Recommended methods, apparatus, devices, and storage media for aggregation pages

Abstract

This application provides a method, apparatus, device, and storage medium for recommending aggregated pages, relating to the field of information retrieval technology. The method includes: acquiring a trained graph model; querying nodes in the graph model based on the items already operated by the user to be recommended, to obtain a first representation of the operated items; querying nodes in the graph model based on candidate item aggregated pages, to obtain a second representation of each candidate item aggregated page; and, if the similarity between the first and second representations meets a first predetermined condition, determining that the candidate item aggregated page should be sent as the recommended aggregated page to the device associated with the user to be recommended. Thus, based on the graph model, the purpose of recommending aggregated pages to the user to be recommended is achieved, improving the accuracy of aggregated page recommendations.

Description

Technical Field

[0001] This application relates to the field of information retrieval technology, and in particular to a method, apparatus, device, and storage medium for recommending aggregated pages. Background Technology

[0002] In related technologies, aggregation page recommendation systems typically analyze users' existing operational behaviors to obtain the similarity between products, and then recommend aggregation pages to users based on the similarity between products. These aggregation pages contain similar items to items that the user has liked. However, this method of recommending aggregation pages lacks sufficient mining of the characteristics of aggregation page materials, resulting in the aggregation pages not being accurately matched with user needs, thus leading to low accuracy in the recommended aggregation pages.

[0003] Application content

[0004] This application provides a recommended method, apparatus, device, and medium for generating aggregated pages to at least address the existing problems.

[0005] The technical solution of this application is as follows:

[0006] This application provides a method for recommending aggregation pages, comprising: acquiring a trained graph model; wherein the graph model includes multiple nodes, each node corresponding to an item or an item aggregation page, used to indicate the representation of the corresponding item or item aggregation page; the representation is generated by the behavioral correlation of the item or item aggregation page in historical access behavior, the behavioral correlation being used to indicate that the item or item aggregation page has a co-occurrence relationship in historical access behavior; querying nodes in the graph model according to the items already operated by the user to be recommended, to obtain a first representation of the operated items; querying nodes in the graph model according to candidate item aggregation pages, to obtain a second representation of each candidate item aggregation page; and determining to send the candidate item aggregation page as a recommended aggregation page to the device associated with the user to be recommended if the similarity between the first representation and the second representation meets a first preset condition.

[0007] Optionally, the step of querying nodes in the graph model based on multiple candidate item aggregation pages to obtain a second representation of each candidate item aggregation page includes: for any candidate item aggregation page, querying the nodes of the corresponding item aggregation page in the graph model; if there are no nodes of the corresponding item aggregation page among the nodes of the corresponding item aggregation page, then querying the nodes of the corresponding items in the graph model based on the target items contained in the candidate item aggregation page to obtain the representation of each target item; and using the aggregation function of the graph model to aggregate the representations of each target item to obtain the second representation.

[0008] Optionally, the target items are multiple and belong to at least two categories; the aggregation of the representations of each target item using the aggregation function of the graph model to obtain the second representation includes: for each category, aggregating the representations of each target item in the corresponding category using the aggregation function of the graph model according to the target items contained in the category, to obtain the second representation in the corresponding category.

[0009] Optionally, determining whether to send the candidate item aggregation page as a recommended aggregation page to the device associated with the user to be recommended based on the similarity between the first representation and the second representation includes: if the similarity between the second representation under any category of the candidate item aggregation page and the first representation meets a first preset condition, then the candidate item aggregation page is sent as the recommended aggregation page to the device associated with the user to be recommended.

[0010] Optionally, before acquiring the trained graph model, the method further includes: acquiring the initial representation of each node in the graph model; taking any node in the graph model as the current node, selecting the k-th order nearest neighbor node of the current node from the graph model, where k is a positive integer with a fixed value; aggregating the initial representation of the k-th order nearest neighbor node and the initial representation of the current node using an aggregation function to obtain the updated representation of the current node; taking at least one nearest neighbor node of the current node in the graph model as a positive sample node, and taking at least one node in the graph model other than the nearest neighbor node as a negative sample node; determining a loss function to indicate the difference between the first representation similarity and the second representation similarity based on the first representation similarity between the updated representation of the current node and the positive sample node, and based on the second representation similarity between the updated representation of the current node and the negative sample node; and adjusting the representation of each node in the graph model and adjusting the parameters of the aggregation function based on the loss function.

[0011] Optionally, obtaining the initial representation of each node in the graph model includes: randomly configuring the initial representation of each node in the graph model; or, using the representation of each node determined by the previous parameter adjustment as the initial representation.

[0012] Optionally, before obtaining the initial representation of each node in the graph model, the method further includes: determining the items and item aggregation pages with operational behavior based on historical access behavior; configuring the nodes corresponding to the items or item aggregation pages; for any two nodes, determining whether there is a co-occurrence relationship between the two nodes based on whether the operational behavior of the items or item aggregation pages corresponding to the two nodes meets a second set condition; and connecting the two nodes with the co-occurrence relationship to obtain the graph model.

[0013] Optionally, the second set condition includes that the operation belongs to the same user and the time interval between operations is not greater than a threshold.

[0014] Optionally, after determining whether there is a co-occurrence relationship between the two nodes based on whether the operation behavior of the items corresponding to the two nodes or the item aggregation page meets the second set condition, the method further includes: for the two nodes that have a co-occurrence relationship, counting the co-occurrence frequency; and determining the weight of the edge between the two nodes based on the co-occurrence frequency; wherein the weight is used to represent the nodes connected by the aggregation function.

[0015] This application provides a method for recommending aggregated pages, which involves obtaining a trained graph model. The graph model includes multiple nodes, each corresponding to an item or an item aggregated page, used to indicate the representation of the corresponding item or item aggregated page. The representation is generated based on the behavioral correlation of items or item aggregated pages in historical access behavior, and the behavioral correlation indicates the co-occurrence relationship of items or item aggregated pages in historical access behavior. Based on the items already operated by the user to be recommended, nodes in the graph model are queried to obtain a first representation of the operated items. Based on candidate item aggregated pages, nodes in the graph model are queried to obtain a second representation of each candidate item aggregated page. If the similarity between the first and second representations meets a first preset condition, the candidate item aggregated page is determined to be sent as the recommended aggregated page to the device associated with the user to be recommended. Therefore, based on the graph model, the purpose of recommending aggregated pages to the user to be recommended is achieved, improving the accuracy of aggregated page recommendations.

[0016] This application embodiment also provides a recommendation device for aggregated pages, comprising: a first acquisition module, configured to acquire a trained graph model; wherein the graph model includes multiple nodes, each node corresponding to an item or an item aggregated page, used to indicate the representation of the corresponding item or item aggregated page; the representation is generated by the behavioral correlation of the item or the item aggregated page in historical access behavior, the behavioral correlation being used to indicate that the item or the item aggregated page has a co-occurrence relationship in historical access behavior; a first representation module, configured to query nodes in the graph model based on the items already operated by the user to be recommended, to obtain a first representation of the operated items; a second representation module, configured to query nodes in the graph model based on candidate item aggregated pages, to obtain a second representation of each candidate item aggregated page; and a recommendation module, configured to determine to send the candidate item aggregated page as a recommended aggregated page to the device associated with the user to be recommended if the similarity between the first representation and the second representation meets a first preset condition.

[0017] This application also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the steps of the method described above.

[0018] This application also provides a computer-readable storage medium storing a computer program thereon, characterized in that the computer program implements the above-described method when executed by a processor.

[0019] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects:

[0020] When recommending aggregated pages to users, a trained graph model is obtained. This model includes multiple nodes, each corresponding to an item or item aggregated page, indicating its representation. The representation is generated based on the behavioral correlations of items or item aggregated pages in historical access behavior, indicating co-occurrence relationships. Based on the items already operated on by the user to be recommended, nodes in the graph model are queried to obtain a first representation of the operated items. Based on candidate item aggregated pages, nodes in the graph model are queried to obtain a second representation of each candidate item aggregated page. If the similarity between the first and second representations meets a first predetermined condition, the candidate item aggregated page is selected as the recommended aggregated page and sent to the device associated with the user to be recommended. Thus, based on the user's operational behavior, aggregated pages are accurately recommended to the user, improving the accuracy of aggregated page recommendations.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0022] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0023] Figure 1 A flowchart illustrating a method for recommending aggregated pages provided in this application embodiment. Figure 1 ;

[0024] Figure 2 A flowchart illustrating a method for recommending aggregated pages provided in this application embodiment. Figure 2 ;

[0025] Figure 3 A flowchart illustrating a method for recommending aggregated pages provided in this application embodiment. Figure 3 ;

[0026] Figure 4A schematic diagram of a recommendation device for an aggregated page provided in an embodiment of this application;

[0027] Figure 5 A schematic diagram of another recommended device for an aggregated page provided in an embodiment of this application;

[0028] Figure 6 This is a schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application. Detailed Implementation

[0029] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0030] The following description, with reference to the accompanying drawings, outlines a recommended method, apparatus, and electronic device for creating an aggregated page according to embodiments of this application.

[0031] Figure 1 A flowchart illustrating a method for recommending aggregated pages provided in this application embodiment. Figure 1 .

[0032] like Figure 1 As shown, the recommendation method for this aggregation page includes the following steps:

[0033] Step 101: Obtain the trained graph model; wherein, the graph model includes multiple nodes, each node corresponding to an item or item aggregation page, used to indicate the representation of the corresponding item or item aggregation page, the representation is generated based on the behavioral correlation of the item or item aggregation page in historical access behavior, and the behavioral correlation is used to indicate that there is a co-occurrence relationship between the item or item aggregation page in historical access behavior.

[0034] It should be noted that the recommendation method for the aggregation page can be executed by a recommendation device for the aggregation page, which can be implemented by software and / or hardware. The aggregation page recommendation can be an electronic device, or it can be configured within an electronic device to perform aggregation page recommendations.

[0035] The electronic device can be any device with computing capabilities, such as a terminal device or a server.

[0036] The terminal device includes, but is not limited to, any one of the following: mobile phone, personal desktop computer, laptop computer, tablet computer.

[0037] In some examples, a graph model is formed based on the co-occurrence relationships of items or item aggregation pages and their corresponding items or item aggregation pages in historical access behavior.

[0038] It is understandable that the item aggregation page contains various corresponding items.

[0039] Step 102: Based on the items already operated by the user to be recommended, query the nodes in the graph model to obtain the first representation of the operated items.

[0040] In some examples, based on the items that the user to be recommended has already operated on, the corresponding nodes of the operated items are queried in the trained graph model, and the first representation of the operated items is obtained based on the nodes of the operated items.

[0041] Step 103: Based on multiple candidate item aggregation pages, query the nodes in the graph model to obtain the second representation of each candidate item aggregation page.

[0042] It is understandable that, based on the user's operational behavior, multiple item aggregation pages can be selected from the material library to obtain multiple candidate item aggregation pages recommended to the user.

[0043] It is understood that the actions in this example may include, but are not limited to, the actions of the user to be recommended and the browsing behavior of the user to be recommended.

[0044] It is understandable that the material library is used to store various items and item aggregation pages.

[0045] In some examples, for any candidate item aggregation page, the corresponding node of the item aggregation page in the graph model is queried. If the node of the corresponding item aggregation page does not contain the node of the candidate item aggregation page, then the node of the corresponding item in the graph model is queried according to the target item contained in the candidate item aggregation page to obtain the representation of each target item. The aggregation function of the graph model is used to aggregate the representations of each target item to obtain the second representation.

[0046] Correspondingly, if a node of a candidate item aggregation page exists among the nodes of the corresponding item aggregation page in the graph model, then the second representation of the candidate item aggregation page is obtained based on the nodes of the candidate item aggregation page in the graph model.

[0047] It is understood that the graph model in this example has been trained, and the trained graph model includes nodes for multiple corresponding item aggregation pages and nodes for multiple corresponding items.

[0048] In this context, it can be understood that the nodes corresponding to the items in the graph model are queried to obtain the nodes corresponding to each target item, and based on the nodes corresponding to each target item, the representation of each target item is obtained.

[0049] In some embodiments, the candidate item aggregation page typically contains multiple target items, and these multiple target items may be of different categories; that is, the target items may be divided into multiple categories belonging to two different categories.

[0050] When the target items belong to at least two categories, in order to accurately obtain the second representation of the candidate item aggregation page, a possible implementation of using the graph model aggregation function to aggregate the representations of each target item to obtain the second representation is as follows: For each category, based on the target items contained in the category, use the graph model aggregation function to aggregate the representations of each target item under the corresponding category to obtain the second representation under the corresponding category.

[0051] In this example, based on the multiple categories to which the target items contained in the candidate item aggregation page belong, for each category, the aggregation function of the graph model is used to aggregate the representations of each target item under the corresponding category to obtain the second representation of the corresponding category of the candidate item aggregation page.

[0052] Step 104: If the similarity between the first representation and the second representation meets the first set condition, determine that the candidate item aggregation page will be sent as the recommended aggregation page to the device associated with the user to be recommended.

[0053] In this embodiment, when the target items on the candidate item aggregation page belong to at least two categories, it can be determined that the similarity between the second representation and the first representation under any category of the candidate item aggregation page meets the first set condition. If the similarity between the second representation and the first representation under any category of the candidate item aggregation page meets the first set condition, the candidate item aggregation page is sent as a recommended aggregation page to the device associated with the user to be recommended.

[0054] It is understandable that the first setting condition is set by staff based on actual needs. For example, the first setting condition could be that the similarity is greater than a preset similarity threshold.

[0055] This application provides a method for recommending aggregated pages. During the process of recommending aggregated pages to users, a trained graph model is obtained. The graph model includes multiple nodes, each corresponding to an item or an item aggregated page, used to indicate the representation of the corresponding item or item aggregated page. The representation is generated based on the behavioral correlation of items or item aggregated pages in historical access behavior, and the behavioral correlation indicates the co-occurrence relationship of items or item aggregated pages in historical access behavior. Based on the items already operated by the user to be recommended, nodes in the graph model are queried to obtain a first representation of the operated items. Based on multiple candidate item aggregated pages, nodes in the graph model are queried to obtain a second representation of each candidate item aggregated page. If the similarity between the first and second representations meets a first preset condition, the candidate item aggregated page is determined to be sent as the recommended aggregated page to the device associated with the user to be recommended. Therefore, based on the graph model, the goal of accurately recommending aggregated pages to users to be recommended is achieved, improving the accuracy of aggregated page recommendations.

[0056] Based on the above embodiments, in order to clearly understand how to train the graph model, the following will combine... Figure 2 The method of this embodiment will be further described exemplarily.

[0057] Figure 2 A flowchart illustrating a method for recommending aggregated pages provided in this application embodiment. Figure 2 .

[0058] like Figure 2 As shown, the method may include the following steps:

[0059] Step 201: Obtain the initial representation of each node in the graph model.

[0060] In this example, after the graph model is formed, each node in the graph model is randomly initialized to obtain an initial representation of the random configuration of each node in the graph model.

[0061] Step 202: Take any node in the graph model as the current node, and select the k-th nearest neighbor node of the current node from the graph model, where k is a positive integer with a fixed value.

[0062] It is understandable that k is determined by the staff based on actual needs, which will not be elaborated here.

[0063] In this example, based on historical access behavior, the nodes that have a connection edge with the current node are determined according to whether the current node has a co-occurrence relationship with any other node in the graph model. Based on the nodes that have a connection edge with the current node, the nearest neighbor nodes of the current node are obtained.

[0064] Step 203: Use an aggregation function to aggregate the initial representations of the k-th order nearest neighbor nodes and the initial representation of the current node, so as to use the updated representation of the current node.

[0065] For example, the aggregation function fagogre can be used to aggregate the initial representations of the k-th order nearest neighbors and the initial representation of the current node.

[0066] For example, the initial representation of each node in a graph model is:<e1,e2,…,eK> Where ei is the representation of the i-th node, K is the total number of nodes in the graph model, and i is a positive integer greater than 1. Randomly select the k-th nearest neighbor of the current node vi.<v1,v2,…,vk> Where k is a fixed positive integer, and the aggregation function fagogre is selected to aggregate the initial representations of the current node and its k-th nearest neighbors to obtain the updated representation of the current node, i.e.: e i =f aggre (e1,e2,…,eK),K∈N(i).

[0067] It should be noted that the specific value of k can be determined according to actual needs, and no limit is imposed here.

[0068] Step 204: Take at least one nearest neighbor node of the current node in the graph model as a positive sample node, and take at least one node other than the nearest neighbor node in the graph model as a negative sample node. Determine a loss function to indicate the difference between the first representation similarity and the second representation similarity based on the first representation similarity between the updated representation of the current node and the positive sample node, and based on the second representation similarity between the updated representation of the current node and the negative sample node.

[0069] As one implementation method, the inner product calculation method can be used to calculate the similarity between the updated representation of the current node and the positive sample node, so as to obtain the first representation similarity between the updated representation of the current node and the positive sample node.

[0070] As one implementation method, the inner product calculation method can be used to calculate the similarity between the updated representation of the current node and the negative sample node, so as to obtain the second representation similarity between the updated representation of the current node and the negative sample node.

[0071] For example, v is the k-th nearest neighbor of the current node u, i.e., a positive sample node, σ is the sigmoid function, and v n Let Q be the total number of negative sample nodes, and e represent the node representation. Then the loss function is...

[0072] Step 205: Based on the loss function, adjust the representation of each node in the graph model and adjust the parameters of the aggregation function.

[0073] In this example, the representations of each node in the graph model are adjusted based on a defined loss function, and the parameters of the aggregation function are tuned to complete the training of the graph model. This ensures that the distance between any two nearest neighbors in the graph model is close to the distance between randomly sampled negative sample nodes, thereby improving the accuracy of the graph model in recommending aggregation pages to users.

[0074] Based on the above embodiments, in order to clearly understand how to form a graph model based on the co-occurrence relationship of items or item aggregation pages and corresponding items or item aggregation pages in historical access behavior, the following will combine... Figure 3 The method of this embodiment will be further described exemplarily.

[0075] Figure 3 A flowchart illustrating a method for recommending aggregated pages provided in this application embodiment. Figure 3 .

[0076] like Figure 3 As shown, the method may include the following steps:

[0077] Step 301: Based on historical access behavior, identify the items and item aggregation pages where there are operational behaviors.

[0078] In this example, historical access behavior is used to collect historical operation behavior data, and based on this data, items and item aggregation pages with observed operations are identified.

[0079] It is understandable that historical access behavior can be directly obtained by the executing entity, and this is not limited here.

[0080] In this example, after determining that the historical operation behavior data contains information about items with operation behavior and item aggregation pages, one possible way to process the historical operation behavior data is to obtain the attribute information and time information of the items with operation behavior and their related information, as well as the item aggregation page with operation behavior and the items it contains.

[0081] It is understandable that the time information includes the time when the operation occurred.

[0082] It is understandable that after processing the historical operation data, the processed data can be stored in the pre-defined material library, which will not be elaborated here.

[0083] Step 302: Configure the node corresponding to the item or item aggregation page.

[0084] Step 303: For any two nodes, determine whether there is a co-occurrence relationship between the two nodes based on whether the operation behavior of the items or item aggregation pages corresponding to the two nodes meets the second set condition.

[0085] In this context, it can be understood that co-occurrence relationship is used to represent the behavioral correlation between items or item aggregation pages in historical access behavior, that is, there is a behavioral correlation between the operation behaviors of two nodes.

[0086] For example, determining whether there is a behavioral correlation between the operations of two nodes can also be done by setting other conditions. These other conditions could include the fact that the operations do not belong to the same user and that the purpose of the operations corresponds to the same item or item aggregation page.

[0087] It is understood that the second set of conditions includes that the operation behaviors belong to the same user and the time interval between operations is not greater than a threshold.

[0088] The threshold values ​​are set by staff based on actual needs, and will not be elaborated here.

[0089] In this example, after determining whether there is a co-occurrence relationship between two nodes based on whether the operation behavior of the items or item aggregation pages corresponding to the two nodes meets the second set condition, the method further includes: counting the co-occurrence frequency for the two nodes that have a co-occurrence relationship; determining the weight of the edge between the two nodes based on the co-occurrence frequency; wherein the weight is used to aggregate the representation of the connected nodes in the aggregation function.

[0090] It is understandable that calculating the co-occurrence frequency can determine the weight of the edge between two nodes.

[0091] For example, extract all items, item aggregation pages, and the times of operation from historical operation data to form a sequence, i.e., U = {(v1,t1),(v2,t2),…,(vk,tk)}, where v is the node corresponding to the configured item or item aggregation page, t is the time of operation, and k is the total number of nodes corresponding to the configured item or item aggregation page. The co-occurrence relationship between nodes constructed in the sequence is S = (P,N), where P is a set of pairs containing two nodes, i.e., (vi,vj). These two nodes must satisfy two conditions to form a pair: 1) vi and vj are adjacent in the same user's operation; 2) the time interval between the two nodes' operations does not exceed 15 minutes, i.e., tj - ti < 15min. N represents the frequency information of each pair appearing in set P, N = ((Vi,Vj),K), where K represents the co-occurrence frequency of each pair appearing in set P.

[0092] It should be noted that the second condition setting refers to the two conditions that need to be met for the two nodes to form a pair in the above example. The 15 minutes in the above example can be represented as a threshold.

[0093] Step 304: Connect the two nodes that have a co-occurrence relationship to obtain the graph model.

[0094] For example, a graph model G = (V, E, W) can be constructed based on whether nodes have co-occurrence relationships. Here, V represents the nodes in the graph model, i.e., the set of items and item aggregation pages. E indicates whether there are connecting edges between nodes, where the presence of connecting edges indicates whether the corresponding nodes have a co-occurrence relationship. If there are connecting edges between corresponding nodes, they have a co-occurrence relationship; if there are no connecting edges, they do not. W represents the weight of the edges.

[0095] In this example, to prevent nodes and connecting edges with high co-occurrence frequency from being trained frequently, resulting in insufficient training of nodes with low co-occurrence frequency, the weights between connecting edges can be exponentially smoothed.

[0096] In some embodiments of this application, in order to accumulate enough data for graph model training, information on newly input items or item aggregation pages with operational behavior can be input into the graph model for training as quickly as possible. This includes: obtaining the representation of the newly input items or item aggregation pages, based on the representation of each node in the graph model, and training the representation of the newly input items or item aggregation pages using an aggregation function, thereby achieving better recommendation results.

[0097] To implement the above embodiments, this application also proposes a recommendation device for aggregated pages.

[0098] Figure 4 This is a schematic diagram of a recommendation device for an aggregated page provided in an embodiment of this application.

[0099] like Figure 4 As shown, the recommendation device 400 of the aggregation page includes a first acquisition module 401, a first characterization module 402, a second characterization module 403, and a recommendation module 404:

[0100] The first acquisition module 401 is used to acquire a trained graph model; wherein, the graph model includes multiple nodes, each node corresponds to an item or item aggregation page, and is used to indicate the representation of the corresponding item or item aggregation page; the representation is generated based on the behavioral correlation of the item or item aggregation page in historical access behavior, and the behavioral correlation is used to indicate that the item or item aggregation page has a co-occurrence relationship in historical access behavior;

[0101] The first representation module 402 is used to query the nodes in the graph model based on the items already operated by the user to be recommended, so as to obtain the first representation of the operated items.

[0102] The second representation module 403 is used to query the nodes in the graph model based on the candidate item aggregation page to obtain the second representation of each candidate item aggregation page;

[0103] The recommendation module 404 is used to determine, when the similarity between the first representation and the second representation meets the first set condition, to send the candidate item aggregation page as the recommended aggregation page to the device associated with the user to be recommended.

[0104] In one embodiment, the second characterization module 403 includes:

[0105] The query unit 4031 is used to query the node of the corresponding item aggregation page in the graph model for any candidate item aggregation page;

[0106] The representation unit 4032 is used to query the nodes of the corresponding items in the graph model based on the target items contained in the candidate item aggregation page if there are no nodes of the corresponding candidate item aggregation page in the nodes of the corresponding item aggregation page, so as to obtain the representation of each target item.

[0107] Aggregation unit 4033 is used to aggregate the representations of each target item using the aggregation function of the graph model to obtain a second representation.

[0108] In one embodiment, the target items are multiple and belong to at least two categories. The aggregation unit 4033 is specifically used for:

[0109] For each category, based on the target items contained in the category, the aggregation function of the graph model is used to aggregate the representations of each target item in the corresponding category to obtain the second representation of the corresponding category.

[0110] In one embodiment, based on the similarity between the first representation and the second representation, it is determined whether to send the candidate item aggregation page as a recommended aggregation page to the device associated with the user to be recommended. The recommendation module 404 is specifically used for:

[0111] If the similarity between the second representation and the first representation under any category of the candidate item aggregation page meets the first set condition, then the candidate item aggregation page will be sent as the recommended aggregation page to the device associated with the user to be recommended.

[0112] Based on the above embodiments, this application also provides another possible implementation of the recommendation device for the aggregation page. Figure 5This is a schematic diagram of another aggregation page recommendation device provided in an embodiment of this application. Based on the previous embodiment, the aggregation page recommendation device 500 further includes: a second acquisition module 505, a selection module 506, an updated representation module 507, a loss function module 508, and an adjustment module 509.

[0113] The second acquisition module 505 is used to acquire the initial representation of each node in the graph model;

[0114] The selection module 506 is used to select any node in the graph model as the current node and select the k-th nearest neighbor node of the current node from the graph model, where k is a positive integer with a fixed value.

[0115] The updated representation module 507 is used to aggregate the initial representations of the k-th order nearest neighbor nodes and the initial representation of the current node using an aggregation function, so as to serve as the updated representation of the current node.

[0116] The loss function module 508 is used to take at least one nearest neighbor node of the current node in the graph model as a positive sample node and at least one node other than the nearest neighbor node in the graph model as a negative sample node. Based on the first representation similarity between the updated representation of the current node and the positive sample node, and based on the second representation similarity between the updated representation of the current node and the negative sample node, a loss function is determined to indicate the difference between the first representation similarity and the second representation similarity.

[0117] The adjustment module 509 is used to adjust the representation of each node in the graph model based on the loss function, and to adjust the parameters of the aggregation function.

[0118] In one embodiment, the initial representation of each node in the graph model is obtained. The second acquisition module 505 is specifically used for:

[0119] The initial representation of each node in a stochastic graph model; or,

[0120] The representations of each node determined in the previous parameter adjustment are used as the initial representations.

[0121] In one embodiment, before obtaining the initial representation of each node in the graph model, the recommendation device 500 of the aggregation page further includes a first determining module 510, a configuration module 511, a second determining module 512, and a graph model module 513:

[0122] The first determination module 510 is used to determine the items and item aggregation pages where there are operational behaviors based on historical access behaviors;

[0123] Configuration module 511 is used to configure the nodes corresponding to items or item aggregation pages;

[0124] The second determining module 512 is used to determine whether there is a co-occurrence relationship between any two nodes based on whether the operation behavior of the corresponding items or item aggregation pages of the two nodes meets the second set conditions.

[0125] Graph model module 513 is used to connect two nodes with a co-occurrence relationship to obtain a graph model.

[0126] In one embodiment, after determining whether there is a co-occurrence relationship between two nodes based on whether the operation behavior of the items or item aggregation pages corresponding to the two nodes meets the second set condition, the second determining module 512 is specifically used for:

[0127] For two nodes that co-occur, count the frequency of co-occurrence.

[0128] The weight of the edge between two nodes is determined based on the co-occurrence frequency; the weight is used to aggregate the representation of the connected nodes in the aggregation function.

[0129] It should be noted that the foregoing explanation of the method embodiment for recommending aggregation pages also applies to the aggregation page recommendation device of this embodiment, and will not be repeated here.

[0130] This application provides an aggregation page recommendation device that acquires a trained graph model. The graph model includes multiple nodes, each corresponding to an item or an item aggregation page, used to indicate the representation of the corresponding item or item aggregation page. The representation is generated by the behavioral correlation of items or item aggregation pages in historical access behavior, and the behavioral correlation indicates the co-occurrence relationship of items or item aggregation pages in historical access behavior. Based on the items already operated by the user to be recommended, nodes in the graph model are queried to obtain a first representation of the operated items. Based on candidate item aggregation pages, nodes in the graph model are queried to obtain a second representation of each candidate item aggregation page. If the similarity between the first and second representations meets a first preset condition, the candidate item aggregation page is determined to be sent as the recommended aggregation page to the device associated with the user to be recommended. Therefore, based on the graph model, the purpose of recommending aggregation pages to the user to be recommended is achieved, improving the accuracy of aggregation page recommendations and enriching the user experience.

[0131] Figure 6 This is a schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application. For example... Figure 6 As shown, the electronic device includes a memory 61 and a processor 62. Additionally, the electronic device also includes a power supply component 63 and a communication component 64.

[0132] Memory 61 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device.

[0133] The memory 61 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0134] Communication component 64 is used for data transmission with other devices.

[0135] Processor 62 executes computer instructions stored in memory 61.

[0136] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program. When the computer-readable storage medium stores a computer program, and the computer program is executed by one or more processors, it causes one or more processors to perform... Figure 1 Each step in the method embodiment.

[0137] Accordingly, embodiments of this application also provide a computer program product, which includes a computer program / instructions, and the computer program / instructions are executed by a processor. Figure 1 Each step in the method embodiment.

[0138] The above Figure 6 The communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA) technology, Ultra-Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0139] The above Figure 6 The power supply component provides power to the various components of the device in which it resides. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which it resides.

[0140] The aforementioned electronic devices also include a display screen.

[0141] The display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touchscreen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0142] Those skilled in the art will understand that the embodiments of this application can be provided as methods or systems. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0144] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0145] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0146] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0147] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A recommendation method for an aggregation page, characterized in that, Includes the following steps: Obtain a trained graph model; wherein the graph model includes multiple nodes, each node corresponding to an item or item aggregation page, used to indicate the representation of the corresponding item or item aggregation page; the representation is generated based on the behavioral correlation of the item or item aggregation page in historical access behavior, the behavioral correlation is used to indicate that the item or item aggregation page has a co-occurrence relationship in historical access behavior; Based on the items already operated by the user to be recommended, query the nodes in the graph model to obtain the first representation of the operated items; Based on multiple candidate item aggregation pages, the nodes in the graph model are queried to obtain a second representation of each candidate item aggregation page; If the similarity between the first representation and the second representation meets the first set condition, it is determined that the candidate item aggregation page will be sent as a recommended aggregation page to the device associated with the user to be recommended.

2. The method according to claim 1, characterized in that, The step of querying nodes in the graph model based on multiple candidate item aggregation pages to obtain a second representation of each candidate item aggregation page includes: For any candidate item aggregation page, query the node corresponding to the item aggregation page in the graph model; If there is no node corresponding to the candidate item aggregation page in the node of the corresponding item aggregation page, then according to the target items contained in the candidate item aggregation page, query the node of the corresponding item in the graph model to obtain the representation of each target item; The aggregation function of the graph model is used to aggregate the representations of each target item to obtain the second representation.

3. The method according to claim 2, characterized in that, The target items are multiple and belong to at least two categories; the aggregation function of the graph model is used to aggregate the representations of each target item to obtain the second representation, including: For each category, based on the target items contained in the category, the aggregation function of the graph model is used to aggregate the representations of each target item in the corresponding category to obtain the second representation of the corresponding category.

4. The method according to claim 3, characterized in that, The step of determining to send the candidate item aggregation page as a recommended aggregation page to the device associated with the user to be recommended, when the similarity between the first representation and the second representation meets a first preset condition, includes: If the similarity between the second representation and the first representation under any category of the candidate item aggregation page meets the first set condition, then the candidate item aggregation page will be sent as the recommended aggregation page to the device associated with the user to be recommended.

5. The method according to any one of claims 1-4, characterized in that, Before obtaining the trained graph model, the process also includes: Obtain the initial representation of each node in the graph model; Take any node in the graph model as the current node, and select the k-th order nearest neighbor node of the current node from the graph model, where k is a positive integer with a fixed value; An aggregation function is used to aggregate the initial representations of the k-th order nearest neighbor nodes and the initial representation of the current node, so as to serve as the updated representation of the current node. The graph model takes at least one nearest neighbor node of the current node as a positive sample node and at least one node other than the nearest neighbor node as a negative sample node. Based on the first representation similarity between the updated representation of the current node and the positive sample node, and based on the second representation similarity between the updated representation of the current node and the negative sample node, a loss function is determined to indicate the difference between the first representation similarity and the second representation similarity. Based on the loss function, the representation of each node in the graph model is adjusted, and the parameters of the aggregation function are adjusted.

6. The method according to claim 5, characterized in that, The process of obtaining the initial representation of each node in the graph model includes: Randomly configure the initial representation of each node in the graph model; or, The representation of each node determined by the previous parameter adjustment is used as the initial representation.

7. The method according to claim 5, characterized in that, Before obtaining the initial representation of each node in the graph model, the process also includes: Based on historical access behavior, identify items and item aggregation pages where actions have been performed; Configure the node corresponding to the item or the item aggregation page; For any two nodes, determine whether there is a co-occurrence relationship between the two nodes based on whether the operation behavior of the items corresponding to the two nodes or the item aggregation page meets the second set condition; Connect the two nodes that have the co-occurrence relationship to obtain the graph model.

8. The method according to claim 7, characterized in that, The second set condition includes that the operation belongs to the same user and the time interval between operations is not greater than a threshold.

9. The method according to claim 7, characterized in that, After determining whether there is a co-occurrence relationship between the two nodes based on whether the operation behavior of the items corresponding to the two nodes or the item aggregation page meets the second set condition, the method further includes: For two nodes that co-occur, count the frequency of co-occurrence. The weight of the edge between the two nodes is determined based on the co-occurrence frequency; wherein the weight is used by the aggregation function to aggregate the representation of the connected nodes.

10. A recommendation device for an aggregated page, characterized in that, include: The first acquisition module is used to acquire a trained graph model; wherein the graph model includes multiple nodes, each node corresponds to an item or an item aggregation page, and is used to indicate the representation of the corresponding item or item aggregation page; the representation is generated based on the behavioral correlation of the item or the item aggregation page in historical access behavior, and the behavioral correlation is used to indicate that the item or the item aggregation page has a co-occurrence relationship in historical access behavior; The first representation module is used to query the nodes in the graph model based on the items already operated by the user to be recommended, so as to obtain the first representation of the operated items; The second representation module is used to query the nodes in the graph model based on multiple candidate item aggregation pages to obtain the second representation of each candidate item aggregation page. The recommendation module is used to determine, when the similarity between the first representation and the second representation meets a first preset condition, to send the candidate item aggregation page as a recommended aggregation page to the device associated with the user to be recommended.

11. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the steps of the method as described in any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-9.

Images