Resource delivery method and device based on propagation chain tree
By using a resource delivery method based on a propagation chain tree, user propagation paths are constructed layer by layer, solving the problems of accuracy and waste in resource distribution in existing technologies, and achieving transparent and efficient resource promotion.
Patent Information
- Application Number
- CN202511575527.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-31
- Publication Date
- 2026-02-10
AI Technical Summary
Existing resource distribution methods rely on tag-based user selection, which suffers from insufficient accuracy and resource waste.
A resource deployment method based on a propagation chain tree is adopted. By acquiring target resource vectors and user vector libraries, a propagation chain tree is constructed layer by layer. The priority of user groups is defined and resources are gradually propagated until the preset level is reached. Resource deployment is optimized by combining user behavior feedback.
It achieves transparency and precise targeting in the resource promotion process, avoids resource waste, maximizes distribution effectiveness, and provides a reference for tracking and analyzing the dissemination path.
Smart Images
Figure CN121504556A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of artificial intelligence and data processing technology, and in particular to a resource delivery method and apparatus based on a propagation chain tree. Background Technology
[0002] In the realm of internet applications, particularly in e-commerce, social media, and advertising, distribution strategies for newly launched resources (such as advertisements, products, or services) are crucial for enhancing user experience and promoting these resources. A robust distribution strategy must ensure the efficient and accurate delivery of new resources to target users who are interested in them, thereby maximizing promotional effectiveness. However, existing distribution methods generally rely on tag-based user segmentation, which, while capable of achieving accurate recommendations to some extent, also has significant limitations. Summary of the Invention
[0003] This application aims to at least partially address one of the technical problems in the related art.
[0004] Therefore, one objective of this application is to propose a resource delivery method based on a propagation chain tree, comprising: obtaining a target resource vector corresponding to the target resource to be delivered; selecting at least one seed user from candidate users based on the target resource vector and a pre-built user vector library, wherein the user vector library stores the candidate user vector corresponding to each candidate user; constructing a propagation chain tree corresponding to the target resource layer by layer with the seed user as the root node, and delivering the target resource to the potential user corresponding to each newly expanded node in the propagation chain tree during the layer-by-layer construction process, until the propagation chain tree stops construction or reaches a preset maximum level; wherein the propagation chain tree is used to represent the propagation path of the target resource from the upper-level user to the lower-level potential user, and each node is generated based on the behavioral feedback of the upper-level user after receiving the target resource and the user vector library mining.
[0005] This application proposes a resource delivery method based on a propagation chain tree. The propagation chain tree not only clearly distinguishes the priorities of different user groups, but also clearly shows the propagation process of the target resource among the population. This feature makes the resource promotion and propagation process more transparent, facilitates the tracking and analysis of the propagation path, and provides an important reference for subsequent optimization of promotion strategies. Furthermore, the layer-by-layer expansion of the propagation chain tree ensures the gradual propagation of the target resource from seed users to potential users, enabling the target resource to be accurately delivered according to the predetermined strategy, avoiding unnecessary resource waste, and maximizing the distribution effect.
[0006] The second objective of this application is to propose a resource delivery device based on a propagation chain tree.
[0007] The third objective of this application is to propose an electronic device.
[0008] The fourth objective of this application is to provide a non-transitory computer-readable storage medium.
[0009] The fifth objective of this application is to provide a computer program product.
[0010] To achieve the above objectives, the first aspect of this application proposes a resource delivery method based on a propagation chain tree, comprising: obtaining a target resource vector corresponding to the target resource to be delivered; selecting at least one seed user from candidate users based on the target resource vector and a pre-built user vector library, wherein the user vector library stores the candidate user vector corresponding to each candidate user; constructing a propagation chain tree corresponding to the target resource layer by layer with the seed user as the root node, and delivering the target resource to the potential user corresponding to each newly expanded node in the propagation chain tree during the layer-by-layer construction process, until the propagation chain tree stops construction or reaches a preset maximum level; wherein the propagation chain tree is used to represent the propagation path of the target resource from the upper-level user to the lower-level potential user, and each node is generated based on the behavioral feedback of the upper-level user after receiving the target resource and the user vector library mining.
[0011] According to one embodiment of this application, a propagation chain tree corresponding to the target resource is constructed layer by layer, with the seed user as the root node. During the layer-by-layer construction of the propagation chain tree, the target resource is distributed to the potential users corresponding to the newly expanded nodes in each layer until the propagation chain tree stops construction or reaches a preset maximum level. The process includes: using the seed user as the root node of the propagation chain tree; distributing the target resource to the users corresponding to all nodes in the current layer, and selecting target users who generate positive feedback behavior towards the target resource from the users corresponding to all nodes in the current layer; obtaining the target user vector corresponding to the target user in the current layer, and based on the target user vector and combined with a user vector library, selecting potential users corresponding to the target resource from candidate users; constructing the nodes of the next layer of the propagation chain tree based on the potential users; taking the next layer as the new current layer, repeating the above steps of distributing the target resource to the users corresponding to all nodes in the current layer and subsequent steps, until the propagation chain tree stops construction or reaches a preset maximum level, and finally obtaining a complete propagation chain tree.
[0012] According to one embodiment of this application, a method for constructing a user vector library includes: obtaining a training sample set, wherein each training sample in the training sample set contains one sample resource, one sample user, and the sample user's preference label for the sample resource; training an initial dual-tower model based on the training sample set until the loss function converges to obtain a trained target dual-tower model, wherein the target dual-tower model includes a user tower and a resource tower, wherein during training, the user tower is used to output a sample user vector based on the sample user features corresponding to the sample user, and the resource tower is used to output a sample resource vector based on the sample resource features corresponding to the sample resource; determining multiple candidate users, and outputting a candidate user vector corresponding to each candidate user based on the candidate user features corresponding to each candidate user according to the user tower in the target dual-tower model; and constructing a user vector library based on the candidate user vectors.
[0013] According to one embodiment of this application, obtaining the target resource vector corresponding to the target resource to be deployed includes: determining the target resource to be deployed, and obtaining the target resource features corresponding to the target resource; inputting the target resource features into the resource tower in the target dual-tower model to obtain the target resource vector corresponding to the target resource.
[0014] According to one embodiment of this application, obtaining the target user vector corresponding to the target user in the current layer includes: obtaining the target user features corresponding to the target user in the current layer; inputting the target user features into the user tower in the target dual-tower model to obtain the target user vector corresponding to the target user.
[0015] According to one embodiment of this application, after obtaining a complete propagation chain tree, the propagation path width of the target resource is determined based on the complete propagation chain tree; the propagation depth of each propagation chain in the complete propagation chain tree is obtained; a first weight corresponding to the propagation path width is obtained, and a second weight corresponding to the propagation depth is obtained; and a comprehensive resource score corresponding to the target resource is calculated based on the propagation path width, propagation depth, first weight, and second weight.
[0016] To achieve the above objectives, a second aspect of this application proposes a resource delivery device based on a propagation chain tree, comprising: a target resource vector acquisition module for acquiring the target resource vector corresponding to the target resource to be delivered; a seed user screening module for selecting at least one seed user from candidate users based on the target resource vector and a pre-built user vector library, wherein the user vector library stores the candidate user vector corresponding to each candidate user; and a propagation chain tree construction module for constructing a propagation chain tree corresponding to the target resource layer by layer with the seed user as the root node, and delivering the target resource to the potential user corresponding to each newly expanded node in the propagation chain tree during the layer-by-layer construction process, until the propagation chain tree stops construction or reaches a preset maximum level; wherein the propagation chain tree represents the propagation path of the target resource from the upper-level user to the lower-level potential user, and each node is generated based on the behavioral feedback of the upper-level user after receiving the target resource and the user vector library.
[0017] According to one embodiment of this application, the propagation chain tree construction module is further configured to: use a seed user as the root node of the propagation chain tree; distribute the target resource to all users corresponding to all nodes in the current layer, and select target users who generate positive feedback behavior towards the target resource from all users corresponding to all nodes in the current layer; obtain the target user vector corresponding to the target user in the current layer, and based on the target user vector and in conjunction with a user vector library, select potential users corresponding to the target resource from candidate users; construct the nodes of the next layer of the propagation chain tree based on the potential users; take the next layer as the new current layer, and repeat the above steps of distributing the target resource to all users corresponding to all nodes in the current layer and subsequent steps until the propagation chain tree stops construction or reaches a preset maximum level, and finally obtains a complete propagation chain tree.
[0018] According to one embodiment of this application, the resource delivery device based on the propagation chain tree further includes a model training module, used for: acquiring a training sample set, wherein each training sample in the training sample set contains one sample resource, one sample user, and the sample user's preference label for the sample resource; training an initial dual-tower model based on the training sample set until the loss function converges, thereby obtaining a target dual-tower model after training, wherein the target dual-tower model includes a user tower and a resource tower, and during the training process, the user tower is used to output a sample user vector based on the sample user features corresponding to the sample user, and the resource tower is used to output a sample resource vector based on the sample resource features corresponding to the sample resource.
[0019] According to one embodiment of this application, the resource delivery device based on the propagation chain tree further includes a user vector library construction module, which is used to: determine multiple candidate users, and output the candidate user vector corresponding to each candidate user based on the candidate user features corresponding to each candidate user according to the user tower in the target dual-tower model; and construct a user vector library based on the candidate user vectors.
[0020] According to one embodiment of this application, the propagation chain tree construction module is further configured to: determine the target resource to be deployed, and obtain the target resource features corresponding to the target resource; input the target resource features into the resource tower in the target dual-tower model to obtain the target resource vector corresponding to the target resource.
[0021] According to one embodiment of this application, the propagation chain tree construction module is further configured to: obtain the target user features corresponding to the target user in the current layer; input the target user features into the user tower in the target dual-tower model to obtain the target user vector corresponding to the target user.
[0022] According to one embodiment of this application, the resource delivery device based on the propagation chain tree further includes a scoring module, configured to: after obtaining the complete propagation chain tree, determine the propagation path width of the target resource based on the complete propagation chain tree; obtain the propagation depth of each propagation chain in the complete propagation chain tree; obtain a first weight corresponding to the propagation path width, and obtain a second weight corresponding to the propagation depth; and calculate the comprehensive resource score corresponding to the target resource based on the propagation path width, propagation depth, first weight, and second weight.
[0023] To achieve the above objectives, a third aspect of this application provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to implement the resource delivery method based on the propagation chain tree as described in the first aspect of this application.
[0024] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the resource delivery method based on the propagation chain tree as described in the first aspect of this application.
[0025] To achieve the above objectives, a fifth aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the resource placement method based on a propagation chain tree as described in the first aspect of this application. Attached Figure Description
[0026] 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: Figure 1 This is a schematic diagram illustrating an exemplary implementation of a resource deployment method based on a propagation chain tree, as shown in one embodiment of this application.
[0027] Figure 2 This is a schematic diagram illustrating a training method for obtaining a target dual-tower model, as shown in one embodiment of this application.
[0028] Figure 3 This is a schematic diagram illustrating an exemplary implementation of a resource deployment method based on a propagation chain tree, as shown in one embodiment of this application.
[0029] Figure 4 This is an exemplary schematic diagram illustrating an embodiment of this application, showing a propagation chain tree corresponding to the target resource constructed layer by layer with the seed user as the root node.
[0030] Figure 5 This is a schematic diagram of a complete propagation chain tree shown in one embodiment of this application.
[0031] Figure 6 This is a schematic diagram of a resource delivery device based on a propagation chain tree, as shown in one embodiment of this application.
[0032] Figure 7 This is a schematic diagram of a resource delivery device based on a propagation chain tree, as shown in one embodiment of this application.
[0033] Figure 8 This is a schematic diagram of an electronic device according to one embodiment of this application. Detailed Implementation
[0034] 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.
[0035] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0036] Figure 1 This is a schematic diagram illustrating an exemplary implementation of a resource deployment method based on a propagation chain tree, as shown in this application. Figure 1 As shown, this resource deployment method based on the propagation chain tree includes the following steps: S101, obtain the target resource vector corresponding to the target resource to be deployed.
[0037] In this application, the target resources to be deployed may be resources such as goods, advertisements or services.
[0038] For example, this solution can be applied to scenarios where new resources need to be promoted to users after they are launched.
[0039] In this application, feature extraction can be performed on the target resource features (including resource features such as resource title, category, and tags) corresponding to the target resource to obtain the target resource vector.
[0040] S102, based on the target resource vector and combined with the pre-built user vector library, at least one seed user is selected from the candidate users. The user vector library stores the candidate user vector corresponding to each candidate user.
[0041] In this application, after obtaining the target resource vector as described above, the Top N candidate user vectors that are closest to the cosine distance of the target resource vector are indexed from the user vector library based on the target resource vector, and the candidate users corresponding to the Top N candidate user vectors are determined. The candidate users corresponding to the Top N candidate user vectors are used as seed users who are highly likely to be interested in the target resource.
[0042] S103: Using seed users as the root node, construct the propagation chain tree corresponding to the target resource layer by layer. During the construction of the propagation chain tree layer by layer, the target resource is deployed to the potential users corresponding to each newly expanded node in each layer until the propagation chain tree stops being constructed or reaches the preset maximum level.
[0043] The propagation chain tree represents the propagation path of the target resource from the upper-level user to the lower-level potential user. Each node is generated based on the behavior feedback of the upper-level user after receiving the target resource and the user vector library.
[0044] This application proposes a resource delivery method based on a propagation chain tree. The propagation chain tree not only clearly distinguishes the priorities of different user groups, but also clearly shows the propagation process of the target resource among the population. This feature makes the resource promotion and propagation process more transparent, facilitates the tracking and analysis of the propagation path, and provides an important reference for subsequent optimization of promotion strategies. Furthermore, the layer-by-layer expansion of the propagation chain tree ensures the gradual propagation of the target resource from seed users to potential users, enabling the target resource to be accurately delivered according to the predetermined strategy, avoiding unnecessary resource waste, and maximizing the distribution effect.
[0045] Furthermore, before building the user vector library, it is first necessary to train and obtain the target dual-tower model. Figure 2 This application illustrates a schematic diagram of a training method for obtaining a target dual-tower model, as shown below. Figure 2 As shown, the steps for training the target dual-tower model are as follows: First, obtain a training sample set (which can be obtained based on log data from a resource delivery platform). Each training sample in the training sample set contains one sample resource, one sample user, and the sample user's preference label for the sample resource (for example, if the sample user clicked or favorited the sample resource, the preference label is set to 1; if the sample user did not click or favorite the sample resource, the preference label is set to 0). Second, train the initial dual-tower model based on the training sample set. The target dual-tower model includes a user tower and a resource tower. During training, the user tower is used to output sample user vectors based on the sample user features (including the sample user's gender, age, interests, user behavior sequences, etc.) through a feature extraction network (which can use a multi-layer DNN). The resource tower is used to output sample resource vectors based on the sample resource features (including the resource title, category, tags, etc.) through a feature extraction network (which can use a multi-layer DNN).
[0046] The aforementioned multi-layer DNN refers to a multi-layer deep neural network, one of the core models of deep learning. Its core characteristic is that it contains multiple hidden layers (neural network layers between the input and output layers), which progressively abstract and extract features from the data through multiple non-linear transformations.
[0047] In this application, after obtaining the sample user vector and sample resource vector corresponding to the training samples during the training process, the cosine distance between the sample user vector and the sample resource vector is calculated. In this application, an in-batch softmax objective function (where a sample user vector is closer to the sample resource vector corresponding to its preferred sample resource and farther from the sample resource vector corresponding to its unpreferred sample resource) is used as the loss function for model training. Finally, the calculated loss value is used to back-update the parameters of the initial dual-tower model until the loss function converges, resulting in the trained target dual-tower model.
[0048] Furthermore, after obtaining the target dual-tower model through training, multiple candidate users are identified, and based on the user tower in the target dual-tower model, a candidate user vector is output for each candidate user according to their respective candidate user features; and a user vector library is constructed based on the candidate user vectors. The construction of the user vector library after training the target dual-tower model can be performed offline.
[0049] The user vector library can be used in the resource deployment method based on the propagation chain tree described in this application.
[0050] The user vector library can utilize the FAISS index. FAISS (Facebook AI SimilaritySearch) is specifically designed for efficiently handling similarity search and clustering tasks with large-scale dense vectors. Through highly optimized algorithms and data structures, FAISS significantly improves the efficiency of similarity search and is widely used in recommender systems, image retrieval, and other fields.
[0051] Figure 3 This is a schematic diagram illustrating an exemplary implementation of a resource deployment method based on a propagation chain tree, as shown in this application. Figure 3 As shown, this resource deployment method based on the propagation chain tree includes the following steps: S301, Obtain the target resource vector corresponding to the target resource to be deployed.
[0052] As described above, a target dual-tower model was obtained through training. In this application, after determining the target resource to be deployed, the target resource features corresponding to the target resource are obtained; the target resource features are input into the resource tower in the target dual-tower model to obtain the target resource vector corresponding to the target resource.
[0053] S302, based on the target resource vector and combined with the pre-built user vector library, select at least one seed user from the candidate users. The user vector library stores the candidate user vector corresponding to each candidate user.
[0054] In this application, after obtaining the target resource vector as described above, the Top N candidate user vectors that are closest to the cosine distance of the target resource vector are indexed from the user vector library based on the target resource vector, and the candidate users corresponding to the Top N candidate user vectors are determined. The candidate users corresponding to the Top N candidate user vectors are used as seed users who are highly likely to be interested in the target resource.
[0055] S303 uses seed users as the root node of the propagation chain tree.
[0056] Using the seed user as the root node of the propagation chain tree can also be understood as the first layer of the propagation chain tree.
[0057] S304: Distribute the target resources to all users corresponding to all nodes in the current layer, and select target users who generate positive feedback behavior towards the target resources from all users corresponding to all nodes in the current layer.
[0058] After determining the first layer where the root node is located, the target resource is distributed to the seed users corresponding to all the root nodes in the first layer. Then, it is determined whether each seed user has generated positive feedback behavior for the target resource (such as clicking, favoriting, commenting, following, etc. can all be considered positive feedback behavior). Seed users who have generated positive feedback behavior for the target resource are designated as target users in the first layer.
[0059] For example, suppose there are 3 seed users in the first layer. The target resource is distributed to the 3 seed users in the first layer. Then, it is determined whether each seed user has generated positive feedback behavior for the target resource. If all 3 seed users have generated positive feedback behavior for the target resource, then all 3 seed users are regarded as target users in the first layer.
[0060] S305: Obtain the target user vector corresponding to the target user in the current layer, and based on the target user vector and the user vector library, filter out the potential users corresponding to the target resources from the candidate users.
[0061] After determining the target users of the first layer, the target user features corresponding to the target users in the current layer (currently the first layer) are obtained; the target user features are input into the user tower in the target dual-tower model to obtain the target user vector corresponding to the target user.
[0062] After obtaining the target user vector corresponding to the target user in the current layer, the TopM candidate user vector with the closest cosine distance to the target user vector is indexed from the user vector library based on the target user vector, and the candidate user corresponding to the TopM candidate user vector is determined. The candidate user corresponding to the TopM candidate user vector is taken as the potential user corresponding to the target resource (users who have already been given the target resource need to be excluded).
[0063] S306, the node of the next layer of the propagation chain tree built based on potential users.
[0064] After filtering out potential users corresponding to the target resources based on the target users in the first layer, the next layer of the propagation chain tree is constructed based on these potential users. That is, the current propagation chain tree has been constructed up to the second layer.
[0065] S307, set the next layer as the new current layer.
[0066] S308, determine whether the propagation chain tree has stopped building or reached the preset maximum level. If not, repeat the above steps of placing the target resource into all users corresponding to the nodes of the current layer and subsequent steps.
[0067] S309, When the propagation chain tree stops building or reaches the preset maximum level, the building stops, and the complete propagation chain tree is finally obtained.
[0068] The above has been constructed up to the second layer. After constructing the nodes of the second layer, the second layer is used as the new current layer and the above steps S304 to S307 are repeated until the propagation chain tree stops being constructed (stopping construction can also be understood as the propagation chain tree stopping growth, that is, the users corresponding to the nodes of the last layer have not generated positive feedback behavior for the target resource, indicating that the mining of potential users who may be interested in the target resource has ended) or the preset maximum level is reached, and construction stops, and finally the complete propagation chain tree is obtained.
[0069] Figure 4 This application illustrates an exemplary diagram of a propagation chain tree corresponding to a target resource, constructed layer by layer with the seed user as the root node. Figure 4 As shown, the target resource is denoted as resource v, and assuming the propagation chain tree contains... This represents the nth user within the m-th node of the k-th layer. Figure 4 The construction process of the propagation chain tree shown is as follows: S1. Generate the root node of the propagation chain tree. Specifically, this includes: the first level of the propagation chain tree is the root node, which includes all seed users, assuming it is formally represented as [ , , The generated tree structure is as follows: Figure 4 As shown in Figure 1, the maximum depth of the tree, layer, is 1.
[0070] S2. Deploy resource v to each user within each node of the layer and collect user feedback data on resource v. Gray out users who are not interested (gray indicates a suppressed state, otherwise it is an active state). Assume users... , , They are all interested in resource v, so they are all active (not grayed out).
[0071] S3. For each user in the active state in step S2, calculate their TopM similar users in turn using the user vector database, and generate the corresponding tree branch nodes; assuming users... Similar users are [ , ],user Similar users are [ , ],user Similar users are [ The generated propagation chain tree structure is as follows: Figure 3 As shown in Figure 2, the maximum depth of the tree is layer = layer + 1.
[0072] S4. Repeat steps S2 to S3 to generate branch nodes for other layers of the propagation chain tree until the tree stops growing (e.g., all users within a branch node are in a suppressed state or the user has no similar users) or reaches the specified maximum number of levels (assuming the item propagation chain tree has a maximum of P levels). The generated propagation chain tree structure is as follows: Figure 3 The legend is P, where the maximum depth of the tree is P.
[0073] This application proposes a resource delivery method based on a propagation chain tree. The propagation chain tree not only clearly distinguishes the priorities of different user groups, but also clearly shows the propagation process of the target resource among the population. This feature makes the resource promotion and propagation process more transparent, facilitates the tracking and analysis of the propagation path, and provides an important reference for subsequent optimization of promotion strategies. Furthermore, the layer-by-layer expansion of the propagation chain tree ensures the gradual propagation of the target resource from seed users to potential users, enabling the target resource to be accurately delivered according to the predetermined strategy, avoiding unnecessary resource waste, and maximizing the distribution effect.
[0074] Furthermore, after obtaining the complete propagation chain tree, the comprehensive resource score corresponding to the target resource can be calculated based on the complete propagation chain tree.
[0075] Figure 5 This application illustrates a complete propagation chain tree, as shown in the diagram. Figure 5 As shown, a depth-first traversal algorithm is used to obtain each propagation chain in the complete propagation chain tree. Figure 5 For example, traversing the path yields the following 13 propagation chains (where the 9th and 13th propagation chains are incomplete due to omissions in the diagram, and are only shown here for illustration): 1): to to to
[0076] 2): to to to
[0077] 3): to to to
[0078] 4): to to to
[0079] 5): to to to
[0080] 6): to to to
[0081] 7): to to to
[0082] 8): to to
[0083] 9): to to
[0084] 10): to
[0085] 11): to to
[0086] 12): to to
[0087] 13): to to
[0088] Specifically, the steps for calculating the comprehensive resource score corresponding to the target resource based on the complete propagation chain tree include: determining the propagation path width of the target resource based on the complete propagation chain tree; obtaining the propagation depth of each propagation chain in the complete propagation chain tree; obtaining the first weight corresponding to the propagation path width and the second weight corresponding to the propagation depth; and calculating the comprehensive resource score corresponding to the target resource based on the propagation path width, propagation depth, first weight, and second weight. The formula for the comprehensive resource score corresponding to the target resource can be expressed as:
[0089] In the above formula, The overall resource score representing resource v; This represents the width of the propagation path of resource v among the population, that is, the number of propagation chains. In other words, one propagation chain is a user propagation path. The more users resource v reaches, the higher the item score. The depth of the k-th propagation chain of resource v is represented by the length of the propagation chain. In other words, the longer a single propagation chain is, the more profound its influence is, and the higher the score of resource v is. The first weight corresponding to the width of the propagation path; The second weight represents the depth of propagation.
[0090] Furthermore, the comprehensive resource score obtained from the above calculations can be used to measure the quality of the target resource, thereby determining the traffic allocation strategy for the target resource. For high-quality target resources with high comprehensive resource scores, more traffic support should be provided to accelerate their market promotion and user coverage; while for target resources with low comprehensive resource scores, decisive traffic restriction measures should be taken, or even taken offline, to avoid the continuous investment of ineffective traffic and ensure the efficient use of traffic resources.
[0091] Furthermore, the comprehensive resource score corresponding to the target resource calculated above can also be used to realize the TOP list statistics for each category of resources.
[0092] Figure 6 This application illustrates a schematic diagram of a resource delivery device based on a propagation chain tree, as shown below. Figure 6 As shown, the resource delivery device 600 based on the propagation chain tree includes a target resource vector acquisition module 601, a seed user screening module 602, and a propagation chain tree construction module 603, wherein: The target resource vector acquisition module 601 is used to acquire the target resource vector corresponding to the target resource to be deployed. The seed user screening module 602 is used to select at least one seed user from the candidate users based on the target resource vector and in combination with a pre-built user vector library, wherein the user vector library stores the candidate user vector corresponding to each candidate user. The propagation chain tree construction module 603 is used to build the propagation chain tree corresponding to the target resource layer by layer with the seed user as the root node, and to distribute the target resource to the potential user corresponding to each newly expanded node in the propagation chain tree during the layer-by-layer construction process, until the propagation chain tree stops construction or reaches the preset maximum level.
[0093] The propagation chain tree is used to represent the propagation path of the target resource from the upper-level users to the lower-level potential users. Each node is generated based on the behavioral feedback of the upper-level users after receiving the target resource and the user vector database.
[0094] In this device, the propagation chain tree not only clearly distinguishes the priorities of different user groups, but also clearly displays the propagation process of target resources among the population. This feature makes the resource promotion and propagation process more transparent, facilitates the tracking and analysis of propagation paths, and provides an important reference for subsequent optimization of promotion strategies. Furthermore, the layer-by-layer expansion of the propagation chain tree ensures the gradual propagation of target resources from seed users to potential users, enabling target resources to be accurately deployed according to the established strategy, avoiding unnecessary resource waste, and maximizing the distribution effect.
[0095] Figure 7 This application illustrates a schematic diagram of a resource delivery device based on a propagation chain tree, as shown below. Figure 7 As shown, the resource delivery device based on the propagation chain tree includes a target resource vector acquisition module, a seed user screening module, a model training module, a user vector library construction module, a propagation chain tree construction module, and a scoring module.
[0096] The target resource vector acquisition module is used to obtain the target resource features corresponding to the target resource after determining the target resource to be deployed; and input the target resource features into the resource tower in the target dual-tower model to obtain the target resource vector corresponding to the target resource.
[0097] The seed user screening module is used to: after obtaining the target resource vector, index the Top N candidate user vectors that are closest to the cosine distance of the target resource vector from the user vector library based on the target resource vector, determine the candidate users corresponding to the Top N candidate user vectors, and take the candidate users corresponding to the Top N candidate user vectors as seed users who are highly likely to be interested in the target resource.
[0098] The model training module is used to: acquire a training sample set, wherein each training sample in the training sample set contains one sample resource, one sample user, and the sample user's preference label for the sample resource; train the initial dual-tower model based on the training sample set until the loss function converges to obtain the trained target dual-tower model, wherein the target dual-tower model includes a user tower and a resource tower. During the training process, the user tower is used to output a sample user vector based on the sample user features corresponding to the sample user, and the resource tower is used to output a sample resource vector based on the sample resource features corresponding to the sample resource.
[0099] The user vector library construction module is used to: determine multiple candidate users, and output the candidate user vector corresponding to each candidate user based on the candidate user features corresponding to each candidate user according to the user tower in the target dual-tower model; and construct the user vector library based on the candidate user vectors.
[0100] The propagation chain tree construction module is further used for: using seed users as the root node of the propagation chain tree; distributing target resources to users corresponding to all nodes in the current layer, and selecting target users who generate positive feedback behavior towards the target resources from all users corresponding to all nodes in the current layer; obtaining target user vectors corresponding to target users in the current layer based on the target user vector acquisition unit, and selecting potential users corresponding to the target resources from candidate users based on the target user vectors and the user vector library based on the potential user acquisition unit; then, constructing nodes for the next layer of the propagation chain tree based on the potential users; using the next layer as the new current layer, repeating the above steps of distributing target resources to users corresponding to all nodes in the current layer and subsequent steps, until the propagation chain tree stops construction or reaches the preset maximum level, and finally obtaining a complete propagation chain tree.
[0101] The steps for the target user vector acquisition unit to obtain the target user vector corresponding to the target user in the current layer are as follows: obtain the target user features corresponding to the target user in the current layer; input the target user features into the user tower in the target dual-tower model to obtain the target user vector corresponding to the target user.
[0102] The step of the potential user acquisition unit in selecting potential users corresponding to the target resource from the candidate users based on the target user vector and the user vector library is as follows: based on the target user vector, index the TopM candidate user vector that is closest to the cosine distance of the target user vector from the user vector library, determine the candidate user corresponding to the TopM candidate user vector, and take the candidate user corresponding to the TopM candidate user vector as the potential user corresponding to the target resource.
[0103] The scoring module is used to: determine the propagation path width of the target resource based on the complete propagation chain tree after obtaining the complete propagation chain tree; obtain the propagation depth of each propagation chain in the complete propagation chain tree; obtain the first weight corresponding to the propagation path width and the second weight corresponding to the propagation depth; and calculate the comprehensive resource score corresponding to the target resource based on the propagation path width, propagation depth, first weight and second weight.
[0104] To implement the above embodiments, this application also proposes an electronic device 800, such as... Figure 8 As shown, the electronic device 800 includes a processor 801 and a memory 802 communicatively connected to the processor. The memory 802 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 801 to implement the resource delivery method based on the propagation chain tree as shown in the above embodiment.
[0105] To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to implement the resource placement method based on the propagation chain tree as shown in the above embodiments.
[0106] To implement the above embodiments, this application also proposes a computer program product, including a computer program that, when executed by a processor, implements the resource placement method based on the propagation chain tree as shown in the above embodiments.
[0107] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.
[0108] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0109] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0110] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.
Claims
1. A resource deployment method based on a propagation chain tree, characterized in that, include: Obtain the target resource vector corresponding to the target resource to be deployed; Based on the target resource vector and combined with a pre-built user vector library, at least one seed user is selected from the candidate users, wherein the user vector library stores the candidate user vector corresponding to each candidate user. Using the seed user as the root node, a propagation chain tree corresponding to the target resource is constructed layer by layer. During the construction of the propagation chain tree, the target resource is distributed to the potential users corresponding to each newly expanded node in each layer until the propagation chain tree stops being constructed or reaches the preset maximum level. The propagation chain tree represents the propagation path of the target resource from upper-level users to lower-level potential users, and each node is generated based on the behavioral feedback of the upper-level user after receiving the target resource and the user vector database.
2. The method according to claim 1, characterized in that, The process of constructing a propagation chain tree corresponding to the target resource layer by layer, with the seed user as the root node, and during the layer-by-layer construction of the propagation chain tree, distributing the target resource to the potential users corresponding to each newly expanded node at each layer, until the propagation chain tree stops construction or reaches a preset maximum level, includes: The seed user is used as the root node of the propagation chain tree; The target resource is distributed to all users corresponding to all nodes in the current layer, and target users who generate positive feedback behavior towards the target resource are selected from all users corresponding to all nodes in the current layer. Obtain the target user vector corresponding to the target user in the current layer, and based on the target user vector and the user vector library, filter out the potential users corresponding to the target resource from the candidate users; Based on the potential users, construct the nodes of the next layer of the propagation chain tree; The next layer is taken as the new current layer, and the above steps of placing the target resource into all the users corresponding to the nodes of the current layer and subsequent steps are repeated until the propagation chain tree stops building or reaches the preset maximum level, and the construction stops, finally obtaining a complete propagation chain tree.
3. The method according to claim 2, characterized in that, The method for constructing the user vector library includes: Obtain a training sample set, wherein each training sample in the training sample set contains one sample resource, one sample user, and the sample user's preference label for the sample resource; The initial dual-tower model is trained based on the training sample set until the loss function converges, resulting in a target dual-tower model. The target dual-tower model includes a user tower and a resource tower. During training, the user tower is used to output a sample user vector based on the sample user features corresponding to the sample user, and the resource tower is used to output a sample resource vector based on the sample resource features corresponding to the sample resource. Multiple candidate users are identified, and based on the user tower in the target dual-tower model, a candidate user vector corresponding to each candidate user is output according to the candidate user features corresponding to each candidate user. The user vector library is constructed based on the candidate user vectors.
4. The method according to claim 3, characterized in that, The step of obtaining the target resource vector corresponding to the target resource to be deployed includes: Determine the target resources to be deployed, and obtain the target resource characteristics corresponding to the target resources; The target resource features are input into the resource tower in the target dual-tower model to obtain the target resource vector corresponding to the target resource.
5. The method according to claim 4, characterized in that, The step of obtaining the target user vector corresponding to the target user in the current layer includes: Obtain the target user characteristics corresponding to the target user in the current layer; The target user features are input into the user tower in the target dual-tower model to obtain the target user vector corresponding to the target user.
6. The method according to claim 2, characterized in that, The method further includes: After obtaining the complete propagation chain tree, the propagation path width of the target resource is determined based on the complete propagation chain tree; Obtain the propagation depth of each propagation chain in the complete propagation chain tree; Obtain the first weight corresponding to the propagation path width, and obtain the second weight corresponding to the propagation depth; Based on the propagation path width, the propagation depth, the first weight, and the second weight, the comprehensive resource score corresponding to the target resource is calculated.
7. A resource delivery device based on a propagation chain tree, characterized in that, include: The target resource vector acquisition module is used to acquire the target resource vector corresponding to the target resource to be deployed; The seed user screening module is used to select at least one seed user from the candidate users based on the target resource vector and in combination with a pre-built user vector library, wherein the user vector library stores the candidate user vector corresponding to each candidate user. The propagation chain tree construction module is used to construct the propagation chain tree corresponding to the target resource layer by layer with the seed user as the root node, and during the construction of the propagation chain tree layer by layer, the target resource is deployed to the potential users corresponding to each newly expanded node in each layer until the propagation chain tree stops construction or reaches the preset maximum level. The propagation chain tree represents the propagation path of the target resource from upper-level users to lower-level potential users, and each node is generated based on the behavioral feedback of the upper-level user after receiving the target resource and the user vector database.
8. An electronic device, comprising: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.
9. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.
10. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1-6.