model parameter matrix, the third model parameter matrix and the historical user information matrix;
determining a loss function according to the product prediction recommendation probability and the actually purchased product;
and when the loss function is smaller than a preset value, establishing a product recommendation model according to a first model parameter matrix, a second model parameter matrix and a third model parameter matrix in the current iteration, otherwise, updating the first model parameter matrix, the second model parameter matrix and the third model parameter matrix according to the loss function, and continuously executing the iteration processing.
4. The deep learning based product pushing method of claim 3, wherein determining a product prediction recommendation probability according to the product matrix, the second model parameter matrix, the third model parameter matrix and the historical user information matrix comprises:
generating a user initial interest characteristic matrix according to the product matrix and the second model parameter matrix;
generating a final user interest characteristic matrix according to the third model parameter matrix, the initial user interest characteristic matrix and the historical user information matrix;
and determining the product prediction recommendation probability according to the final interest feature matrix of the user and the product matrix.
5. A product pushing system based on deep learning is characterized by comprising:
the current user information matrix unit is used for constructing a current user information matrix according to the user identity information;
the current product information matrix unit is used for creating a current product information matrix according to the sequence of products clicked by users;
the product recommendation probability unit is used for inputting the current user information matrix and the current product information matrix into a product recommendation model established based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;
and the product pushing unit is used for determining a pushed product according to the product recommendation probability and pushing the pushed product.
6. The deep learning based product push system of claim 5, wherein the current product information matrix comprises a product ID matrix, a product category matrix, and a product price matrix;
the current product information matrix unit is specifically configured to:
creating a product ID sequence according to the order of clicking products by a user;
determining a product category sequence and a product price sequence according to the product ID sequence;
generating a product ID matrix according to the product ID sequence and the residence time of the user on each product;
and generating a product category matrix according to the product category sequence, and generating a product price matrix according to the product price sequence.
7. The deep learning based product push system according to claim 5, further comprising a product recommendation model creation unit configured to:
the following iterative process is performed:
generating a product matrix according to the historical product information matrix and the first model parameter matrix;
determining a product prediction recommendation probability according to the product matrix, the second model parameter matrix, the third model parameter matrix and the historical user information matrix;
determining a loss function according to the product prediction recommendation probability and the actually purchased product;
and when the loss function is smaller than a preset value, establishing a product recommendation model according to a first model parameter matrix, a second model parameter matrix and a third model parameter matrix in the current iteration, otherwise, updating the first model parameter matrix, the second model parameter matrix and the third model parameter matrix according to the loss function, and continuously executing the iteration processing.
8. The deep learning based product pushing system according to claim 7, wherein the product recommendation model creation unit is specifically configured to:
generating a user initial interest characteristic matrix according to the product matrix and the second model parameter matrix;
generating a final user interest characteristic matrix according to the third model parameter matrix, the initial user interest characteristic matrix and the historical user information matrix;
and determining the product prediction recommendation probability according to the final interest feature matrix of the user and the product matrix.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor implements the steps of the deep learning based product push method according to any one of claims 1 to 4 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored, wherein the computer program, when being executed by a processor, implements the steps of the deep learning based product push method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011508268.8A CN112612955A (en) | 2020-12-18 | 2020-12-18 | Product pushing method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011508268.8A CN112612955A (en) | 2020-12-18 | 2020-12-18 | Product pushing method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112612955A true CN112612955A (en) | 2021-04-06 |
Family
ID=75240774
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011508268.8A Pending CN112612955A (en) | 2020-12-18 | 2020-12-18 | Product pushing method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112612955A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113313562A (en) * | 2021-07-29 | 2021-08-27 | 太平金融科技服务(上海)有限公司深圳分公司 | Product data processing method and device, computer equipment and storage medium |
-
2020
- 2020-12-18 CN CN202011508268.8A patent/CN112612955A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113313562A (en) * | 2021-07-29 | 2021-08-27 | 太平金融科技服务(上海)有限公司深圳分公司 | Product data processing method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10958748B2 (en) | Resource push method and apparatus | |
| CN110278175B (en) | Graph structure model training and garbage account identification method, device and equipment | |
| CN107491432B (en) | Low-quality article identification method and device based on artificial intelligence, equipment and medium | |
| US11521221B2 (en) | Predictive modeling with entity representations computed from neural network models simultaneously trained on multiple tasks | |
| US20150269495A1 (en) | Methods and system for providing simultaneous multi-task ensemble learning | |
| CN110889463A (en) | Sample labeling method and device, server and machine-readable storage medium | |
| CN111966914A (en) | Content recommendation method and device based on artificial intelligence and computer equipment | |
| CN111382283A (en) | Resource category label labeling method and device, computer equipment and storage medium | |
| CN109165691A (en) | Training method, device and the electronic equipment of the model of cheating user for identification | |
| CN113343091A (en) | Industrial and enterprise oriented science and technology service recommendation calculation method, medium and program | |
| CN115358809A (en) | Multi-intention recommendation method and device based on graph comparison learning | |
| CN112612955A (en) | Product pushing method and system based on deep learning | |
| CN109308332B (en) | Target user acquisition method and device and server | |
| CN111680213A (en) | Information recommendation method, data processing method and device | |
| US20230316106A1 (en) | Method and apparatus for training content recommendation model, device, and storage medium | |
| CN112818234A (en) | Network public opinion information analysis processing method and system | |
| CN115794898B (en) | Financial information recommendation method and device, electronic equipment and storage medium | |
| CN116501979A (en) | Information recommendation method, information recommendation device, computer equipment and computer readable storage medium | |
| CN111861545B (en) | User behavior portrait construction method, device, equipment and storage medium | |
| CN113553501A (en) | Method and device for user portrait prediction based on artificial intelligence | |
| CN113837807A (en) | Heat prediction method and device, electronic equipment and readable storage medium | |
| CN115114462A (en) | Model training method and device, multimedia recommendation method and device and storage medium | |
| CN113076471A (en) | Information processing method and device and computing equipment | |
| CN113627513A (en) | Training data generation method and system, electronic device and storage medium | |
| CN113837836A (en) | Model recommendation method, device, equipment and storage medium |
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 |