Product push method and system based on deep learning

Abstract

The present invention provides a product push method and system based on deep learning, belonging to the field of artificial intelligence technology. The product push method based on deep learning includes: constructing a current user information matrix according to user identity information; creating a current product information matrix according to the order in which users click on products; inputting the current user information matrix and the current product information matrix into a product recommendation model created based on a historical user information matrix and a historical product information matrix to obtain a product recommendation probability; determining a push product according to the product recommendation probability and pushing the push product. The present invention can push specific products to different users, improve the precision and accuracy of product push, and further improve user satisfaction.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a product push method and system based on deep learning. Background Art

[0002] With the rapid development of technologies such as cloud computing, big data, and the Internet of Things, the emergence of various applications in the Internet space has triggered an explosive growth in the scale of data. While big data has brought revolutionary development to human society, it has also brought about a serious "information overload" problem. Product push is an effective way to solve this problem. It can analyze user preferences and interests based on user historical behaviors to predict what users may be interested in next, and then push corresponding products to users. However, most of the existing push methods only analyze from the user's perspective, and do not consider the impact of the product itself on user preferences, which greatly affects the precision and accuracy of product push and reduces user satisfaction. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide a product push method and system based on deep learning, so as to push specific products to different users, improve the precision and accuracy of product push, and further improve user satisfaction.

[0004] In order to achieve the above object, an embodiment of the present invention provides a product push method based on deep learning, comprising:

[0005] Construct the current user information matrix based on the user identity information;

[0006] Create a current product information matrix based on the order in which users click on products;

[0007] Input the current user information matrix and the current product information matrix into the product recommendation model created based on the historical user information matrix and the historical product information matrix to obtain the product recommendation probability;

[0008] Determine the push products based on the product recommendation probability and push the pushed products.

[0009] The embodiment of the present invention also provides a product push system based on deep learning, including:

[0010] A current user information matrix unit is used to construct a current user information matrix according to user identity information;

[0011] The current product information matrix unit is used to create the current product information matrix according to the order in which users click on the products;

[0012] A product recommendation probability unit is used to input the current user information matrix and the current product information matrix into a product recommendation model created based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;

[0013] The product push unit is used to determine the push products according to the product recommendation probability and push the push products.

[0014] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the product push method based on deep learning are implemented.

[0015] An embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the product push method based on deep learning are implemented.

[0016] The deep learning-based product push method and system of the embodiment of the present invention first constructs a current user information matrix based on user identity information, creates a current product information matrix based on the order in which users click on products, and then inputs the current user information matrix and the current product information matrix into a product recommendation model to obtain product recommendation probabilities. Finally, the push products are determined based on the product recommendation probabilities and the push products are pushed. Specific products can be pushed to different users, thereby improving the precision and accuracy of product push and further improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 is a flow chart of a product push method based on deep learning in an embodiment of the present invention;

[0019] Figure 2 is a flowchart of a product push method based on deep learning in another embodiment of the present invention;

[0020] Figure 3 is a flow chart of S102;

[0021] Figure 4 is a flow chart of creating a product recommendation model in an embodiment of the present invention;

[0022] Figure 5 is a flow chart of S302 in an embodiment of the present invention;

[0023] Figure 6 is a schematic diagram of constructing a directed graph in an embodiment of the present invention;

[0024] Figure 7 is a schematic diagram of constructing a product information matrix in an embodiment of the present invention;

[0025] Figure 8 is a structural block diagram of a product push system based on deep learning in an embodiment of the present invention;

[0026] Fig. 9 It is a structural block diagram of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, device, apparatus, method or computer program product. Therefore, the present disclosure may be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0029] In view of the fact that the prior art does not consider the influence of the product itself on user preferences, which greatly affects the precision and accuracy of product push and reduces user satisfaction, the embodiment of the present invention provides a product push method based on deep learning, which overcomes the defects of single-angle modeling, insufficient feature utilization and incomplete behavior sequence analysis in the existing recommendation algorithm. The present invention has the following advantages: First, modeling is performed from the perspectives of users and products respectively, and the user's own long-term preferences and the user's short-term purchasing interests can be considered at the same time, so as to more accurately capture the user's current interests. Second, by using multiple features to analyze users and products respectively, a more accurate representation of users and products can be obtained. Third, the time interval information and the number of clicks in the user's click behavior are explicitly modeled, and the user's preference for the product is determined by the user's stay time on a certain product. In summary, the purpose of the present invention is to use a deep learning method to integrate multiple features, and model the user's behavior records from the two perspectives of users and products respectively, so as to improve the accuracy of product push, thereby improving user satisfaction and increasing product purchase rate. The present invention is described in detail below in conjunction with the accompanying drawings.

[0030] Figure 1It is a flowchart of a product push method based on deep learning in an embodiment of the present invention. Figure 2 FIG. 1 is a flow chart of a product push method based on deep learning in another embodiment of the present invention. Figure 1-Figure 2 As shown in the figure, the product push method based on deep learning includes:

[0031] S101: construct a current user information matrix according to user identity information.

[0032] During specific implementation, the current user information matrix may be constructed by referring to the method of constructing the historical user information matrix.

[0033] The steps to construct the historical user information matrix are as follows:

[0034] a. Get the user representation vector u according to the user's ID number i1 First, we get the user set U = [u1,u2,…,u K ], where K is the number of users. Use one-hot vectors to get the vector representation of each user, and then represent it in the d-dimensional representation space to get u i1 .

[0035] b. Vectorize the user's identity (student, office worker, farmer, etc.) to get the vector u i2 ;

[0036] c. Segment the user's income and represent it in vector form. For example, if the income range of all users is [2000,5000], it can be segmented into [2000,3000), [3000,4000) and [4000,5000]. These three segments are numbered 1, 2 and 3 respectively. Then, users can be marked as 1, 2 or 3 according to their income situation, and the representation vector u of the user's income situation is obtained. i3 .

[0037] d. Fuse the three vectors obtained in steps a, b, and c to obtain the user's own long-term preference (user information matrix), that is, the historical user information matrix u i =[u i1 ;u i2 ;u i3 ], where [;] represents vector concatenation.

[0038] S102: Create a current product information matrix according to the order in which users click on products.

[0039] The current product information matrix can be constructed by referring to the method of constructing the historical product information matrix. Both the current product information matrix and the historical product information matrix include a product ID matrix, a product category matrix and a product price matrix.

[0040] Figure 3 This is a flow chart of S102. Figure 3 As shown, S102 includes:

[0041] S201: Create a product ID sequence according to the order in which users click on products.

[0042] In the specific implementation, the products are mapped to a unified representation space according to the product ID number. That is, firstly, all the non-repeated products that the users have interacted with are extracted from the interaction data between the users and the products, and the product ID sequence X = [x1, x2, ..., x n ], where n is the number of unique products. Use one-hot vectors to get the representation of each product, and then represent it in a d-dimensional representation space to get the representation v of the product ID i .

[0043] S202: Determine a product category sequence and a product price sequence according to the product ID sequence.

[0044] In specific implementation, the product categories can be mapped to a unified representation space according to the category ID. That is, firstly, non-repeated product categories are extracted from the product data set to obtain the product category sequence C = [c1, c2, ..., c m ], where m represents the number of non-repeated categories, and m≤n. Use one-hot vectors to get the representation of each category, and then represent it in a d-dimensional representation space to get the representation m of the product category. i .

[0045] Similarly, the product price is processed in the same way as the user income, that is, the product price is mapped to the corresponding interval to obtain the product price sequence R i =[r1,r2,......r p ], where p represents the number of price ranges, and p≤n. Use one-hot vectors to get the representation of each price range, and then represent it in a d-dimensional representation space to get the representation p of the product price i .

[0046] S203: Generate a product ID matrix according to the product ID sequence and the user's stay time on each product.

[0047] Figure 6 Schematic diagram of the construction of a directed graph in an embodiment of the present invention. Figure 6 As shown in Figure 1, the historical interaction behavior sequence of the user (current user or historical user) can be constructed according to the time when the click event occurs. Figure 6 The directed graph in . Figure 6 The user1, user2, and user3 represent different users.

[0048] like Figure 6 As shown, according to the product ID sequence X = [x1, x2, ..., x n ]Build a directed graph of product ID sequences. The i-th node x in the directed graph i represents the product clicked by the user i, each edge (x i-1 ,x i ) means the user clicked x first i-1 Click the x again i .

[0049] Figure 7 FIG. 1 is a schematic diagram of constructing a product information matrix in an embodiment of the present invention. Figure 7 As shown, Figure 6 For example, user 1 in the example, considering that the time a user stays on a product reflects the user's preference for each product, that is, the longer the stay time, the more interested the user is in the product. Therefore, when constructing a directed graph of product ID sequences, each edge in the directed graph can be weighted according to the user's stay time on each product (i.e., time interval). The weight calculation formula is as follows:

[0050]

[0051] Among them, χ 1 is the weight of the edge in the directed graph of product ID sequence. Figure 7 As shown in the figure, the product ID out-edge matrix (Out-edge Matrix) can be constructed based on the weight of each edge in the product ID sequence directed graph. and product ID in-edge matrix Then and Splicing to generate product ID matrix A x .

[0052] For example, Figure 7 In the example, x1 points to x2, x2 points to x3, and x3 points to x4. The time a user spends on products x1, x2, and x3 is 20s, 40s, and 60s, respectively. Therefore, the weight of the x1x2 edge is 1 / 6, the weight of the x2x3 edge is 1 / 3, and the weight of the x3x1 edge is 1 / 2. The product ID out of the edge matrix The element in the first row and second column is 1 / 6, the element in the second row and third column is 1 / 3, the element in the third row and fourth column is 1 / 2, and the other elements in the product ID outgoing edge matrix are 0; the product ID incoming edge matrix The element in the second row and first column is 1 / 6, the element in the third row and second column is 1 / 3, the element in the fourth row and third column is 1 / 2, and the other elements in the product ID input edge matrix are 0.

[0053] S204: Generate a product category matrix according to the product category sequence, and generate a product price matrix according to the product price sequence.

[0054] like Figure 6 As shown, according to the product category sequence C = [c1, c2, ..., c m ] Construct a directed graph of product category sequences. That is, the category of product x1 is c1, and product x n The category is c m The product category sequence is constructed as a directed graph of product category sequence, where each node represents a category and each edge (c i-1 ,c i ) represents that the user clicked on the category c i-1 After that, I clicked on the product of category c i products.

[0055] like Figure 6 As shown, according to the product price sequence R i =[r1,r2,…r p ] Construct a directed graph of product price series. That is, the price range of product x1 is r1, and product x n The price range is r p The product price sequence is constructed as a directed graph of product price sequence, where each node represents a price range and each edge (r i-1 ,r i ) represents the user clicked the price r i-1 After clicking on the product with price r i products.

[0056] like Figure 7 As shown, 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, as well as the number of occurrences 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 by the following formula:

[0057]

[0058] Among them, χ 2 is the weight of the edge in the directed graph of product category sequence or product price sequence.

[0059] like Figure 7 As shown in the figure, the 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 category in-edge matrix Then and Splicing to generate product category matrix A c When the denominator in the formula is the out-degree of the node, the corresponding weight is the element in the product category out-edge matrix; when the denominator in the formula is the in-degree of the node, the corresponding weight is the element in the product category in-edge matrix. For example, Figure 7 c1 points to c2, c2 points to c3, and c3 points to c1, so the product category out-edge matrix The elements in the first row and second column, the elements in the second row and third column, and the elements in the third row and first column are all 1, and the other elements in the product category out-edge matrix are 0; the product category in-edge matrix (In-edge Matrix) The elements in the first row and third column, the elements in the second row and first column, and the elements in the third row and second column are all 1, and the other elements in the product category input edge matrix are 0.

[0060] like Figure 7 As shown in the figure, the product price out-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 Then and Splicing to generate product price matrix A r When the denominator in the formula is the out-degree of the node, the corresponding weight is the element in the product price out-edge matrix; when the denominator in the formula is the in-degree of the node, the corresponding weight is the element in the product price in-edge matrix. For example, Figure 7 In the matrix, r1 points to r2, r2 points to r1, and r1 points to r3, so the product price out-edge matrix The elements in the first row and second column and the elements in the first row and third column are both 1 / 2, the elements in the second row and first column are 1, and the other elements in the product price out-edge matrix are 0; the product category in-edge matrix (In-edge Matrix) The elements in the first row and second column, the elements in the second row and first column, and the elements in the third row and first column are all 1, and the other elements in the product category input edge matrix are 0.

[0061] S103: Input the current user information matrix and the current product information matrix into a product recommendation model created based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability.

[0062] In specific implementation, the product recommendation model outputs the product recommendation probability corresponding to each current product.

[0063] S104: Determine a push product based on the product recommendation probability and push the push product.

[0064] During specific implementation, the current product corresponding to the maximum value of the product recommendation probability may be determined as the push product, and the push product may be pushed to the current user.

[0065] Figure 1 The execution subject of the product push method based on deep learning shown in the figure can be a computer. Figure 1 As can be seen from the process shown, the product push method based on deep learning in the embodiment of the present invention first constructs a current user information matrix based on user identity information, creates a current product information matrix based on the order in which users click on products, and then inputs the current user information matrix and the current product information matrix into the product recommendation model to obtain the product recommendation probability. Finally, the push product is determined based on the product recommendation probability and the push product is pushed. Specific products can be pushed to different users, thereby improving the precision and accuracy of product push and further improving user satisfaction.

[0066] Figure 4 is a flow chart of creating a product recommendation model in an embodiment of the present invention. Figure 4 As shown, the steps of creating a product recommendation model based on the historical user information matrix and the historical product information matrix include:

[0067] Perform the following iterative processing:

[0068] S301: Generate a product matrix according to the historical product information matrix and the first model parameter matrix.

[0069] The first model parameter matrix includes the first parameter matrix H and the second parameter matrix W z , the third parameter matrix W r , the fourth parameter matrix W o , the fifth parameter matrix U z , the sixth parameter matrix U r And the seventh parameter matrix U o .

[0070] In the specific implementation, suppose the historical product information matrix includes the product ID matrix A x , Product Category Matrix A c and product price matrix A r As mentioned above, the representation of product ID can be obtained by using graph neural network to learn node representation for product ID sequence, product category sequence and product price sequence respectively. i , Product category representation i and product price representation i . Use gated graph neural network to learn node representation, with product ID representation v i For example, the formula for node update is as follows:

[0071]

[0072]

[0073]

[0074]

[0075]

[0076] in, is the representation of the i-th node at time t, A is the product ID matrix x In the i-th row of , when the nodes represent product categories or product prices, can be replaced by the product category matrix A c The i-th row Or product price matrix A r The i-th row represents the product ID sequence clicked by the user at time t-1, b is the bias vector, σ() is the Sigmoid function, is the representation of the user clicking on the i-th product at time t-1, In other words, the representation of the user clicking on the i-th product at time t. Similarly, the representation of the user clicking on the i-th category product at time t can be obtained. and the representation of the price range of the i-th product clicked by the user at time t

[0077] The representation of the product learned by the graph neural network and Perform concat fusion to obtain the product matrix q of the i-th product i ,Right now

[0078] S302: Determine 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.

[0079] Figure 5 is a flow chart of S302 in an embodiment of the present invention. Figure 5 As shown, S302 includes:

[0080] S401: Generate a user initial interest feature matrix based on the product matrix and the second model parameter matrix.

[0081] Among them, the second model parameter matrix includes an eighth parameter matrix w, a ninth parameter matrix W2, a tenth parameter matrix W3 and an eleventh parameter matrix W4.

[0082] In the specific implementation, we first combine the attention mechanism to obtain the user's global preference sglobal ; Then the product matrix q1 of the first product clicked by the user and the product matrix q of the last product clicked by the user are n As the user's local preference S local , the formula is as follows:

[0083] α i =w T σ(W2q n +W3Q i +c);

[0084]

[0085] s local =q1+q n ;

[0086] Where c is a constant.

[0087] Finally, the user's global preference s global and the user's local preference S local Perform concat fusion to obtain the user's initial interest feature matrix (user purchase interest representation) s p ,as follows:

[0088] s p =W4[s global ;s local ].

[0089] S402: Generate a user final interest feature matrix according to the third model parameter matrix, the user initial interest feature matrix and the historical user information matrix.

[0090] The third model parameter matrix includes a twelfth parameter matrix W1 and a thirteenth parameter matrix W5.

[0091] In specific implementation, the user's long-term preference and the user's purchase interest expression can be integrated through the following formula to obtain the user's final interest feature matrix u: fi :

[0092] u' i =W1u i =W1[u i1 ;u i2 ;u i3 ];

[0093] u fi =W5[u' i ;s p ].

[0094] S403: Determine the product prediction recommendation probability according to the user's final interest feature matrix and the product matrix.

[0095] In specific implementation, the product prediction recommendation probability can be determined by the following formula:

[0096]

[0097] in, is the predicted recommendation probability of the i-th product.

[0098] S303: Determine a loss function based on the predicted recommendation probability of the product and the actual purchased product.

[0099] In specific implementation, the loss function can be determined by the following formula:

[0100]

[0101] in, is the loss function, y i is the one-hot vector of the actual purchased product, λ is the first preset parameter, and θ is the second preset parameter.

[0102] S304: Determine whether the loss function is less than a preset value.

[0103] S305: When the loss function is less than a preset value, a product recommendation model is created according to the first model parameter matrix, the second model parameter matrix, and the third model parameter matrix in the current iteration.

[0104] S306: When the loss function is greater than or equal to a preset value, the first model parameter matrix, the second model parameter matrix and the third model parameter matrix are updated according to the loss function, and the iterative process continues.

[0105] In a specific implementation, when the loss function is greater than or equal to a preset value, the back propagation through time (BPTT) algorithm can be used to update the first model parameter matrix, the second model parameter matrix, and the third model parameter matrix to train the product recommendation model.

[0106] The specific process of the embodiment of the present invention is as follows:

[0107] 1. Generate a product matrix based on the historical product information matrix and the first model parameter matrix.

[0108] 2. Generate the user initial interest feature matrix based on the product matrix and the second model parameter matrix.

[0109] 3. Generate the user's final interest feature matrix based on the third model parameter matrix, the user's initial interest feature matrix and the historical user information matrix.

[0110] 4. Determine the product prediction recommendation probability based on the user's final interest feature matrix and product matrix.

[0111] 5. Determine the loss function based on the predicted recommendation probability of the product and the actual purchase of the product.

[0112] 6. When the loss function is less than the preset value, create a product recommendation model based on the first model parameter matrix, the second model parameter matrix and the third model parameter matrix in the current iteration; otherwise, update the first model parameter matrix, the second model parameter matrix and the third model parameter matrix according to the loss function and return to step 1.

[0113] 7. Construct the current user information matrix based on the user identity information.

[0114] 8. Create a product ID sequence based on the order in which users click on products, and determine the product category sequence and product price sequence based on the product ID sequence.

[0115] 9. Generate a product ID matrix based on the product ID sequence and the user’s stay time on each product.

[0116] 10. Generate a product category matrix based on the product category sequence, and generate a product price matrix based on the product price sequence.

[0117] 11. Input the current user information matrix, product ID matrix, product category matrix and product price matrix into the product recommendation model to obtain the product recommendation probability.

[0118] 12. Determine the current product corresponding to the maximum value of the product recommendation probability as the push product, and push the push product to the current user.

[0119] In summary, the present invention models the user's identity and the user's product click sequence to explore the user's interests, predict the product that the user may buy next, and then push products to the user, which has the following beneficial effects:

[0120] 1. The present invention models from the user perspective and the product perspective respectively, discovers the user's long-term preferences by modeling the user's own characteristics, and explores the user's short-term purchasing interests by modeling the products that the user has interacted with. This method of combining the user's identity information and purchase history helps to better discover the user's purchase characteristics, and pushes products to the user in consideration of the user's actual situation, which can greatly improve user satisfaction, improve transaction efficiency, and thus increase product purchase rates.

[0121] 2. The present invention incorporates multiple features when modeling users and products, and analyzes user behaviors and products from different dimensions, which can reduce the detail noise caused by too much user behavior and the cold start problem caused by too little user behavior, thereby obtaining a more accurate user representation.

[0122] 3. The present invention constructs the product ID sequence, product category sequence and product price sequence into a directed graph, and uses a graph neural network to learn the representation of nodes, so as to capture the complex transfer relationship between the products clicked by the user, and then discover the transfer of user interests.

[0123] 4. Since users tend to spend more time on products they are interested in, the present invention explicitly models the user's dwell time and click times on the product to discover the products that the user really likes.

[0124] 5. The first product and the last product clicked by the user are often closely related to the user's intention. The first product represents the user's browsing direction (i.e., what kind of products the user wants to view), while the last product represents the user's purchase direction (i.e., what kind of products the user will eventually buy). Therefore, the present invention uses the representation of the first product and the last product as a local representation of the user to more accurately capture the user's true intention.

[0125] Based on the same inventive concept, an embodiment of the present invention also provides a product push system based on deep learning. Since the principle of solving the problem by this system is similar to that of the product push method based on deep learning, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be repeated.

[0126] Figure 8 is a structural block diagram of a product push system based on deep learning in an embodiment of the present invention. Figure 8 As shown in the figure, the product push system based on deep learning includes:

[0127] A current user information matrix unit is used to construct a current user information matrix according to user identity information;

[0128] The current product information matrix unit is used to create the current product information matrix according to the order in which users click on the products;

[0129] A product recommendation probability unit is used to input the current user information matrix and the current product information matrix into a product recommendation model created based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability;

[0130] The product push unit is used to determine the push products according to the product recommendation probability and push the push products.

[0131] In one embodiment, the current product information matrix includes a product ID matrix, a product category matrix, and a product price matrix;

[0132] The current product information matrix unit is specifically used for:

[0133] Create a sequence of product IDs based on the order in which users click on products;

[0134] Determine product category sequence and product price sequence according to product ID sequence;

[0135] Generate a product ID matrix based on the product ID sequence and the user's stay time on each product;

[0136] A product category matrix is ​​generated according to the product category sequence, and a product price matrix is ​​generated according to the product price sequence.

[0137] In one embodiment, a product recommendation model creation unit is further included, which is used to:

[0138] Perform the following iterative processing:

[0139] Generate a product matrix according to the historical product information matrix and the first model parameter matrix;

[0140] Determine 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;

[0141] Determine the loss function based on the predicted recommendation probability of the product and the actual purchase product;

[0142] When the loss function is less than a preset value, a product recommendation model is created based on the first model parameter matrix, the second model parameter matrix and the third model parameter matrix in the current iteration; otherwise, the first model parameter matrix, the second model parameter matrix and the third model parameter matrix are updated according to the loss function, and the iterative processing continues.

[0143] In one embodiment, the product recommendation model creation unit is specifically used to:

[0144] Generate a user initial interest feature matrix based on the product matrix and the second model parameter matrix;

[0145] Generate a user final interest feature matrix according to the third model parameter matrix, the user initial interest feature matrix and the historical user information matrix;

[0146] Determine the product prediction recommendation probability based on the user's final interest feature matrix and product matrix.

[0147] In summary, the deep learning-based product push system of the embodiment of the present invention first constructs a current user information matrix based on user identity information, creates a current product information matrix based on the order in which users click on products, and then inputs the current user information matrix and the current product information matrix into the product recommendation model to obtain the product recommendation probability. Finally, the push product is determined based on the product recommendation probability and the push product is pushed. Specific products can be pushed to different users, thereby improving the precision and accuracy of product push and further improving user satisfaction.

[0148] An embodiment of the present invention also provides a specific implementation of a computer device that can implement all steps in the deep learning-based product push method in the above embodiment. Fig. 9 is a block diagram of a computer device in an embodiment of the present invention, see Fig. 9 , the computer equipment specifically includes the following contents:

[0149] Processor (processor) 901 and memory (memory) 902 .

[0150] The processor 901 is used to call the computer program in the memory 902. When the processor executes the computer program, all steps in the product push method based on deep learning in the above embodiment are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0151] Construct the current user information matrix based on the user identity information;

[0152] Create a current product information matrix based on the order in which users click on products;

[0153] Input the current user information matrix and the current product information matrix into the product recommendation model created based on the historical user information matrix and the historical product information matrix to obtain the product recommendation probability;

[0154] Determine the push products based on the product recommendation probability and push the pushed products.

[0155] In summary, the computer device of the embodiment of the present invention first 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 on the products, and then inputs the current user information matrix and the current product information matrix into the product recommendation model to obtain the product recommendation probability. Finally, the push product is determined according to the product recommendation probability and the push product is pushed. Specific products can be pushed to different users, thereby improving the precision and accuracy of product push and further improving user satisfaction.

[0156] The embodiment of the present invention also provides a computer-readable storage medium capable of implementing all the steps in the product push method based on deep learning in the above embodiment. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, all the steps of the product push method based on deep learning in the above embodiment are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0157] Construct the current user information matrix based on the user identity information;

[0158] Create a current product information matrix based on the order in which users click on products;

[0159] Input the current user information matrix and the current product information matrix into the product recommendation model created based on the historical user information matrix and the historical product information matrix to obtain the product recommendation probability;

[0160] Determine the push products based on the product recommendation probability and push the pushed products.

[0161] In summary, the computer-readable storage medium of the embodiment of the present invention first constructs a current user information matrix based on the user identity information, creates a current product information matrix based on the order in which users click on products, and then inputs the current user information matrix and the current product information matrix into the product recommendation model to obtain the product recommendation probability. Finally, the push product is determined based on the product recommendation probability and the push product is pushed. Specific products can be pushed to different users, thereby improving the precision and accuracy of product push and further improving user satisfaction.

[0162] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0163] Those skilled in the art may also understand that the various illustrative logical blocks, units, and steps listed in the embodiments of the present invention may be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly demonstrate the interchangeability of hardware and software, the various illustrative components, units, and steps described above have generally described their functions. Whether such functions are implemented by hardware or software depends on the specific application and the design requirements of the entire system. Those skilled in the art may use various methods to implement the described functions for each specific application, but such implementation should not be understood as exceeding the scope of protection of the embodiments of the present invention.

[0164] The various illustrative logic blocks, or units, or devices described in the embodiments of the present invention can be implemented or operated by a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of the above. The general-purpose processor can be a microprocessor, and optionally, the general-purpose processor can also be any conventional processor, controller, microcontroller or state machine. The processor can also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration.

[0165] The steps of the method or algorithm described in the embodiments of the present invention can be directly embedded in hardware, a software module executed by a processor, or a combination of the two. The software module can be stored in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or other storage media of any form in the art. Exemplarily, the storage medium can be connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Optionally, the storage medium can also be integrated into the processor. The processor and the storage medium can be arranged in an ASIC, and the ASIC can be arranged in a user terminal. Optionally, the processor and the storage medium can also be arranged in different components in the user terminal.

[0166] In one or more exemplary designs, the above functions described in the embodiments of the present invention can be implemented in hardware, software, firmware or any combination of the three. If implemented in software, these functions can be stored on a computer-readable medium, or transmitted in the form of one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media that facilitate the transfer of computer programs from one place to another. The storage medium can be any available medium that can be accessed by any general or special computer. For example, such computer-readable media can include but are not limited to RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store program codes in the form of instructions or data structures and other forms that can be read by general or special computers, or general or special processors. In addition, any connection can be appropriately defined as a computer-readable medium, for example, if the software is transmitted from a website site, server or other remote resource through a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) or wirelessly, such as infrared, wireless and microwave, it is also included in the defined computer-readable medium. The disk and disc include compact disk, laser disk, optical disk, DVD, floppy disk and blue-ray disk. Disks usually copy data magnetically, while discs usually copy data optically with lasers. The above combination can also be included in computer readable media.

Claims

1. A product push method based on deep learning, characterized in that: include: Construct the current user information matrix based on the user identity information; Create a current product information matrix based on the order in which users click on products; Inputting the current user information matrix and the current product information matrix into a product recommendation model created based on the historical user information matrix and the historical product information matrix to obtain a product recommendation probability; Determine a push product according to the product recommendation probability and push the push product; The current product information matrix includes a product ID matrix, a product category matrix and a product price matrix; Creating a current product information matrix based on the order in which users click on products includes: Create a sequence of product IDs based on the order in which users click on products; Determine a product category sequence and a product price sequence according to the product ID sequence; Generate a product ID matrix based on the product ID sequence and the user's stay time on each product; Generate a product category matrix according to the product category sequence, and generate a product price matrix according to the product price sequence; The user's historical interaction behavior sequence is constructed into a directed graph according to the time when the click event occurs. The edges of the directed graph represent the order in which the user clicks on the product.

2. The product push method based on deep learning according to claim 1, characterized in that: The steps of creating a product recommendation model based on the historical user information matrix and the historical product information matrix include: Perform the following iterative processing: Generate a product matrix according to the historical product information matrix and the first model parameter matrix; Determine 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; Determine a loss function based on the predicted recommendation probability of the product and the actual purchased product; When the loss function is less than a preset value, a product recommendation model is created based on the first model parameter matrix, the second model parameter matrix and the third model parameter matrix in the current iteration; otherwise, the first model parameter matrix, the second model parameter matrix and the third model parameter matrix are updated according to the loss function, and the iterative processing continues.

3. The product push method based on deep learning according to claim 2, characterized in that: 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 includes: Generate a user initial interest feature matrix according to the product matrix and the second model parameter matrix; Generate a user final interest feature matrix according to the third model parameter matrix, the user initial interest feature matrix and the historical user information matrix; The product prediction recommendation probability is determined according to the user final interest feature matrix and the product matrix.

4. A product push system based on deep learning, characterized in that: include: A current user information matrix unit is used to construct a current user information matrix according to user identity information; The current product information matrix unit is used to create the current product information matrix according to the order in which users click on the products; A product recommendation probability unit, used to input the current user information matrix and the current product information matrix into a product recommendation model created based on a historical user information matrix and a historical product information matrix to obtain a product recommendation probability; A product push unit, configured to determine a push product according to the product recommendation probability and push the push product; The current product information matrix includes a product ID matrix, a product category matrix and a product price matrix; The current product information matrix unit is specifically used for: Create a sequence of product IDs based on the order in which users click on products; Determine a product category sequence and a product price sequence according to the product ID sequence; Generate a product ID matrix based on the product ID sequence and the user's stay time on each product; Generate a product category matrix according to the product category sequence, and generate a product price matrix according to the product price sequence; The user's historical interaction behavior sequence is constructed into a directed graph according to the time when the click event occurs. The edges of the directed graph represent the order in which the user clicks on the product.

5. The product push system based on deep learning according to claim 4, characterized in that: It also includes a product recommendation model creation unit for: Perform the following iterative processing: Generate a product matrix according to the historical product information matrix and the first model parameter matrix; Determine 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; Determine a loss function based on the predicted recommendation probability of the product and the actual purchased product; When the loss function is less than a preset value, a product recommendation model is created based on the first model parameter matrix, the second model parameter matrix and the third model parameter matrix in the current iteration; otherwise, the first model parameter matrix, the second model parameter matrix and the third model parameter matrix are updated according to the loss function, and the iterative processing continues.

6. The product push system based on deep learning according to claim 5, characterized in that: The product recommendation model creation unit is specifically used for: Generate a user initial interest feature matrix according to the product matrix and the second model parameter matrix; Generate a user final interest feature matrix according to the third model parameter matrix, the user initial interest feature matrix and the historical user information matrix; The product prediction recommendation probability is determined according to the user final interest feature matrix and the product matrix.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the deep learning-based product push method described in any one of claims 1 to 3 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the deep learning-based product push method described in any one of claims 1 to 3 are implemented.

Images