Product recommendation method and device and related equipment

Abstract

The invention provides a product recommendation method and device and related equipment, and relates to the technical field of artificial intelligence, the method comprises the following steps: determining a node corresponding to a target user in a pre-constructed knowledge graph, the knowledge graph being used for describing a user, a product and interaction information between the user and the product; determining a to-be-recommended product list corresponding to the target user based on the knowledge graph; and inputting the to-be-recommended product list into a pre-trained product recommendation model, and outputting recommended products. According to the invention, more accurate and personalized product recommendation can be provided for the user.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a product recommendation method, device, and related equipment. Background Art

[0002] Traditional recommendation systems often rely on simple statistical analysis or rule-based methods, and are unable to combine and utilize various information such as the relationship between sales items, the relationship between users and sales items, and the relationship between users to analyze users' personalized preferences and behavior patterns. Moreover, traditional recommendation systems usually only consider the recommendation of a single sales item and ignore the user's potential demand for other sales items. In addition, although some methods have discussed the use of the above-mentioned related information to recommend sales items to users, these methods focus on theoretical design and ignore the efficiency problems caused by massive data in actual production (for example, an e-commerce platform may have tens of millions or even hundreds of millions of users). Therefore, the recommendation methods in related technologies have problems such as not fully utilizing various information and being unable to handle the massive data in actual production.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the Invention

[0004] The present disclosure provides a product recommendation method, apparatus, and related equipment, which, at least to a certain extent, overcome the problem in related art recommendation methods that various information is not fully utilized and the products recommended to users are not accurate.

[0005] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by practice of the present disclosure.

[0006] According to one aspect of the present disclosure, a product recommendation method is provided, comprising: determining a node corresponding to a target user in a pre-constructed knowledge graph, wherein the knowledge graph is used to describe users, products, and interaction information between users and products; determining a list of to-be-recommended products corresponding to the target user based on the knowledge graph; inputting the list of to-be-recommended products into a pre-trained product recommendation model, and outputting recommended products.

[0007] In some exemplary embodiments of the present disclosure, based on the aforementioned scheme, before determining the node corresponding to the target user in the pre-constructed knowledge graph, the method also includes: obtaining multiple users, multiple products, and interaction information between multiple users and multiple products; constructing an initialized knowledge graph based on multiple users, multiple products, and interaction information between multiple users and multiple products; and compressing the initialized knowledge graph to obtain the knowledge graph.

[0008] In some exemplary embodiments of the present disclosure, based on the aforementioned scheme, the list of products to be recommended corresponding to the target user is determined based on the knowledge graph, including: inputting the knowledge graph into a pre-trained graph embedding model, outputting the node vector of each node in the knowledge graph, the node vector including the node vector of the product to be recommended; and determining the list of products to be recommended corresponding to the target user based on the node vector of the product to be recommended.

[0009] In some exemplary embodiments of the present disclosure, based on the aforementioned scheme, the list of products to be recommended corresponding to the target user is determined according to the node vectors of the products to be recommended, including: expanding the list of products to be recommended for the target user based on the node vectors of the products that the target user has ordered in the knowledge graph; or, expanding the list of products to be recommended for the target user based on the node vectors of the products that similar users have ordered.

[0010] In some exemplary embodiments of the present disclosure, based on the aforementioned scheme, the initialized knowledge graph is compressed to obtain the knowledge graph, including: compressing the initialized knowledge graph based on the graph structure to obtain the knowledge graph; or compressing the initialized knowledge graph based on a clustering analysis algorithm to obtain the knowledge graph.

[0011] In some exemplary embodiments of the present disclosure, based on the aforementioned scheme, the initialized knowledge graph is compressed based on the graph structure to obtain the knowledge graph, including: extracting features from each node in the initialized knowledge graph; calculating the similarity of each node based on the features of each node in the initialized knowledge graph; and compressing the nodes in the initialized knowledge graph based on the similarity of each node to obtain the knowledge graph.

[0012] In some exemplary embodiments of the present disclosure, based on the aforementioned scheme, the product recommendation model also outputs the recommendation probability of each recommended product, inputs the list of products to be recommended into the pre-trained product recommendation model, and outputs the recommended products, including: sorting according to the recommendation probability of each recommended product; and outputting the recommended products to the target user in sequence based on the sorting.

[0013] According to another aspect of the present disclosure, a product recommendation device is also provided, including: a user node determination module, used to determine the node corresponding to the target user in a pre-constructed knowledge graph, wherein the knowledge graph is used to describe the interaction information between users, products, and users and products; a to-be-recommended product list determination module, used to determine the to-be-recommended product list corresponding to the target user based on the knowledge graph; and a recommended product output module, used to input the to-be-recommended product list into a pre-trained product recommendation model and output recommended products.

[0014] According to another aspect of the present disclosure, an electronic device is provided, including: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform any of the above-mentioned product recommendation methods by executing the executable instructions.

[0015] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the computer program implements any of the above-mentioned product recommendation methods.

[0016] According to another aspect of the present disclosure, a computer program product is provided, including: a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the method for recommending any one of the above products is implemented.

[0017] A product recommendation method, apparatus, and related equipment provided in the embodiments of the present disclosure visualize users, products, and interaction information between users and products through a knowledge graph, thereby better capturing the interests and preferences of target users. Furthermore, the knowledge graph can also be used to determine products related to the target user, forming a list of products to be recommended as a basis for personalized recommendations. Finally, the list of products to be recommended is input into a pre-trained product recommendation model to output more accurate, intelligent, and more suitable recommended products for the target user.

[0018] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0020] Figure 1 A schematic diagram illustrating an exemplary application system architecture of a product recommendation method according to an embodiment of the present disclosure;

[0021] Figure 2 A schematic diagram showing a recommended method for a product according to an embodiment of the present disclosure is shown;

[0022] Figure 3 An example diagram of an initialized knowledge graph according to an embodiment of the present disclosure is shown;

[0023] Figure 4 An example diagram of a knowledge graph after compression processing in an embodiment of the present disclosure is shown;

[0024] Figure 5 A schematic diagram of a process for compressing a very large knowledge graph based on the K-means algorithm in an embodiment of the present disclosure is shown;

[0025] Figure 6 A schematic diagram showing a recommended device for a product according to an embodiment of the present disclosure;

[0026] Figure 7 A schematic diagram of an electronic device illustrating a method for recommending an application product according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0027] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0028] In addition, the described features, structures or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, well-known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present disclosure.

[0029] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, while others may be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0030] Figure 1 FIG. 1 shows an exemplary application system architecture diagram to which the product recommendation method according to the embodiment of the present disclosure can be applied. Figure 1 As shown, the system architecture may include a terminal device 101 , a network 102 and a server 103 .

[0031] The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103 , and can be a wired network or a wireless network.

[0032] Optionally, the above-mentioned wireless network or wired network uses standard communication technologies and / or protocols. The network is typically the Internet, but can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network or any combination of a virtual private network). In some embodiments, technologies and / or formats including Hypertext Markup Language (HTML), Extensible Markup Language (XML), etc. are used to represent data exchanged over the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec), etc. can be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0033] The terminal device 101 can be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, wearable devices, augmented reality devices, virtual reality devices, etc.

[0034] Optionally, the client of the application installed in different terminal devices 101 is the same, or the client of the same type of application based on different operating systems. Based on the different terminal platforms, the specific form of the client of the application can also be different, for example, the application client can be a mobile phone client, a PC client, etc.

[0035] The server 103 may be a server that provides various services, such as a background management server that provides support for devices operated by users using the terminal device 301. The background management server may analyze and process received request data and feed back the processing results to the terminal device.

[0036] Optionally, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The terminal can be a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited to these. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in this application.

[0037] Those skilled in the art will know that Figure 1 The number of terminal devices, networks, and servers in the embodiment is merely illustrative, and any number of terminal devices, networks, and servers may be provided based on actual needs. This embodiment of the present disclosure does not limit this.

[0038] Under the above system architecture, an embodiment of the present disclosure provides a product recommendation method, which can be executed by any electronic device with computing and processing capabilities.

[0039] In some embodiments, the product recommendation method provided in the embodiments of the present disclosure can be executed by the terminal device of the above-mentioned system architecture; in other embodiments, the product recommendation method provided in the embodiments of the present disclosure can be executed by the server in the above-mentioned system architecture; in other embodiments, the product recommendation method provided in the embodiments of the present disclosure can be implemented by the terminal device and server in the above-mentioned system architecture through interaction.

[0040] First, in response to the above-mentioned problems, the present disclosure provides a method for recommending sales products that can be applied but is not limited to sales products. First, a large amount of user data, sales product data, and data on user browsing and ordering packages are combined to construct a super-large knowledge graph. By constructing the knowledge graph, the relationship between users, sales products, sales product packages, and user behaviors can be visually represented, thereby better capturing the user's interests and preferences. However, the initial knowledge graph constructed based on massive data is too large. Adding, deleting, modifying, and querying the knowledge graph containing hundreds of millions or even billions of nodes requires a large amount of computing and storage resources and cannot be directly applied to subsequent processes. Therefore, the present disclosure designs a method for compressing the scale of the knowledge graph. This method combines the graph structure of the knowledge graph and the semantic information of the nodes, reduces the scale of the knowledge graph by merging nodes with similar graph structures and similar semantic information, and at the same time retains the information contained in the knowledge graph as much as possible, thereby improving the data expression efficiency; then, the relational graph convolutional network (RelationalGraph ConvolutionalNet) is trained based on the compressed knowledge graph. In this paper, we use R-GCN (Regression-Gated CN) to obtain node embeddings. By training R-GCN based on knowledge graphs, we can effectively learn the relationships between nodes and generate corresponding node vectors for each node. These vectors contain rich semantic information and graph structure information, which can serve as the basis for personalized recommendations. Finally, we design a method for recommending packages to target users based on node embeddings. By building a more accurate, intelligent, and efficient package recommendation system, we can improve user experience, increase product sales, reduce recommendation costs, and provide companies with more precise marketing strategies.

[0041] Figure 2 A schematic diagram of a method for recommending a product according to an embodiment of the present disclosure is shown, and the method includes the following steps:

[0042] S202, determining the node corresponding to the target user in a pre-built knowledge graph, wherein the knowledge graph is used to describe users, products, and interaction information between users and products.

[0043] It should be noted that the product in the embodiments of the present disclosure can be any item that the user needs, for example, a sales item that the user needs to purchase, or a package containing items that the user needs to purchase, etc.; in addition, the knowledge graph in the embodiments of the present disclosure is a model for representing the relationship between information and data. It organizes information in a structured manner to facilitate machine understanding and processing. It is often composed of entities and the relationships between entities. The embodiments of the present disclosure regard users and products as entities, which are represented by nodes in the knowledge graph, and regard the interaction information between users and products as relationships, which are represented by edges in the knowledge graph.

[0044] In some embodiments, the present disclosure embodiment first needs to define the node data and relationship data of the knowledge graph in order to construct the knowledge graph, such as Figure 3 As shown, specifically, the user node is defined to describe user data, including user ID, nickname, registration date, gender, date of birth and other information; the sales product node is defined to describe sales product information, including sales product ID, name, description and other information; the package node is defined to describe package information, which is used to describe the package purchased by the user, including package ID, name, pricing, sales date and other information; the page node is defined, including page ID, resource link, sales package and other information, which is used to describe the page viewed by the user, and the package sold is displayed on the page.

[0045] In some embodiments, as Figure 3 As shown, the embodiment of the present disclosure defines relationship data including: defining a browsing relationship between users and page nodes, indicating that the user has browsed the page of a certain package; defining an ordering relationship between users and package nodes, indicating that the user has ordered a certain package; defining a display relationship between pages and packages, indicating the package displayed on the page; defining an inclusion relationship between pages and sales items, indicating the sales items included in the package. If some sales items cannot be sold in combination with other sales items, the package only contains one sales item.

[0046] It should be noted that the knowledge graph in the embodiment of the present disclosure can be further expanded to consider user purchase and collection package data, description of friendship relationships between users, etc.

[0047] S204: Determine a list of recommended products corresponding to the target user based on the knowledge graph.

[0048] It should be noted that the knowledge graph in the embodiment of the present disclosure contains a large number of packages, which cannot all be used as products to be recommended. Directly performing recommendation calculations on the complete knowledge graph may be very time-consuming and resource-intensive. The embodiment of the present disclosure can significantly reduce the amount of data that needs to be processed by pre-determining a list containing only a small number of products to be recommended, thereby speeding up the recommendation speed and reducing computing costs. In addition, narrowing the scope of candidate products can help focus on those products that are truly potential, avoid recommending irrelevant or low-quality products to users, and thus improve user experience and satisfaction.

[0049] S206: Input the list of products to be recommended into a pre-trained product recommendation model, and output the recommended products.

[0050] It should be noted that the product recommendation model in the embodiment of the present disclosure can be a model obtained by pre-training various artificial intelligence algorithm models (for example, neural network models) or their combination models through machine learning. The model can automatically identify the list of products to be recommended and match them to obtain related recommended products. The input data of the model is the data in the list of products to be recommended, and the output data is the products recommended to the target user.

[0051] Through the above embodiment, the product recommendation model obtained through machine learning training is used to automatically identify the target user's list of products to be recommended to obtain corresponding recommended products, thereby enabling rapid identification of the list of products to be recommended.

[0052] The product recommendation method provided in the embodiments of the present disclosure first determines the node corresponding to the target user in the pre-constructed knowledge graph; then, based on the knowledge graph, determines the list of products to be recommended corresponding to the target user; finally, inputs the list of products to be recommended into the pre-trained product recommendation model to output the recommended products. Compared with the recommendation methods in related technologies, which have the problem of not fully utilizing various information and the accuracy of the products recommended to users is not high, the embodiments of the present disclosure use the knowledge graph to visualize the user, the product, and the interaction information between the user and the product, so as to better capture the interests and preferences of the target user. In addition, the knowledge graph can also be used to determine the products related to the target user, forming a list of products to be recommended as the basis for personalized recommendations. Finally, the list of products to be recommended is input into the pre-trained product recommendation model to output more accurate, intelligent, and more suitable recommended products for the target user.

[0053] In some embodiments, after defining the node data and relationship data of the knowledge graph, the embodiment of the present disclosure also designs a method for extracting implicit information in the knowledge graph in subsequent steps, and designs a package recommendation method based on the extracted implicit information. Since there are a large number of users, sales products and packages in the ultra-large knowledge graph, the amount of data may reach more than one million. Initialization through one-hot encoding will result in a large amount of computing and storage overhead, so the embodiment of the present disclosure designs a more efficient method to represent nodes with low-dimensional dense vectors.

[0054] Specifically, the embodiment of the present disclosure first converts the node into text, splices the node's features and feature values after the node name, distinguishes different attribute values with semicolons, and handles features with missing values with the special identifier null. For example, a user node can be converted into the text "User: ID 999; Nickname: Zhang San; Date of Birth: null; Registration Date: January 1, 2024"; then, the text is input into the text embedding model to obtain an initialization vector. For example, the text is input into the "bge-small-zh-v1.5" text embedding model to obtain a 512-dimensional vector. The embodiment of the present disclosure inputs the text information of the node into the text embedding model to obtain the initial vector, which can retain the semantic information of the node, so that the vectors corresponding to nodes with similar information are also similar.

[0055] In some embodiments, before determining the node corresponding to the target user in the pre-built knowledge graph, the product recommendation method in the embodiment of the present disclosure further includes: obtaining multiple users, multiple products, and interaction information between multiple users and multiple products; constructing an initialized knowledge graph based on multiple users, multiple products, and interaction information between multiple users and multiple products; compressing the initialized knowledge graph to obtain a knowledge graph. Specifically, by compressing the initialized knowledge graph, the number of nodes and edges is reduced, thereby significantly reducing storage requirements and saving memory resources. Moreover, the compressed knowledge graph is usually smaller and more concise, and the number of nodes and edges that need to be traversed for query and reasoning operations is reduced, which speeds up the search speed on the graph and improves the real-time response capability. In addition, the compressed knowledge graph reduces the amount and complexity of data, thereby reducing the computational burden of algorithms (such as cluster analysis, graph embedding, machine learning, etc.), improving the scalability of the algorithm, and enabling it to process larger-scale data.

[0056] In some embodiments, many organizations have a large number of customers, sales products and packages, which leads to the generation of massive amounts of relevant data. The initialized knowledge graph is too large. If it is used directly to train the graph neural network, it will lead to high model training costs, long training time, slow model convergence or even failure to converge. Therefore, the embodiment of the present disclosure compresses the initialized knowledge graph while retaining the knowledge graph features as much as possible, reducing the scale of the knowledge graph data and improving the efficiency of knowledge graph data expression. In order to simplify subsequent processing, the subsequent steps only consider merging nodes of the same type, and do not discuss merging nodes of different types. For example, only the case of merging user node 1 and user node 2 is considered, and the case of merging user nodes and product nodes is not considered. Specifically, all user nodes are divided into user node sets, all package nodes are divided into package node sets, and all sales product nodes are divided into sales product node sets. It should be noted that the subsequent steps are limited to merging and compressing nodes of the same type.

[0057] In some embodiments, the disclosed embodiments determine a list of products to be recommended corresponding to a target user based on a knowledge graph, including: inputting the knowledge graph into a pre-trained graph embedding model, outputting a node vector for each node in the knowledge graph, the node vector including the node vector of the product to be recommended; and determining a list of products to be recommended corresponding to the target user based on the node vector of the product to be recommended. Specifically, the disclosed embodiments map the nodes in the knowledge graph to low-dimensional vectors through a graph embedding model, avoiding direct calculations on a complex graph structure. Compared to directly operating the original knowledge graph, the vectorized representation greatly reduces the demand for computing resources. Moreover, the graph embedding model can capture the semantic relationships and structural information between nodes in the knowledge graph, thereby generating a more representative node vector. The node vector not only contains local connection information, but also reflects the global topological structure and semantic association, making the recommendation more in line with user needs. By comparing the vector of the target user with the vector of the product to be recommended, the disclosed embodiments can accurately find the product most relevant to the user's interests.

[0058] In some embodiments, the node vector of each node obtained by the present disclosure embodiment through the knowledge graph can make the Euclidean distance of user vectors with similar user information closer, but because these vectors do not contain the knowledge graph structure information during initialization, it cannot be guaranteed that the Euclidean distance of users with similar package preferences is closer. For example, user 1 and user 2 may be similar in user information, but it does not mean that they have similar package preferences. Therefore, the present disclosure embodiment performs training based on the node vectors in the knowledge graph, so that the node vectors contain the structural information of the knowledge graph, so that the vectors of nodes with similar structures are closer, that is, the Euclidean distance of the vectors of user nodes connected to the same package is closer. For example, when user 1 and user 2 both show interest in the same page and package (browse or order), the vectors of these two user nodes should be closer.

[0059] In some embodiments, the disclosed embodiments use the R-GCN training graph embedding model to update the node vector so that the node vector contains the structural information of the knowledge graph: the vectors around the node are

[0060]

[0061] in, represents the hidden state of node i in the lth layer of the neural network, represents the hidden state of node i in the l+1th layer of the neural network, σ is the activation function, is the set of adjacent nodes of node i under relation r, c i,r is a regularization parameter that can be specified as R represents a set of relationship types, r represents a relationship, Represents the parameter matrix of relation r in the (lth) layer of the neural network, A parameter matrix representing a special relationship, that is, the parameter matrix of the relationship between each node and itself in the (l)th layer of the neural network.

[0062] In some embodiments, the disclosed embodiment inputs the implicit vector h into the subsequent network and outputs the classification of the node by calculating the loss function of the node's true label category (such as user node, sales product node, etc.) and the estimated label. For example, the disclosed embodiment can use the cross entropy loss formula (2) below to optimize the model. The specific loss function is:

[0063]

[0064] Among them, Y represents the node set, i represents one of the samples, k represents a type of node, t ik Represents the value of node i on type k, that is, if node i is type, then t ik =1, otherwise 0; Represents the predicted probability of whether node i is of type k output by the neural network.

[0065] The above R-GCN network can be updated to converge the network. It should be noted that the graph embedding model in the embodiment of the present disclosure only requires the first half of the network, that is, the hidden state h, as the vector representation of the node.

[0066] In some embodiments, the embodiments of the present disclosure provide a graph embedding model training method based on R-GCN, which can be trained based on a compressed knowledge graph to obtain the implicit vector of each node as the embedding of the node. The node embedding contains the knowledge graph structure information, so that the Euclidean distance of the vectors of nodes with similar graph structures is closer.

[0067] In some embodiments, the disclosed embodiment determines the list of products to be recommended corresponding to the target user based on the node vector of the product to be recommended, including: expanding the target user's list of products to be recommended based on the node vector of the target user's subscription in the knowledge graph; or, expanding the target user's list of products to be recommended based on the node vector of the product to be recommended that has been subscribed by similar users. Specifically, since the knowledge graph contains a large number of packages, they cannot all be used as candidate packages to be recommended, otherwise the response speed of the package recommendation algorithm will be greatly limited. Therefore, the disclosed embodiment needs to determine in advance a list containing only a small number of packages to be recommended, which is used as the input of the product recommendation model. The disclosed embodiment expands the list of products to be recommended through two methods. One is to expand the list of products to be recommended based on the sales items included in the package that the user has subscribed to: according to the package that the user has subscribed to, the sales items in the package are taken into account, and then the packages where these sales items are located are obtained and put into the list of products to be recommended. For example, Figure 3In the example, user 2 has ordered package C, which contains product d. Product d is included in both package C and package D, so package C and package D can be used as recommended packages. Another approach is to expand the list of recommended products based on packages ordered by similar users: Based on the graph embedding model, user nodes can be converted into corresponding low-dimensional dense vectors. The vector of user node i is recorded as v i , the vector of user node j is recorded as v j , the embodiment of the present disclosure uses vector inner product to measure the similarity between two nodes, that is, the similarity s between user i and user j ij , where the similarity s ij Determined by the following formula (3):

[0068] s ij =v i ×v j (3)

[0069] It should be noted that the embodiment of the present disclosure can calculate similar user nodes in advance, and does not need to calculate user similarity in real time, so it will not slow down the recommendation speed. After obtaining similar user nodes, the packages ordered by users with high similarity are taken and put into the list of packages to be recommended. For example, Figure 3 In the example, if user 1 is similar to user 2, when recommending packages for user 2, user 1's subscribed packages, namely, package A and package B, can be recommended to user 2.

[0070] In some embodiments, the disclosed embodiments first determine whether the graph structures of the nodes in the knowledge graph are consistent. The nodes with consistent graph structures are aggregated using the mean method, and the nodes with the same original graph structure are compressed into a new node. This method compresses the scale of ultra-large knowledge graphs while retaining the original information in the knowledge graph as much as possible, improves the data expression efficiency of the knowledge graph, reduces the calculation and storage costs of the knowledge graph, and facilitates the subsequent training of the graph embedding model.

[0071] In some embodiments, the embodiments of the present disclosure compress the initialized knowledge graph to obtain a knowledge graph, including: compressing the initialized knowledge graph based on the graph structure to obtain the knowledge graph; or compressing the initialized knowledge graph based on the cluster analysis algorithm to obtain the knowledge graph. Specifically, when the graph structures of the nodes in the initialized knowledge graph are similar, the two nodes can be considered similar, and therefore can be merged and compressed. For example, if there are two packages containing the same sales products, the two packages can be considered to be the same package, so the two package nodes can be merged. The specific steps of the compression processing method based on the graph structure used in the embodiments of the present disclosure include: dividing the package nodes associated with the same sales product nodes into the same cluster, and then merging the nodes in the same cluster by the following formula (4); then, retaining the union of all edges of the original nodes in the cluster and the external nodes as the edges of the new node; finally, deleting the original nodes and edge associations in the cluster.

[0072]

[0073] Where A represents the number of all nodes in the cluster, v a The vectors representing these nodes, v new A new vector representing the retained nodes.

[0074] In some embodiments, as Figure 4 As shown, in the embodiment of the present disclosure, through the compression processing method as above, the package C node and the package D node only contain the sales item d, so they can be merged to obtain the node sales item C'. Since user 2 purchased package C and user 4 purchased package D, after the package C node and the package D node are merged, it is obtained that both user 2 and user 4 purchased package C'.

[0075] In some embodiments, the embodiments of the present disclosure divide user nodes associated with exactly the same package nodes into the same cluster through the above compression processing method, and merge the nodes in the same cluster.

[0076] In some embodiments, the embodiment of the present disclosure can reduce the complexity of the knowledge graph to a certain extent through the compression processing method as described above, but it cannot guarantee that the data magnitude of the knowledge graph is reduced to the required level. The embodiment of the present disclosure pays attention to the semantic information contained in the node when initializing the node vector, so the vector can be clustered to reduce the nodes with similar semantics. For this purpose, the embodiment of the present disclosure designs a knowledge graph compression method based on a clustering analysis algorithm (such as the K-means clustering algorithm), such as Figure 5 As shown, the steps include:

[0077] S502: Determine the number of nodes in each cluster (eg, N=10), and determine the number of clusters M to be divided based on the current number of nodes, and randomly select M nodes as initial central nodes.

[0078] S504: N nodes with the closest Euclidean distance to the central node vector are grouped into a cluster.

[0079] S506: Update the central node vector in the cluster to the average value of all node vectors in the cluster.

[0080] S508 , repeatedly iterate S504 and S506 until the cluster converges.

[0081] S510: Merge the nodes in the same cluster. At this time, the number of graph nodes is reduced to 1 / N of the original number.

[0082] In some embodiments, the disclosed embodiments place similar nodes into the same cluster through the K-means algorithm, and then use the mean method to compress the nodes in the same cluster. This method compresses the scale of ultra-large knowledge graphs while retaining the original information in the knowledge graphs as much as possible, thereby improving the data expression efficiency of the knowledge graphs, reducing the computational and storage costs of the knowledge graphs, and facilitating subsequent training of graph embedding models.

[0083] In some embodiments, the disclosed embodiments compress the initialized knowledge graph based on the graph structure to obtain a knowledge graph, including: extracting features from each node in the initialized knowledge graph; calculating the similarity of each node based on the features of each node in the initialized knowledge graph; compressing the nodes in the initialized knowledge graph based on the similarity of each node to obtain a knowledge graph. The disclosed embodiments achieve the purpose of streamlining the knowledge graph structure while retaining key information through feature extraction, similarity calculation and node compression processing. Specifically, by extracting features from each node in the initialized knowledge graph and merging or deleting nodes based on similarity, the number of nodes in the graph can be significantly reduced, storage requirements can be reduced, and storage costs can be reduced because the amount of data to be processed is smaller, thereby improving query speed and computing efficiency. Moreover, by merging similar nodes, duplicate information is reduced, and the consistency and quality of data are improved. The streamlined knowledge graph is more focused on key information, thereby improving the accuracy of recommendation systems or other applications.

[0084] In some embodiments, the product recommendation model in the embodiment of the present disclosure also outputs the recommendation probability of each recommended product, inputs the list of products to be recommended into the pre-trained product recommendation model, and outputs the recommended products, including: sorting according to the recommendation probability of each recommended product; and outputting the recommended products to the target user in sequence based on the sorting.

[0085] In some embodiments, before inputting the list of products to be recommended into a pre-trained product recommendation model and outputting the recommended products, the embodiment of the present disclosure further includes: training the product recommendation model. Specifically, the embodiment of the present disclosure constructs a fully connected neural network and maps the node vectors to the relationship types between the nodes using the following formula (5):

[0086]

[0087] Among them, d v is the node embedding dimension, L = {0, 1}, the input is a vector concatenated from any two nodes, and the output is whether there is an ordering relationship between the two nodes, 0 means no, 1 means yes.

[0088] In some embodiments, the present disclosure embodiment trains the neural network to obtain the product recommendation model, further comprising: in the knowledge graph, a (head node, relationship, tail node) triple can be obtained, for example, (user 1, order, package A), the node vector of "user 1" and the node vector of "package A" are spliced and input into the fully connected neural network to obtain a prediction As the output, it indicates whether there is a subscription relationship between user 1 and package A. Finally, the actual result y is calculated. true and The loss function is used to update the neural network model, and the product recommendation model is obtained by repeated iterations until the model converges.

[0089] In some embodiments, the embodiments of the present disclosure obtain a product recommendation model by training a neural network. By inputting the vectors of any two nodes into the product recommendation model, it can output whether there is an ordering relationship between the two nodes. When the vectors of the user node and the package node are input into the product recommendation model, when it is determined that there is an ordering relationship, the recommended product is finally output.

[0090] In some embodiments, the disclosed embodiments incorporate information such as users, sales products, packages containing sales products, browsing and purchasing into the same ultra-large knowledge graph, which can consider the user's potential demand for multiple sales products, make full use of various information to analyze the user's personalized preferences and behavior patterns, and realize an end-to-end sales product package recommendation method, which can improve user experience, increase sales of sales products, and provide enterprises with more accurate marketing strategies.

[0091] Based on the same inventive concept, the present disclosure also provides a product recommendation device, such as the following embodiment. Since the principle of solving the problem in the device embodiment is similar to that in the above method embodiment, the implementation of the device embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0092] Figure 6A schematic diagram of a recommended device for a product according to an embodiment of the present disclosure is shown, the device comprising:

[0093] A user node determination module 601 is used to determine the node corresponding to the target user in a pre-built knowledge graph, where the knowledge graph is used to describe users, products, and interaction information between users and products;

[0094] A module 602 for determining a list of products to be recommended corresponding to a target user based on the knowledge graph;

[0095] The recommended product output module 603 is used to input the list of products to be recommended into the pre-trained product recommendation model and output the recommended products.

[0096] A product recommendation device provided in an embodiment of the present disclosure determines the node corresponding to the target user in a pre-constructed knowledge graph through a user node determination module; determines the list of products to be recommended corresponding to the target user based on the knowledge graph through a to-be-recommended product list determination module; and inputs the list of products to be recommended into a pre-trained product recommendation model through a recommended product output module to output recommended products. Compared with the recommendation methods in related technologies that do not fully utilize various information and the accuracy of the products recommended to users is not high, the embodiment of the present disclosure uses a knowledge graph to visually represent the user, product, and the interaction information between the user and the product, thereby better capturing the interests and preferences of the target user. In addition, the knowledge graph can also be used to determine products related to the target user, forming a list of products to be recommended as a basis for personalized recommendations. Finally, the list of products to be recommended is input into a pre-trained product recommendation model to output more accurate, intelligent, and more suitable recommended products for the target user.

[0097] In some embodiments, the product recommendation device in the embodiments of the present disclosure further includes: an information acquisition module, which is used to obtain multiple users, multiple products, and interaction information between multiple users and multiple products before determining the node corresponding to the target user in a pre-constructed knowledge graph; an initialized knowledge graph construction module, which is used to construct an initialized knowledge graph based on multiple users, multiple products, and interaction information between multiple users and multiple products; and a compression processing module, which is used to compress the initialized knowledge graph to obtain a knowledge graph.

[0098] In some embodiments, the module for determining the list of products to be recommended in the embodiments of the present disclosure is also used to input the knowledge graph into a pre-trained graph embedding model, output the node vector of each node in the knowledge graph, and the node vector includes the node vector of the product to be recommended; and determine the list of products to be recommended corresponding to the target user based on the node vector of the product to be recommended.

[0099] In some embodiments, the module for determining the list of products to be recommended in the embodiments of the present disclosure is also used to expand the target user's list of products to be recommended based on the node vectors that the target user has ordered in the knowledge graph; or, to expand the target user's list of products to be recommended based on the node vectors that similar users have ordered.

[0100] In some embodiments, the compression processing module in the embodiments of the present disclosure is also used to compress the initialized knowledge graph based on the graph structure to obtain a knowledge graph; or, to compress the initialized knowledge graph based on a clustering analysis algorithm to obtain a knowledge graph.

[0101] In some embodiments, the compression processing module in the embodiments of the present disclosure is also used to extract features from each node in the initialized knowledge graph; calculate the similarity of each node based on the features of each node in the initialized knowledge graph; and compress the nodes in the initialized knowledge graph based on the similarity of each node to obtain a knowledge graph.

[0102] In some embodiments, the recommended product output module in the embodiments of the present disclosure is further used to sort the recommended products according to the recommendation probability of each recommended product; and output the recommended products to the target user in sequence based on the sorting.

[0103] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods, or program products. Therefore, various aspects of the present disclosure may be implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits," "modules," or "systems."

[0104] Based on the same inventive concept, an embodiment of the present disclosure further provides an electronic device, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any of the aforementioned product recommendation methods by executing the executable instructions. Because the principles for solving the problem in this electronic device embodiment are similar to those in the aforementioned method embodiment, the implementation of this electronic device embodiment can refer to the implementation of the aforementioned method embodiment, and any repetitive details will not be repeated.

[0105] Refer to the following Figure 7 700 according to this embodiment of the present disclosure will be described. Figure 7 The electronic device 700 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0106] like Figure 7As shown, electronic device 700 is implemented as a general-purpose computing device. Components of electronic device 700 may include, but are not limited to, at least one processing unit 701, at least one storage unit 702, and a bus 703 connecting different system components (including storage unit 702 and processing unit 701).

[0107] The storage unit stores program codes, which can be executed by the processing unit 701, so that the processing unit 701 executes the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification.

[0108] In some embodiments, when the electronic device is used to control the recommendation method of the product disclosed above, the processing unit 701 may perform the following steps of the above method embodiment:

[0109] Determine the node corresponding to the target user in the pre-built knowledge graph, where the knowledge graph is used to describe users, products, and the interaction information between users and products; determine the list of recommended products corresponding to the target user based on the knowledge graph; input the list of recommended products into the pre-trained product recommendation model and output the recommended products.

[0110] The storage unit 702 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 7021 and / or a cache memory unit 7022 , and may further include a read-only memory unit (ROM) 7023 .

[0111] The storage unit 702 may also include a program / utility 7024 having a set (at least one) of program modules 7025, such program modules 7025 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0112] Bus 703 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0113] The electronic device 700 may also communicate with one or more external devices 704 (e.g., a keyboard, a pointing device, a Bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device 700, and / or any device that enables the electronic device 700 to communicate with one or more other computing devices (e.g., a router, a modem, etc.). Such communication may be performed via an input / output (I / O) interface 705. Furthermore, the electronic device 700 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 706. As shown, the network adapter 706 communicates with other modules of the electronic device 700 via a bus 703. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 700, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0114] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0115] Based on the same inventive concept, embodiments of the present disclosure also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements any of the aforementioned product recommendation methods. Because the principles underlying the problem solved by this computer-readable storage medium embodiment are similar to those of the aforementioned method embodiment, the implementation of this computer-readable storage medium embodiment can be referenced to the implementation of the aforementioned method embodiment, and any repetitions will not be repeated.

[0116] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0117] In the present disclosure, a computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0118] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination thereof.

[0119] In a specific implementation, the program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, and the like, as well as conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0120] Based on the same inventive concept, embodiments of the present disclosure further provide a computer program product, including a computer program or instructions, which, when executed by a processor, implements the recommended method for implementing any of the aforementioned method embodiments. Because the principles underlying the problem solved by this computer program product embodiment are similar to those of the aforementioned method embodiments, the implementation of this computer program product embodiment can be referenced to the implementation of the aforementioned method embodiments, and any repetitions will not be repeated.

[0121] It should be noted that although several modules or units of the device for action execution are mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0122] Furthermore, although the steps of the method of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in this particular order, or that all steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0123] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0124] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the appended claims.

Claims

1. A product recommendation method, characterized in that: include: Determine the node corresponding to the target user in a pre-built knowledge graph, where the knowledge graph is used to describe users, products, and interactions between users and products; Determine a list of products to be recommended corresponding to the target user based on the knowledge graph; Input the list of products to be recommended into the pre-trained product recommendation model and output the recommended products.

2. The product recommendation method according to claim 1, characterized in that: Before determining the node corresponding to the target user in the pre-built knowledge graph, the method further includes: Obtaining information about multiple users, multiple products, and interactions between multiple users and multiple products; Build an initial knowledge graph based on multiple users, multiple products, and the interaction information between multiple users and multiple products; The initialized knowledge graph is compressed to obtain the knowledge graph.

3. The product recommendation method according to claim 1, characterized in that: Determining a list of recommended products corresponding to the target user based on the knowledge graph includes: Input the knowledge graph into a pre-trained graph embedding model, and output a node vector for each node in the knowledge graph, wherein the node vector includes a node vector for the product to be recommended; A list of products to be recommended corresponding to the target user is determined according to the node vectors of the products to be recommended.

4. The product recommendation method according to claim 3, characterized in that: Determining a list of products to be recommended corresponding to the target user according to the node vectors of the products to be recommended includes: Expanding the target user's list of products to be recommended based on the node vectors that the target user has ordered in the knowledge graph; Alternatively, the list of products to be recommended for the target user is expanded based on the node vectors ordered by similar users.

5. The product recommendation method according to claim 2, characterized in that: The initialized knowledge graph is compressed to obtain the knowledge graph, including: Compressing the initialized knowledge graph based on the graph structure to obtain the knowledge graph; Alternatively, the initialized knowledge graph is compressed based on a clustering analysis algorithm to obtain the knowledge graph.

6. The product recommendation method according to claim 5, characterized in that: The initialization knowledge graph is compressed based on the graph structure to obtain the knowledge graph, including: Performing feature extraction on each node in the initialized knowledge graph; Calculate the similarity of each node based on the characteristics of each node in the initialized knowledge graph; The nodes in the initialized knowledge graph are compressed according to the similarity of each node to obtain the knowledge graph.

7. The product recommendation method according to claim 1, characterized in that: The product recommendation model also outputs the recommendation probability of each recommended product. The list of products to be recommended is input into the pre-trained product recommendation model to output the recommended products, including: Sort by the recommendation probability of each recommended product; Recommended products are output to the target user in sequence based on the ranking.

8. A product recommendation device, characterized in that: include: A user node determination module is used to determine the node corresponding to the target user in a pre-built knowledge graph, wherein the knowledge graph is used to describe users, products, and the interaction information between users and products; A module for determining a list of products to be recommended, configured to determine a list of products to be recommended corresponding to the target user based on the knowledge graph; The recommended product output module is used to input the list of products to be recommended into the pre-trained product recommendation model and output the recommended products.

9. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; The processor is configured to execute the product recommendation method according to any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the product recommendation method according to any one of claims 1 to 7 is implemented.

11. A computer program product comprising: A computer program or instruction, characterized in that when the computer program or instruction is executed by a processor, it implements the product recommendation method according to any one of claims 1 to 7.

Images