Promotion content recall method and device and computer storable medium

By integrating promotional content delivery conditions into a target graph structure using user and item vectors, the method enhances recall efficiency and accuracy, addressing inefficiencies in existing online advertising systems.

CN119067724APending Publication Date: 2024-12-03BEIJING WODONG TIANJUN INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310637155.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-31
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing online advertising systems suffer from inefficiencies and inaccuracies in recalling promotional content due to the separation of algorithm capabilities across different data domains, leading to wasted computing power and reduced revenue.

Method used

A method and device that integrate promotional content delivery conditions into a target graph structure, using user and item vectors to directly recall relevant content based on user information, enhancing recall efficiency and accuracy by associating algorithm capabilities with promotional content delivery data.

Benefits of technology

Improves the efficiency, accuracy, and reliability of promotional content recall by integrating promotional content delivery conditions into a graph structure, reducing resource waste and enhancing user experience and advertiser interests.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067724A_ABST
    Figure CN119067724A_ABST
Patent Text Reader

Abstract

The invention relates to a promotion content recall method and device and a computer readable storage medium. The promotion content recall method comprises the steps that a target graph structure is obtained, the target graph structure comprises a promotion content node representing promotion content delivery data, an article node representing an article vector and a first edge between the promotion content node and the article node which have an incidence relation, and the attribute of the first edge comprises a promotion content delivery condition; in response to the online request of the user, determining a user vector of the user according to the user information of the user; according to a distance between an article vector corresponding to an article node in the target graph structure and a user vector, determining promotion content delivery data corresponding to a promotion content node, meeting a promotion content delivery condition, of a user from the promotion content nodes in the target graph structure, and taking the promotion content delivery data as candidate promotion content; and according to the candidate promotion content, determining target promotion content for recall of the promotion content.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method and device for recalling promotional content, and a computer storable medium. Background Art

[0002] Online advertising is an important revenue source for the Internet. On the one hand, it must ensure the user experience of the platform, and on the other hand, it must protect the interests of advertisers, which has given rise to a complex advertising search architecture. In the field of e-commerce, the advertising search system includes two important functions: user and product correlation modeling; and advertising placement rule filtering.

[0003] In the related art, the product recall module first recalls a candidate set of products with high relevance to the user from the product set based on the user vector; then the advertising unit recall module determines the advertising units corresponding to the candidate set of products, and filters them according to the user information and the advertising delivery rules set in the advertising unit and the platform rules of the advertising delivery platform to obtain the advertising units used for advertising recall. Summary of the invention

[0004] In the related technology, the online advertising retrieval system logically separates the algorithm capability from the advertising delivery. In the related product recall module, the products are filtered according to the platform rules due to the limitations of recall diversity, time consumption, bandwidth, etc., and then the advertising unit recall module performs advertising unit attachment and advertising delivery rule filtering. The two modules are oriented to different data domains, which will lead to waste of computing power and reduce the overall revenue of platform advertising. For example, the related product fails to attach an advertisement, or the attached advertisement is filtered by the advertising delivery rules, resulting in invalid recalls and insufficient number of output advertising unit sets, which in turn leads to a decrease in the accuracy and reliability of advertising recall.

[0005] In view of the above technical problems, the present disclosure proposes a solution that can improve the recall efficiency, accuracy and reliability of promotional content.

[0006] According to a first aspect of the present disclosure, a method for recalling promotional content is provided, comprising: obtaining a target graph structure, wherein the target graph structure comprises a promotional content node representing promotional content delivery data, an item node representing an item vector, and a first edge between the promotional content node and the item node having an associated relationship, wherein an attribute of the first edge comprises a promotional content delivery condition; in response to an online request of a user, determining a user vector of the user according to user information of the user; determining, from the promotional content nodes in the target graph structure, the promotional content delivery data corresponding to the promotional content node that satisfies the promotional content delivery condition for the user, as candidate promotional content, according to a distance between the item vector corresponding to the item node in the target graph structure and the user vector; and determining, based on the determined candidate promotional content, a target promotional content for recalling the promotional content.

[0007] In some embodiments, the target graph structure also includes a second edge between different item nodes having an associated relationship. According to the distance between the item node in the target graph structure and the user vector, the promotional content delivery data corresponding to the promotional content node for which the user meets the promotional content delivery conditions is determined from the promotional content nodes in the target graph structure as candidate promotional content, including: selecting, from the candidate item nodes, a candidate item node whose distance between the item vector and the user vector is less than a distance threshold as a reference item node; according to the target graph structure, determining an item node having a second edge with the reference item node as a candidate item node; according to the target graph structure, determining the promotional content delivery data corresponding to the promotional content node having a first edge with the reference item node and for which the user meets the promotional content delivery conditions corresponding to the first edge as the candidate promotional content; and repeatedly performing the operations of selecting reference item nodes, determining candidate item nodes, and determining candidate promotional content until there are no candidate item nodes in the target graph structure.

[0008] In some embodiments, in an initial state, the candidate item nodes include one or more item nodes randomly selected from the target graph structure.

[0009] In some embodiments, obtaining a target graph structure includes: obtaining item data including multiple items, multiple promotional content delivery data and the corresponding relationship between them; determining an item vector for each item; for each item, determining the distance between the item vector of each item and the item vectors of other items; adding a first edge to a promotional content node corresponding to each promotional content delivery data, for connecting the item node corresponding to the item corresponding to each promotional content delivery data; determining a promotional content delivery condition corresponding to each promotional content delivery data, which is an attribute of the first edge of each promotional content delivery data; and adding a second edge to the item node corresponding to each item according to the distance between the item vector of each item and the item vectors of other items.

[0010] In some embodiments, there are multiple other items, and adding a second edge to the item node corresponding to each item based on the distance between the item vector of each item and the item vectors of other items includes: generating an item relationship pair based on each item node and each other item node; constructing the second edge between the item nodes corresponding to other items in the item relationship pair and the item node corresponding to each item in order of distance from small to large, wherein, in the process of constructing the second edge of the item node corresponding to each item, the angle between the newly added second edge and the already added second edge is greater than an angle threshold.

[0011] In some embodiments, there are multiple candidate promotional contents determined, and based on the determined candidate promotional contents, target promotional content is determined for promotional content recall, including: obtaining weights of promotional content delivery data corresponding to multiple candidate promotional contents; and based on at least one candidate promotional content having a weight greater than a weight threshold, determining the target promotional content for promotional content recall.

[0012] In some embodiments, determining the target promotional content based on at least one candidate promotional content having a weight greater than a weight threshold includes: selecting, from at least one candidate promotional content having a weight greater than a weight threshold, a candidate promotional content that meets a promotional content filtering condition of a promotional content delivery platform as the target promotional content.

[0013] In some embodiments, the user information includes user portrait information and user historical behavior information related to the item, and determining the user vector of the user based on the user information of the user includes: determining the user vector of the user based on the user portrait information and the user historical behavior information using the user sub-network in the dual-tower model.

[0014] According to a second aspect of the present disclosure, a device for recalling promotional content is provided, comprising: an acquisition module, configured to acquire a target graph structure, wherein the target graph structure comprises a promotional content node representing promotional content delivery data, an item node representing an item vector, and a first edge between the promotional content node and the item node having an associated relationship, wherein the attribute of the first edge comprises a promotional content delivery condition; a first determination module, configured to determine a user vector of the user according to user information of the user in response to an online request of the user; a second determination module, configured to determine, according to a distance between an item vector corresponding to an item node in the target graph structure and the user vector, promotional content delivery data corresponding to a promotional content node that satisfies the promotional content delivery condition of the user from the promotional content nodes in the target graph structure as candidate promotional content; and a third determination module, configured to determine, according to the determined candidate promotional content, a target promotional content for recalling the promotional content.

[0015] According to a third aspect of the present disclosure, a promotional content recall device is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the promotional content recall method described in any of the above embodiments based on instructions stored in the memory.

[0016] According to a fourth aspect of the present disclosure, a computer storable medium is provided, on which computer program instructions are stored, and when the instructions are executed by a processor, the method for recalling promotional content described in any of the above embodiments is implemented.

[0017] In the above embodiments, the recall efficiency, accuracy and reliability of promotional content can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0019] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0020] Figure 1 is a flow chart showing a method for recalling promotional content according to some embodiments of the present disclosure;

[0021] Figure 2 is a flowchart illustrating a method for recalling promotional content according to some embodiments of the present disclosure;

[0022] Figure 3 is a block diagram showing a device for recalling promotional content according to some embodiments of the present disclosure;

[0023] Figure 4is a block diagram showing a device for recalling promotional content according to other embodiments of the present disclosure;

[0024] Figure 5 is a block diagram illustrating a computer system for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0025] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0026] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0027] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0028] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0029] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0030] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0031] Figure 1 is a flow chart illustrating a method for recalling promotional content according to some embodiments of the present disclosure.

[0032] like Figure 1 As shown, the promotion content recall method includes steps S110 to S140. For example, the promotion content recall method is executed by a promotion content recall device.

[0033] In step S110, a target graph structure is obtained. The target graph structure includes a promotion content node representing promotion content delivery data, an item node representing an item vector, and a first edge between a promotion content node and an item node having an associated relationship. The attribute of the first edge includes a promotion content delivery condition. In some embodiments, the target graph structure also includes a second edge between different item nodes having an associated relationship.

[0034] In some embodiments, the promotional content includes advertisements, and the items may include commodities or products on the e-commerce platform. The promotional content delivery data includes, for example, advertisement delivery data. The advertisement delivery data mainly refers to the advertisement unit information. An advertisement unit information is set by the advertiser, and specifies the corresponding item set and the advertisement delivery rule set. The advertisement unit and the item are in a many-to-many relationship. For example, the item is represented by Item, and Item is the granularity of the algorithm recall, generally SKU (Stock Keeping Unit, the smallest available unit for inventory control), material, etc.

[0035] In some embodiments, the target graph structure may be constructed through the following steps 1) to 5).

[0036] In step 1), item data including a plurality of items, a plurality of promotional content delivery data and the corresponding relationship between them are obtained. The relationship between the promotional content delivery data and the items may be many-to-many.

[0037] In step 2), the item vector of each item is determined. Taking the item represented by Item as an example, the item vector is an Item vector, which can be expressed as ItemEmbedding.

[0038] In step 3), a first edge is added to the promotion content node corresponding to each piece of promotion content delivery data, for connecting the item node corresponding to the item corresponding to each piece of promotion content delivery data.

[0039] In step 4), the promotion content delivery condition corresponding to each piece of promotion content delivery data is determined, which is the attribute of the first edge of each piece of promotion content delivery data.

[0040] In some embodiments, constructing the target graph structure also includes step 5) and step 6).

[0041] In step 5), for each item, the distance between the item vector of each item and the item vectors of other items is determined.

[0042] In step 6), a second edge is added to the item node corresponding to each item according to the distance between the item vector of each item and the item vectors of other items.

[0043] In some embodiments, there are multiple other items. In this case, the above step 6) can be performed in the following manner.

[0044] First, an item relationship pair is generated based on each item node and each other item node.

[0045] Then, in order of distance from small to large, the second edge between the item node corresponding to the other items in the item relationship pair and the item node corresponding to each item is constructed. In the process of constructing the second edge of the item node corresponding to each item, the angle between the newly added second edge and the already added second edge is greater than the angle threshold.

[0046] In step S120, in response to the user's online request, the user vector of the user is determined according to the user information of the user. In some embodiments, a user's visit to the e-commerce platform is an online request.

[0047] In some embodiments, the user information includes user portrait information and user historical behavior information related to the item. In this case, determining the user vector of the user according to the user information of the user includes: determining the user vector of the user according to the user portrait information and the user historical behavior information using the user sub-network in the double tower model.

[0048] In some embodiments, during offline training, the dual-tower model uses the user sub-network and the item sub-network to extract user vectors and item vectors respectively. The two sub-networks are exported separately. The user vector can be determined by forward calculation of the model. During the forward calculation of the model, the user portrait information and the filtered historical behavior information are first processed into model features in a specified format, and the user sub-network is used for forward calculation to obtain the user vector, which is used for vectorized retrieval, that is, to determine the candidate promotion content node. For example, the data for offline training is generally the portrait data and behavior data of all users on the e-commerce platform in the past period of time. After collecting user information, the data will be de-identified through technical means for privacy protection.

[0049] In some embodiments, the user information in the online request can be filtered according to the features used by the model for determining the user vector during offline training (for example, using the user's gender as a feature after hash calculation). For example, the filtering basis includes information such as time, frequency, and behavior category. Different models will select different user information, such as a simple filtering logic that only selects the 20 items that the user has browsed most recently.

[0050] In some embodiments, in response to a user's online request, the user portrait information and user historical behavior information are obtained by accessing the user service module. The user portrait information includes, for example, age, gender, etc. The user historical behavior information includes, for example, historically browsed items, added items, etc. The user service module is used to obtain user information.

[0051] In step S130, based on the distance between the item vector corresponding to the item node in the target graph structure and the user vector, the promotional content delivery data corresponding to the promotional content node for which the user meets the promotional content delivery conditions is determined from the promotional content nodes in the target graph structure as candidate promotional content.

[0052] In some embodiments, taking the case where the target graph structure also includes a second edge between different item nodes having an association relationship as an example, the above step S130 may be implemented in the following manner.

[0053] First, from the candidate item nodes, the candidate item nodes whose distance between the item vector and the user vector is less than the distance threshold are selected as reference item nodes. In the initial state, the candidate item nodes include one or more item nodes randomly selected from the target graph structure.

[0054] In some embodiments, from the candidate item nodes, the candidate item node with the smallest distance between the item vector and the user vector is selected as the reference item node. In this case, the distances between the item vector and the user vector in the candidate item nodes are sorted in ascending order, and the distance with the second largest sorting result is determined as the distance threshold.

[0055] Secondly, according to the target graph structure, the item nodes having the second edge with the reference item node are determined as candidate item nodes.

[0056] Then, according to the target graph structure, promotional content delivery data corresponding to a promotional content node having a first edge with the reference item node and whose user satisfies the promotional content delivery condition corresponding to the first edge is determined as candidate promotional content.

[0057] Finally, the operations of selecting reference item nodes, determining candidate item nodes, and determining candidate promotion content are repeated until there are no candidate item nodes in the target graph structure.

[0058] In the above embodiment, by further building associations between items, the recall efficiency, accuracy and reliability of promotional content can be further improved.

[0059] In step S140, target promotion content is determined based on the determined candidate promotion content for recalling the promotion content.

[0060] In some embodiments, there are multiple candidate promotion content nodes determined. In this case, the above step S140 can be implemented in the following manner.

[0061] First, the weights of the promotion content delivery data corresponding to a plurality of candidate promotion contents are obtained.

[0062] Then, according to at least one candidate promotion content whose weight is greater than a weight threshold, a target promotion content is determined for promotion content recall.

[0063] In some embodiments, candidate promotional content that meets a promotional content filtering condition of the promotional content delivery platform may be selected from at least one candidate promotional content having a weight greater than a weight threshold as the target promotional content.

[0064] In the above embodiment, the association between items and promotional content delivery data is constructed through a graph structure, and the promotional content delivery conditions of each promotional content delivery data are integrated into the graph structure, and then the user vector is distance-associated with the item vector, and the promotional content delivery conditions are integrated when determining candidate promotional content, thereby sinking the association between algorithm capabilities and promotional content delivery information into the graph structure index, and directly recalling promotional content units that meet the promotional content delivery rules through user vectors and / or user information during online retrieval, avoiding platform resource waste and interest damage due to different data domains, and improving the recall efficiency, accuracy and reliability of promotional content.

[0065] The following will be combined Figure 2 A method for recalling promotional content in some embodiments of the present disclosure is described.

[0066] Figure 2 The figure is a flowchart showing a method for recalling promotional content according to some embodiments of the present disclosure.

[0067] like Figure 2 As shown, the recall method of promoted content includes graph index construction, request parsing, model forward calculation and vectorized retrieval (i.e., vectorized graph index).

[0068] In the construction of graph index, for example, it can be based on the SSG (Satellite System Graph) graph index algorithm, and it can also be compatible with various graph algorithms in practical applications.

[0069] The input of the graph construction algorithm includes the item vector data and advertising data trained offline using the machine learning algorithm. SSG graph construction usually presets an angle setting value, which is also used as the input of the graph construction algorithm. The output of the graph construction algorithm is the graph structure data that can be used for retrieval. The graph construction process is mainly divided into building item node-item node (Item-Item) connection edges and building item node-advertisement node (Item-Ad) connection edges.

[0070] In some embodiments, taking Item as an example, the item node-item node connection edge can be constructed in the following manner.

[0071] Step (1), loop steps (2)-(4) for each item, and the corresponding item is denoted as itemA.

[0072] Step (2), calculate the distance between itemA and all other item nodes and generate itemA-item relationship pairs.

[0073] Step (3), sort the itemA-item relationship pairs in ascending order according to the distance.

[0074] Step (4), according to the sorted itemA-item relationship pairs, add edges to itemA in sequence. The adding process needs to meet the following conditions: the angle between the newly added edge and the added edge must be greater than the angle setting value. The maximum angle setting value is 180 degrees.

[0075] In some embodiments, taking Item as an example, the item node-advertisement node connection edge can be constructed in the following manner.

[0076] For each advertisement information, loop through the item set corresponding to the advertisement and construct a connection edge. The connection edge attribute needs to identify the connection as Item-Ad and store the advertisement unit information. The advertisement unit information includes promotion content delivery conditions such as advertisement targeting or filtering information.

[0077] In some embodiments, the graph structure data is represented by a map, and the key corresponds to the starting point of the connection edge. The value has four attributes: the first attribute is the end point of the connection edge; the second attribute indicates the distance; the third attribute indicates whether the corresponding edge is Item-Item (value is 0) or Item-Ad (value is 1); when the corresponding edge is Item-Ad, the fourth attribute indicates the unit information of the advertisement, including advertisement targeting or filtering information. Ad targeting information is, for example, a whitelist, and advertisement filtering information is, for example, a blacklist.

[0078] The above method is used to complete the graph construction index. Figure 2 The vectorized graph index shown.

[0079] like Figure 2 As shown in FIG, in the request parsing stage, the input of the entire recall method includes the user's online request. By parsing the user's online request, the user's user profile information and user historical behavior information are obtained.

[0080] like Figure 2 As shown, in the forward calculation stage of the model, based on the user portrait information and the user historical behavior information, the user vector of the user is obtained through the forward calculation of the model, which is used to vectorize the graph index.

[0081] like Figure 2As shown, in the vectorized retrieval (vectorized graph index) stage, a set of ad units that can be placed is obtained through vectorized recall and advertising placement rule filtering, namely the aforementioned candidate promotional content.

[0082] The input of vectorized search includes: the constructed graph structure data, the user vector generated by the user sub-network, the candidate pool size, user information and the starting search point, and its output is a specified number (K, K is a positive integer) of advertising units. The logic of vectorized search is as follows:

[0083] a) Randomly set an Item node as the starting query point, denoted as ItemA. In practical applications, multiple starting query points can be set, and steps c)-f) are performed in parallel at different query points;

[0084] b) Loop c)-f) until all nodes in the candidate pool are in the checked nodes;

[0085] c) Initialize the advertisement set Ads, the candidate pool S, the checked node set CN, and add ItemA to the candidate pool S;

[0086] d) Select the first item in the candidate pool that does not appear in the checked set (you can add a distance threshold from the user vector as a filtering condition), add it to the checked set, and record it as Item B;

[0087] e) Find all the connection points of ItemB from the graph structure data, and loop operations i)-iii):

[0088] i) If ItemB has been checked, skip it;

[0089] ii) When the Node is an Item node, it is added to the candidate pool;

[0090] iii) When the Node is an Ad node and the user meets the targeting conditions of the advertisement, the advertisement unit corresponding to the Ad node is added to the advertisement result set;

[0091] f) Sort all items in the candidate pool in ascending order according to the distance from the user vector. If the number is greater than the size of the candidate pool, truncate;

[0092] g) Sort the advertising result set in descending order according to the weight of the advertisement, then truncate it according to the set TopK, and finally return the result.

[0093] In the above embodiment, the advertising unit set is directly recalled based on the A-Star algorithm, which realizes fast and efficient retrieval in the vectorized graph index. The retrieval process adds the judgment of node attributes on the basis of A-Star, unifies the item retrieval and advertising attachment logic, and avoids the attachment failure in the two-stage recall (first recall the item, and then use the item to attach the advertisement) and other problems.

[0094] In some embodiments, Figure 2 As shown, the method for recalling promotional content may also include platform rule filtering, that is, using the rules set by the promotional content delivery platform to filter the acquired set of ad units that can be delivered, and output the recalled ad unit set. For example, the rules set by the promotional content delivery platform are used to ensure the experience and interests of users and advertisers, including page deduplication, overbroadcast control, etc.

[0095] In the above embodiment, the association between algorithm capabilities and advertising placement is sunk to the index construction stage. During online retrieval, the advertising unit set can be directly recalled through vectorized retrieval. Compared with the existing implementation, it avoids the waste of computing power and damage to platform interests caused by the different data domains caused by first recalling the item by the algorithm and then attaching advertisements through the item. It is a more advanced advertising retrieval concept. In terms of implementation, the vectorized graph retrieval scheme proposed by the present invention makes full use of the characteristics of the graph, and can efficiently complete the retrieval using the improved version of the A-Star algorithm (a direct search algorithm for solving the shortest path in a static road network). It has extremely high feasibility and can improve the recall efficiency, accuracy and reliability of the promotional content.

[0096] The above different embodiments complement and support each other.

[0097] Figure 3 is a block diagram showing a device for recalling promotional content according to some embodiments of the present disclosure.

[0098] like Figure 3 As shown, the promotion content recall device 3 includes an acquisition module 31 , a first determination module 32 , a second determination module 33 and a third determination module 34 .

[0099] The acquisition module 31 is configured to acquire a target graph structure, wherein the target graph structure includes a promotion content node representing promotion content delivery data, an item node representing an item vector, and a first edge between the promotion content node and the item node having an associated relationship, and the attribute of the first edge includes a promotion content delivery condition, such as executing Figure 1 Step S110 is shown. In some embodiments, the target graph structure further includes a second edge between different item nodes having an association relationship.

[0100] In some embodiments, taking the target graph structure including a promotional content node, an item node and a first edge and a second edge as an example, the acquisition module 31 is also configured to acquire item data including multiple items, multiple promotional content delivery data and the correspondence between them; determine the item vector of each item; for each item, determine the distance between the item vector of each item and the item vectors of other items; add a first edge to the promotional content node corresponding to each promotional content delivery data, for connecting the item node corresponding to the item corresponding to each promotional content delivery data; determine the promotional content delivery condition corresponding to each promotional content delivery data, which is the attribute of the first edge of each promotional content delivery data; and add a second edge to the item node corresponding to each item according to the distance between the item vector of each item and the item vectors of other items.

[0101] In some embodiments, there are multiple other items, and the acquisition module 31 is further configured to generate an item relationship pair based on each item node and each other item node; construct the second edge between the item nodes corresponding to the other items in the item relationship pair and the item nodes corresponding to each item in order of distance from small to large, wherein, in the process of constructing the second edge of the item node corresponding to each item, the angle between the newly added second edge and the already added second edge is greater than the angle threshold.

[0102] The first determination module 32 is configured to determine the user vector of the user according to the user information of the user in response to the user's online request, for example, by performing the following steps: Figure 1 Step S120 is shown.

[0103] In some embodiments, the user information includes user portrait information and user historical behavior information related to the item. The first determination module 32 is configured to determine the user vector of the user based on the user portrait information and the user historical behavior information using the user sub-network in the dual-tower model.

[0104] The second determination module 33 is configured to determine, from the promotion content nodes in the target graph structure, the promotion content delivery data corresponding to the promotion content nodes for which the user meets the promotion content delivery conditions, as candidate promotion content, based on the distance between the item vector corresponding to the item node in the target graph structure and the user vector, for example, by performing the following steps: Figure 1 Step S130 is shown.

[0105] In some embodiments, taking the case where the target graph structure also includes a second edge between different item nodes having an associated relationship, the second determination module 33 is further configured to select, from the candidate item nodes, a candidate item node whose distance between the item vector and the user vector is less than a distance threshold, as a reference item node; determine, according to the target graph structure, an item node having a second edge with the reference item node as a candidate item node; determine, according to the target graph structure, promotional content delivery data corresponding to a promotional content node having a first edge with the reference item node and for which the user satisfies the promotional content delivery condition corresponding to the first edge as a candidate promotional content; and repeat the operations of selecting reference item nodes, determining candidate item nodes, and determining candidate promotional content until there are no more candidate item nodes in the target graph structure.

[0106] In some embodiments, in an initial state, the candidate item nodes include one or more item nodes randomly selected from the target graph structure.

[0107] The third determination module 34 is configured to determine the target promotion content according to the determined candidate promotion content, for recalling the promotion content, for example, performing the following steps: Figure 1 Step S140 is shown.

[0108] In some embodiments, the third determination module 34 is further configured to obtain weights of the promotional content delivery data corresponding to the plurality of candidate promotional contents; and determine the target promotional content for promotional content recall according to at least one candidate promotional content having a weight greater than a weight threshold. In some embodiments, the third determination module 34 is further configured to select, from at least one candidate promotional content having a weight greater than a weight threshold, candidate promotional content that meets the promotional content filtering condition of the promotional content delivery platform as the target promotional content.

[0109] Figure 4 is a block diagram showing a device for recalling promotional content according to other embodiments of the present disclosure.

[0110] like Figure 4 As shown, the promotion content recall device 4 includes a memory 41; and a processor 42 coupled to the memory 41. The memory 41 is used to store instructions for executing the corresponding embodiment of the promotion content recall method. The processor 42 is configured to execute the promotion content recall method in any of the embodiments of the present disclosure based on the instructions stored in the memory 41.

[0111] Figure 5 is a block diagram illustrating a computer system for implementing some embodiments of the present disclosure.

[0112] like Figure 5As shown, the computer system 50 may be embodied in the form of a general purpose computing device. The computer system 50 includes a memory 510, a processor 520, and a bus 500 that connects the various system components.

[0113] The memory 510 may include, for example, a system memory, a non-volatile storage medium, etc. The system memory may store, for example, an operating system, an application program, a boot loader, and other programs. The system memory may include a volatile storage medium, such as a random access memory (RAM) and / or a cache memory. The non-volatile storage medium may store, for example, instructions for executing at least one of the corresponding embodiments of the method for recalling promotional content. The non-volatile storage medium includes, but is not limited to, a disk memory, an optical memory, a flash memory, etc.

[0114] The processor 520 can be implemented by a general processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistors, etc. Discrete hardware components. Accordingly, each module such as the judgment module and the determination module can be implemented by a central processing unit (CPU) running instructions in a memory that execute corresponding steps, or can be implemented by a dedicated circuit that executes corresponding steps.

[0115] The bus 500 may use any of a variety of bus architectures, including, but not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, and a Peripheral Component Interconnect (PCI) bus.

[0116] The computer system 50 may also include an input / output interface 530, a network interface 540, a storage interface 550, etc. These interfaces 530, 540, 550, the memory 510, and the processor 520 may be connected via a bus 500. The input / output interface 530 may provide a connection interface for input / output devices such as a display, a mouse, and a keyboard. The network interface 540 may provide a connection interface for various networked devices. The storage interface 550 may provide a connection interface for external storage devices such as a floppy disk, a USB flash drive, and an SD card.

[0117] It should be noted that in the technical solution of the present disclosure, the collection, use, storage, sharing and transfer of user personal information involved are in compliance with the provisions of relevant laws and regulations, and it is necessary to inform the user and obtain the user's consent or authorization. When applicable, the user's personal information is de-identified and / or anonymized and / or encrypted.

[0118] Here, various aspects of the present disclosure are described with reference to flowcharts and / or block diagrams of methods, devices, and computer program products according to embodiments of the present disclosure. It should be understood that each frame of the flowchart and / or block diagram and the combination of frames can be implemented by computer-readable program instructions.

[0119] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable device to produce a machine, so that the processor executes the instructions to produce means for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.

[0120] These computer-readable program instructions may also be stored in a computer-readable memory, which cause the computer to work in a specific manner to produce an article of manufacture, including instructions for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.

[0121] The present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.

[0122] The recall method and device for promotional content and the computer storable medium in the above-mentioned embodiments can improve the recall efficiency, accuracy and reliability of promotional content.

[0123] So far, the method and device for recalling promotional content according to the present disclosure, and the computer storable medium have been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed here.

Claims

1. A method for recalling promotional content, comprising: Acquire a target graph structure, wherein the target graph structure includes a promotion content node representing promotion content delivery data, an item node representing an item vector, and a first edge between the promotion content node and the item node having an associated relationship, wherein an attribute of the first edge includes a promotion content delivery condition; In response to an online request of a user, determining a user vector of the user according to user information of the user; According to the distance between the item vector corresponding to the item node in the target graph structure and the user vector, determining, from the promotion content nodes in the target graph structure, promotion content delivery data corresponding to the promotion content node for which the user meets the promotion content delivery condition as candidate promotion content; Based on the determined candidate promotional content, target promotional content is determined for recalling the promotional content.

2. The method for recalling promotional content according to claim 1, wherein: The target graph structure further includes a second edge between different item nodes having an associated relationship. According to the distance between the item node in the target graph structure and the user vector, promotion content delivery data corresponding to the promotion content node for which the user meets the promotion content delivery condition is determined from the promotion content nodes in the target graph structure, and the candidate promotion content includes: From the candidate item nodes, a candidate item node whose distance between the item vector and the user vector is less than a distance threshold is selected as a reference item node; According to the target graph structure, determining an item node having a second edge with the reference item node as a candidate item node; Determine, according to the target graph structure, promotional content delivery data corresponding to a promotional content node having a first edge with the reference item node and for which the user satisfies a promotional content delivery condition corresponding to the first edge as candidate promotional content; The operations of selecting a reference item node, determining a candidate item node, and determining a candidate promotion content are repeatedly performed until there are no more candidate item nodes in the target graph structure.

3. The method for recalling promotional content according to claim 2, wherein: In an initial state, the candidate item nodes include one or more item nodes randomly selected from the target graph structure.

4. The method for recalling promotional content according to claim 2, wherein: Obtaining the target graph structure includes: Acquire item data including multiple items, multiple promotional content delivery data and the corresponding relationship between them; Determine the item vector for each item; For each item, determining the distance between the item vector of each item and the item vectors of other items; Adding a first edge to the promotion content node corresponding to each piece of promotion content delivery data, for connecting the item node corresponding to the item corresponding to each piece of promotion content delivery data; Determine a promotion content delivery condition corresponding to each piece of promotion content delivery data, which is an attribute of the first edge of each piece of promotion content delivery data; According to the distance between the item vector of each item and the item vectors of other items, a second edge is added to the item node corresponding to each item.

5. The method for recalling promotional content according to claim 4, wherein: There are multiple other items, and according to the distance between the item vector of each item and the item vectors of other items, adding a second edge to the item node corresponding to each item includes: Generate an item relationship pair according to each item node and each other item node; Constructing the second edge between the item nodes corresponding to other items in the item relationship pair and the item node corresponding to each item in the order of distance from small to large, wherein, in the process of constructing the second edge of the item node corresponding to each item, the angle between the newly added second edge and the already added second edge is greater than the angle threshold.

6. The method for recalling promotional content according to claim 1, wherein: There are multiple candidate promotional contents determined. According to the determined candidate promotional contents, target promotional contents are determined. The recall of promotional contents includes: Obtaining weights of promotion content delivery data corresponding to multiple candidate promotion contents; The target promotion content is determined according to at least one candidate promotion content whose weight is greater than a weight threshold, and is used for recalling the promotion content.

7. The method for recalling promotional content according to claim 6, wherein: Determining the target promotion content according to at least one candidate promotion content having a weight greater than a weight threshold includes: From at least one candidate promotion content having a weight greater than a weight threshold, a candidate promotion content satisfying a promotion content filtering condition of the promotion content delivery platform is selected as the target promotion content.

8. The method for recalling promotional content according to claim 1, wherein: The user information includes user portrait information and user historical behavior information related to items. Determining the user vector of the user according to the user information of the user includes: According to the user portrait information and the user historical behavior information, the user vector of the user is determined by utilizing the user sub-network in the double-tower model.

9. A device for recalling promotional content, comprising: an acquisition module configured to acquire a target graph structure, wherein the target graph structure includes a promotion content node representing promotion content delivery data, an item node representing an item vector, and a first edge between the promotion content node and the item node having an associated relationship, wherein an attribute of the first edge includes a promotion content delivery condition; A first determination module is configured to determine a user vector of the user according to user information of the user in response to an online request of the user; A second determination module is configured to determine, from the promotion content nodes in the target graph structure, promotion content delivery data corresponding to the promotion content nodes for which the user meets the promotion content delivery conditions, as candidate promotion content, based on the distance between the item vector corresponding to the item node in the target graph structure and the user vector; The third determination module is configured to determine target promotion content based on the determined candidate promotion content for recalling the promotion content.

10. A device for recalling promotional content, comprising: Memory; as well as A processor coupled to the memory, wherein the processor is configured to execute the promotion content recall method according to any one of claims 1 to 8 based on instructions stored in the memory.

11. A computer storable medium having computer program instructions stored thereon, wherein when the instructions are executed by a processor, the method for recalling promotional content according to any one of claims 1 to 8 is implemented.