Propagation evaluation method of recommended object and propagation planning method of recommended object

Abstract

The embodiment of the invention provides a recommendation object propagation evaluation method and a recommendation object propagation planning method. The recommendation object propagation evaluation method comprises the steps of obtaining an initial network value of a propagation network of a target recommendation object and a propagation queue; selecting a first propagation probability from the propagation queue; based on the first propagation probability, determining a propagation value gain of the initiating user; determining the target user value of the initiating user based on the initial user value of the initiating user and the propagation value gain, and returning to execute the step of selecting the first propagation probability from the propagation queue until the propagation queue is empty, and determining the target network value of the propagation network based on the target user values of the multiple users; and determining a network value gain of the propagation network based on the target network value and the initial network value. And the real contribution of the dynamic propagation network is quantified, so that the network value evaluation is more accurate and reliable.

Description

Technical Field

[0001] The embodiments in this specification relate to the technical field of project evaluation, and in particular to a method for evaluating the dissemination of recommendations and a method for planning the dissemination of recommendations. Background Technology

[0002] With the rapid development of internet technology, especially social media platforms, information dissemination networks have become a key carrier for information dissemination, product promotion, and public opinion management. These networks typically consist of multiple user nodes, each representing a user, interconnected through dissemination behaviors (such as reading, clicking, sharing, and forwarding), forming dynamic dissemination paths. Valuing the value of dissemination networks is crucial for optimizing resource allocation, improving dissemination efficiency, and enhancing network resilience.

[0003] Currently, the valuation of communication networks mainly employs static or simplified models. For example, the initial network value is calculated directly based on users' basic attributes (such as the number of followers and activity level), or a fixed propagation coefficient is applied to estimate the spread of influence.

[0004] However, because propagation behavior is inherently dynamic, the probability of propagation between different users is influenced by various factors (such as the strength of user relationships and content relevance), and the propagation process exhibits chain reaction characteristics. Ignoring this dynamic propagation path can lead to distorted network value assessment, thereby limiting the optimization of propagation strategies and reducing the efficiency of network enhancement. Summary of the Invention

[0005] In view of this, embodiments of this specification provide a method for evaluating the propagation of recommended objects. One or more embodiments of this specification also relate to a method for planning the propagation of recommended objects, a method for propagating recommended objects, a platform for recommended objects, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0006] According to a first aspect of the embodiments of this specification, a method for evaluating the propagation of a recommended object is provided, comprising:

[0007] Obtain the initial network value of the propagation network of the target recommendation object, and the propagation queue, wherein the propagation network includes user nodes corresponding to multiple users, the initial network value includes the initial user value of multiple users, and the propagation queue includes multiple propagation probabilities;

[0008] Select a first propagation probability from the propagation queue, where the first propagation probability represents the probability that the initiating user influences the reaching user to initiate a second propagation behavior through the first propagation behavior;

[0009] Based on the first propagation probability, determine the propagation value gain of the initiating user;

[0010] Based on the initial user value and propagation value gain of the initiating user, the target user value of the initiating user is determined, and the step of selecting the first propagation probability from the propagation queue is returned to be executed until the propagation queue is empty. Based on the target user values ​​of multiple users, the target network value of the propagation network is determined.

[0011] Based on the target network value and the initial network value, determine the network value gain of the propagation network.

[0012] According to a second aspect of the embodiments of this specification, a method for planning the propagation of recommendation objects is provided, comprising:

[0013] Obtain the network value gain of the propagation network of the target recommendation object, wherein the network value gain is the difference between the target network value and the initial network value obtained after iteratively updating the user value based on each propagation probability in the propagation queue; the propagation queue includes multiple propagation probabilities.

[0014] Based on network value gain, a target user reach sequence for the target recommendation object is generated, wherein the target user reach sequence includes a sequence of user nodes arranged according to network value gain.

[0015] According to a third aspect of the embodiments of this specification, a method for propagating a recommendation object is provided, comprising:

[0016] Obtain the target user reach sequence of the target recommendation object, wherein the target user reach sequence is a sequence of user nodes reached by the target recommendation object generated based on the network value gain of the propagation network, the network value gain is the difference between the target network value obtained after iteratively updating the user value of each propagation probability in the propagation queue and the initial network value, and the propagation queue includes multiple propagation probabilities;

[0017] Based on the target user reach sequence, the content of the target recommendation object is propagated to users in a sequential manner.

[0018] According to a fourth aspect of the embodiments of this specification, a recommendation object platform is provided, including an offline module and an online module;

[0019] Offline module, used for:

[0020] Obtain the initial network value of the propagation network of the target recommendation object, and the propagation queue, wherein the propagation network includes user nodes corresponding to multiple users, the initial network value includes the initial user value of multiple users, and the propagation queue includes multiple propagation probabilities;

[0021] Select a first propagation probability from the propagation queue, where the first propagation probability represents the probability that the initiating user influences the reaching user to initiate a second propagation behavior through the first propagation behavior;

[0022] Based on the first propagation probability, determine the propagation value gain of the initiating user;

[0023] Based on the initial user value and propagation value gain of the initiating user, the target user value of the initiating user is determined, and the step of selecting the first propagation probability from the propagation queue is returned to be executed until the propagation queue is empty. Based on the target user values ​​of multiple users, the target network value of the propagation network is determined.

[0024] Based on the target network value and the initial network value, determine the network value gain of the propagation network;

[0025] Based on network value gain, a target user reach sequence for the target recommendation object is generated;

[0026] Online module, used for:

[0027] Based on the target user reach sequence, the content of the target recommendation object is propagated to users in a sequential manner.

[0028] According to a fifth aspect of the embodiments of this specification, a computing device is provided, comprising:

[0029] memory and processor;

[0030] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the above method.

[0031] According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0032] According to a seventh aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0033] In one embodiment of this specification, the initial network value of the propagation network of the target recommendation object and a propagation queue are obtained. The propagation queue contains multiple propagation probabilities, each representing the probability that an initiating user will influence a reaching user to initiate a subsequent second propagation action through a first propagation action, thereby directly modeling the dynamic path of the propagation chain. Based on this, the propagation probabilities are iteratively selected and the propagation value gain of the initiating user of a single user node is calculated. The progressively accumulated gain simulates the dynamic diffusion mechanism of the propagation action, and this gain reflects its actual influence in triggering a chain reaction. Then, the target user value of a single user node is updated, and finally the target network value of the entire propagation network is obtained. By comparing the target network value with the initial network value, the network value gain is obtained. This gain quantifies the true contribution of the dynamic propagation network, making the network value assessment more accurate and reliable. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the architecture of a recommendation platform;

[0035] Figure 2 This is a flowchart illustrating a method for planning the propagation of recommendations.

[0036] Figure 3 This is a flowchart illustrating a method for evaluating the propagation of recommendations.

[0037] Figure 4 This is a flowchart illustrating a method for evaluating the propagation of a recommended object, as provided in one embodiment of this specification.

[0038] Figure 5 This is a schematic diagram illustrating the propagation behavior in a propagation evaluation method for a recommended object provided in one embodiment of this specification;

[0039] Figure 6 This is one of the schematic diagrams illustrating the probability in a method for evaluating the propagation of a recommendation object provided in one embodiment of this specification;

[0040] Figure 7 This is a second schematic diagram illustrating the probability in a propagation evaluation method for a recommended object provided in one embodiment of this specification;

[0041] Figure 8 This is one of the structural schematic diagrams of the prediction model in a propagation evaluation method for recommending objects provided in one embodiment of this specification;

[0042] Figure 9 This is the second schematic diagram of the prediction model in a propagation evaluation method for recommending objects provided in one embodiment of this specification;

[0043] Figure 10This is a schematic diagram of user clustering in a method for evaluating the propagation of a recommendation object provided in one embodiment of this specification;

[0044] Figure 11 This is a timing diagram illustrating a method for evaluating the propagation of a recommended object, provided in one embodiment of this specification.

[0045] Figure 12 This is a flowchart illustrating a method for planning the propagation of a recommended object, as provided in one embodiment of this specification.

[0046] Figure 13 This is a flowchart illustrating a method for propagating a recommended object according to one embodiment of this specification;

[0047] Figure 14 This is a flowchart illustrating the process of a method for evaluating the propagation of recommended objects in advertising distribution on a content community platform, provided by one embodiment of this specification.

[0048] Figure 15 This is a schematic diagram of the structure of a recommendation object platform provided in one embodiment of this specification;

[0049] Figure 16 This is a structural block diagram of a computing device provided by one embodiment of this specification. Detailed Implementation

[0050] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0051] The terminology used in one or more embodiments of the present invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of the invention. The singular forms “a,” “the,” and “the” used in one or more embodiments of the invention are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” used in one or more embodiments of the invention refers to and includes any or all possible combinations of one or more associated listed items.

[0052] It should be understood that although various information may be described using terms such as first, second, etc., in one or more embodiments of the present invention, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of the present invention, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0053] Furthermore, it should be noted that the data involved in one or more embodiments of the present invention are all information and data authorized by the user or fully authorized by all parties, and the statistics, use and processing of the relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0054] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0055] Recommended Target: Refers to the specific entity in a transaction or analysis. In different scenarios, it can be represented as: Advertisement: Content carriers such as images / videos that serve as the target of advertising; SPU Collection: A standardized product cluster that serves as a product management unit (e.g., all color versions of the iPhone 15 are collectively referred to as one SPU); SKU: The smallest inventory management unit in retail, e-commerce, and warehousing management; Brand: A company's intellectual property assets that serve as the target of value assessment.

[0056] Standard Product Unit (SPU): A collection of marketable goods that share the same key attributes (such as name, function, model).

[0057] Stock Keeping Unit (SKU): The smallest unit of inventory management in retail, e-commerce and warehousing management. It is used to uniquely identify the specific specifications of a product and is usually composed of letters, numbers or a combination of codes.

[0058] Human-to-Human (H2H) marketing is a sharing marketing method based on community content platforms. It aims to enhance the influence and reach of recommended objects and the long-term effect of recommended content through users’ spontaneous and large-scale content creation and dissemination.

[0059] User-generated content (UGC) is community content created spontaneously by users. It is one of the core elements of word-of-mouth marketing and aims to inspire users to produce user-generated content materials through high-quality marketing content, thereby creating a wider range of influence within the community content platform than a single recommended object.

[0060] White-box person-to-person marketing referral distribution: A referral distribution technique that utilizes person-to-person marketing. This method explicitly models the content sharing and dissemination process within the propagation network of a community content platform, and optimizes referral distribution through this network modeling.

[0061] Expected Cost Per Mille (ECPM) is a metric that measures the revenue efficiency of ad placements. It represents the estimated average revenue per thousand ad placement impressions for the ad placement advertiser. The calculation formula is (Total Revenue / Total Impressions) × 1000, used to compare the ad placement value across different communication channels.

[0062] Click-Through Rate (CTR): This measures the percentage of times a recommended product or content is clicked. It is calculated by dividing the number of clicks by the number of impressions and is one of the core metrics for evaluating the effectiveness of digital marketing. For example, if a recommended product is displayed 1000 times and generates 50 clicks, then the CTR is 5%.

[0063] Predicted Click-Through Rate (pCtr): This is the probability of content being clicked estimated by a machine learning model, used for optimizing recommendation ranking and delivery. This metric makes real-time predictions based on user characteristics, content characteristics, and contextual features. For example, the estimated click probability for a user on a particular recommended item might be pCtr = 3.2%.

[0064] Currently, the essence of recommendation platforms is to maximize immediate value, and word-of-mouth marketing is one important path: authentic content and sharing are the characteristics of community content platforms. Recommendation platforms can use word-of-mouth and viral spread among people to allow excellent content to be repeatedly disseminated within the community content platform's dissemination network, reaching high-value users and optimizing the value of the recommended objects. Figure 1 This diagram illustrates the architecture of a recommendation platform:

[0065] On traditional traffic-driven recommendation platforms, recommenders expose their products to users, thereby achieving conversions and generating direct / immediate value.

[0066] On a recommendation platform based on a word-of-mouth marketing model, the recommender exposes the recommended object to users. After receiving the recommended object, users can spread it through social networks, user-generated content, and direct communication, reaching multiple users such as user 1, user 2, user 3, and user n, generating dissemination value / viral value. Based on this, the value of the recommended object can be optimized by combining direct value and immediate value.

[0067] The valuation of a dissemination network primarily employs static or simplified models. For example, the initial network value can be calculated directly based on basic user attributes (such as the number of followers and activity level), or a fixed dissemination coefficient can be applied to estimate the diffusion of influence.

[0068] However, because propagation behavior is inherently dynamic, the probability of propagation between different users is influenced by various factors (such as the strength of user relationships and content relevance), and the propagation process exhibits chain reaction characteristics. Ignoring this dynamic propagation path can lead to distorted network value assessment, thereby limiting the optimization of propagation strategies and reducing the efficiency of network enhancement.

[0069] This specification provides a method for planning the dissemination of recommendations and a method for evaluating the dissemination of recommendations.

[0070] The propagation planning method for the recommended users pre-plans the target user reach sequence offline to reach multiple user nodes in the propagation network, and then reaches users sequentially according to the offline plan in the online propagation part. Figure 2 A flowchart illustrating a method for planning the propagation of recommendations is shown below:

[0071] Each user node has an initial value. According to a certain algorithm, it is determined that users need to be reached in the order of U2, U3, U5, U3. The network propagation effect can be relied upon to obtain the maximum propagation value. After that, the online part will follow the offline plan to generate sequential reach to the corresponding users.

[0072] However, such a method of planning the propagation of recommendations is too complex and requires a simplified approach: degrading to the changes in the propagation network after reaching a single user. Figure 3 A flowchart illustrating a method for evaluating the propagation of recommendations is shown below:

[0073] When a user receives a target recommendation (reaching U2), various propagation behaviors may occur in the network, increasing the propagation value of each user in the network, ultimately generating a network value gain of 0.34 for the entire propagation network. With this basic propagation network model, we can obtain the network value gain generated in the network by a single propagation behavior of a recommendation reach, i.e., the expected value (NetAdvv), and use it to calculate the expected value per thousand impressions online, such as bid perturbation, boost, or directly use it to design advertising products based on expected value.

[0074] This specification also relates to a method for disseminating recommendations, a platform for recommending recommendations, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail in the following embodiments.

[0075] See Figure 4 , Figure 4 A flowchart illustrating a method for evaluating the propagation of a recommended object according to an embodiment of this specification is shown, including the following specific steps:

[0076] Step 402: Obtain the initial network value of the propagation network of the target recommendation object, and the propagation queue, wherein the propagation network includes user nodes corresponding to multiple users, the initial network value includes the initial user value of multiple users, and the propagation queue includes multiple propagation probabilities.

[0077] The target recommendation object is a quantifiable entity that serves as the core analysis object in the propagation network, and its value is increased through the dynamic propagation behavior of user nodes. The target recommendation object is the triggering medium of the propagation network, and its value assessment depends on the chain propagation effect among users. The target recommendation object includes, but is not limited to, advertising content (image / text / video), standardized product units (SPU), minimum stock units (SKU), or brand intellectual property assets.

[0078] A propagation network is a dynamic topology structure composed of user nodes and their inter-node propagation behaviors, used to simulate the diffusion path of information or recommended objects. Edge weights in the network represent propagation probabilities, and node values ​​are iteratively updated with propagation behaviors. For example, the hierarchical connections formed by users through forwarding and sharing in a community content platform. The user node corresponding to a user is a vertex in the propagation network representing the user entity and its attributes, including initial value and dynamic propagation capabilities. Optionally, node attributes include user value. For example, a buyer account node in a product promotion chain in a community content platform. A user's initial user value is a quantitative indicator of the user node's baseline value before being affected by propagation behaviors. A user's initial user value can be statically calculated based on historical behavioral data (such as click-through rate and conversion rate). For example, the initial value of user U1 is its number of followers × average content interaction rate. The initial network value of the propagation network is the sum of the network's baseline values ​​before being affected by dynamic propagation behaviors, aggregated from the initial user values ​​of all user nodes. The initial network value is calculated based on the initial user values ​​of multiple users, without considering the cascading effects of the propagation chain. For example, in a promotional campaign on a community content platform, the initial network value is the sum of the number of followers of all participating users multiplied by the average content interaction rate.

[0079] A propagation queue is a set of propagation behaviors to be processed, containing the propagation probabilities and relationships between user nodes. The propagation queue is used to dynamically simulate propagation paths; each element can contain a tuple of user and propagation probability. For example, user A has a 15% probability of sharing product content to user B, and user B has a 10% probability of propagating it to user C: (User B, 15%) → (User C, 10%). Propagation probability is the conditional probability that a user node influences other nodes through a specific propagation behavior (such as forwarding or sharing). Propagation probability is dynamically adjusted by the strength of user relationships and content relevance. For example, after user U4 posts beauty content, their follower U5 has a 12% probability of secondary propagation.

[0080] One possible way to obtain the initial network value of the propagation network of the target recommendation object is to sum the initial user values ​​of multiple users. Another possible way is to predict the initial network value of the propagation network of the target recommendation object based on the sum of the initial user values ​​of multiple users using a prediction model. No specific method is specified here.

[0081] One possible way to obtain the propagation queue is to obtain the propagation network of the target recommendation object, iteratively generate multiple propagation probabilities based on the influence probabilities between multiple users in the propagation network, and construct a propagation queue based on the multiple propagation probabilities. Another possible way is to analyze the historical propagation data of the target recommendation object, count the propagation frequency between multiple users in the propagation network, calculate multiple propagation probabilities, and construct a propagation queue based on the multiple propagation probabilities. Yet another possible way is to use user relationship graphs and machine learning models to predict the propagation probabilities between multiple users and construct a propagation queue based on the multiple propagation probabilities. No particular method is specified here.

[0082] For example, the propagation network of an advertisement is represented as G = (U, E, Ω, T).

[0083] The user set is represented as U = {u i Each user has a corresponding user value. The content set is represented as T = {t} i} represents the content within a propagation network, such as a set of SPUs. The set of propagation behaviors is represented as... This represents the dissemination behavior (channels) of certain content, such as user-generated content exposure, user-generated content reading, comment reading, and sharing reading. The set of influence probabilities is represented as... Impact probability Indicates the second user u j by the first user u i First propagation behavior ω t The probability of influence, for example, the first user u i After publishing user-generated content, u j There is a 10% probability that the user-generated content will be read.

[0084] Initial user value for multiple users Summing yields the initial network value of the propagation network for the target recommendation object. And a propagation queue Q, which contains multiple tuples. Any binary tuple includes user u and propagation probability

[0085] The propagation generation probability is characterized as Indicates that the user was u i A certain propagation behavior ω t Influenced and initiated dissemination behavior The probability of propagation, for example, the probability that a user will share content about a cosmetic product after reading user-generated content about it. Accordingly, the propagation probability is represented as... This indicates that the initiating user u uses the first propagation action ω.t Impact on reaching users Initiating a second dissemination campaign The probability of propagation.

[0086] In step 402, the initial network value of the propagation network of the target recommendation object is obtained to provide a benchmark reference. A propagation queue is obtained, which contains multiple propagation probabilities. Each propagation probability represents the probability that an initiating user will influence the reaching user to initiate subsequent propagation behaviors through propagation behavior, thereby directly modeling the dynamic path of the propagation chain.

[0087] Step 404: Select a first propagation probability from the propagation queue, wherein the first propagation probability represents the probability that the initiating user will influence the reaching user to initiate a second propagation behavior through the first propagation behavior.

[0088] The first propagation probability is the propagation probability selected from the propagation queue in the current iteration. It represents the conditional probability that the initiating user's first propagation action influences the reaching user to initiate a second propagation action within a specific propagation path. The first propagation probability reflects dynamic factors such as user relationship strength, content relevance, and propagation channel characteristics. For example, after user A shares product content, their follower user B has a 15% probability of reposting it.

[0089] The initiating user is the user node that actively initiates a propagation behavior in the propagation chain. As the previous node in the propagation path, the initiating user's propagation behavior will trigger subsequent chain reactions. For example, a high-influence user who publishes product review content for the first time on a community platform. The first propagation behavior is the initial propagation action that the initiating user influences and reaches users. The first propagation behavior includes, but is not limited to, content generation, content sharing, forwarding, mentioning, and other propagation behaviors. For example, a user clicking the "Share to social media" function on a social platform.

[0090] The "reaching user" refers to the receiving user node directly affected by the propagation behavior of the initial user. It's important to note that the reaching user may become the initiator of the next level of propagation, forming a propagation chain. The second propagation behavior is a subsequent propagation action initiated by the reaching user after being influenced by the first propagation behavior. The second propagation behavior reflects the cascading effect of propagation, and its probability of occurrence is positively correlated with the probability of the first propagation. Second propagation behaviors include, but are not limited to, content generation, content sharing, forwarding, and mentioning. For example, a user forwarding a product link shared by a friend.

[0091] For example, if the propagation queue Q is not empty, the first tuple is selected from the propagation queue Q. The first binary pair Including reaching user u and the probability of first propagation

[0092] In step 404, the first propagation probability of the current iteration is selected, which lays the foundation for updating the target user value of the initiating user in the current iteration.

[0093] Step 406: Based on the first propagation probability, determine the propagation value gain of the initiating user.

[0094] The propagation value gain of the initiating user is the increase in network value caused by the initiating user's first propagation behavior, quantifying the contribution of their propagation behavior to network value. The propagation value gain value is jointly determined by the propagation probability and the value of reaching users, reflecting the potential influence of the propagation behavior. For example, after user A shares product content, based on a 15% propagation probability and the value of reaching user B, a value gain of 0.23 is calculated.

[0095] Based on the first propagation probability, the propagation value gain of the initiating user can be determined. One possible approach is to calculate the user value increment of the initiating user through an activation function, and then determine the propagation value gain of the initiating user based on the first propagation probability and the user value increment of the initiating user. Another possible approach is to predict the propagation value gain of the initiating user based on the first propagation probability through a machine learning model. No specific method is specified here.

[0096] For example, the activation function is This indicates the increase in user value after a user is reached by media activities.

[0097] Through activation function Calculate the incremental user value of the initiating user Based on the first propagation probability And the incremental user value of the initiating user Determine the propagation value gain of the initiating user

[0098] In step 406, the propagation value gain of the initiating user is calculated through the first propagation probability, which realizes the accurate quantification of the value of a single propagation behavior, transforms the abstract propagation probability into a specific value increment, dynamically calculates the potential gain of each propagation, and provides an accurate incremental basis for subsequent iterations to update user value.

[0099] Step 408: Based on the initial user value and propagation value gain of the initiating user, determine the target user value of the initiating user, return to the step of selecting the first propagation probability from the propagation queue, until the propagation queue is empty, and determine the target network value of the propagation network based on the target user values ​​of multiple users.

[0100] The target user value of the initiating user is the dynamic value of the user node after iterative updates based on the propagation value gain. The target user value of the initiating user is the incremental value accumulated from the initial value and the propagation chain effect, reflecting the user's actual influence in the current propagation path. User U4's initial value is 0.04; due to triggering U3's propagation behavior (gain +0.01), its target value is updated to 0.05. The target network value of the propagation network is the sum of updated values ​​after being affected by dynamic propagation behavior, aggregated from the target user values ​​of all user nodes. The target network value of the propagation network quantifies the effect of the dynamic propagation path on the overall network value. For example, if the initial network value is 0.24, after the propagation chain update, the target network value increases to 0.58.

[0101] Based on the initial user value and propagation value gain of the initiating user, the target user value of the initiating user is determined. One possible approach is to sum the initial user value and propagation value gain of the initiating user to obtain the target user value of the initiating user. Another possible approach is to weight the propagation value gain with a decay factor and then merge it with the initial user value to obtain the target user value of the initiating user. No specific method is specified here.

[0102] Based on the target user value of multiple users, the target network value of the propagation network is determined. One possible approach is to sum the target user values ​​of multiple users to obtain the initial network value of the propagation network of the target recommendation object. Another possible approach is to predict the initial network value of the propagation network of the target recommendation object based on the sum of the target user values ​​of multiple users through a prediction model. No specific method is specified here.

[0103] For example, the initial user value of the initiating user and the gain of communication value By combining these elements, the target user value of the initiating user can be obtained. If the propagation queue Q is not empty, select the first tuple from the propagation queue Q. The first binary pair Including reaching user u and the probability of first propagation The target user value for multiple users until the propagation queue Q is empty. Summing yields the target network value of the propagation network for the target recommendation object.

[0104] In step 408, the propagation probability is iteratively selected and the propagation value gain of the initiating user is calculated. The progressive accumulation of gains simulates the dynamic diffusion mechanism of the propagation behavior. This gain reflects the actual impact of the chain reaction it triggers. In turn, the user value is updated and the target network value is finally obtained, providing data support for the subsequent determination of the network value gain of the propagation network.

[0105] Step 410: Determine the network value gain of the propagation network based on the target network value and the initial network value.

[0106] The network value gain of a propagation network is the difference between the target network value and the initial network value. The network value gain quantifies the net increase in network value due to propagation dynamics. For example, target network value 0.58 - initial value 0.24 = network value gain 0.34.

[0107] Based on the target network value and the initial network value, the network value gain of the propagation network is determined. One possible approach is to calculate the arithmetic difference between the target network value and the initial network value to obtain the network value gain of the propagation network. Another possible approach is to normalize the target network value and the initial network value, calculate the arithmetic difference between the normalized target network value and the initial network value to obtain the network value gain of the propagation network. No particular limitation is made here.

[0108] For example, calculating the target network value With initial network value The arithmetic difference is used to obtain the network value gain for each recommending object in the propagation network.

[0109] In the embodiments of this specification, a propagation queue contains multiple propagation probabilities, each representing the probability that an initiating user will influence a reaching user to initiate subsequent propagation actions through propagation behavior, thereby directly modeling the dynamic path of the propagation chain. Based on this, the propagation probabilities are iteratively selected and the propagation value gain of the initiating user is calculated. The progressive accumulation of gains simulates the dynamic diffusion mechanism of propagation behavior, and this gain reflects its actual influence in triggering a chain reaction. Then, the user value is updated and the target network value is finally obtained. By comparing the target network value with the initial network value, the network value gain is obtained. This gain quantifies the true contribution of propagation dynamics, making network value assessment more accurate and reliable.

[0110] Corresponding to steps 402 to 410 above, Figure 5 The diagram illustrates the propagation behavior in a propagation evaluation method for a recommended object provided in one embodiment of this specification:

[0111] In the offline component, an algorithm module is deployed to update the propagation network using a message queue of acquired user behavior events. This process enhances the propagation network from its initial state to an improved one, and the updated data is cached for use by the online single-point network efficiency estimation module. The delivery system utilizes the single-point network efficiency estimation algorithm to recommend targets. The expected value per thousand impressions of the delivery system is calculated as f(propagation behavior, target network value of the propagation network), and then applied according to the strategy design.

[0112] In one optional embodiment of this specification, before step 406, the following specific steps are further included: calculating the user value increment of the initiating user based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user through an activation function;

[0113] Correspondingly, step 406 includes the following specific steps: determining the propagation value gain of the initiating user based on the first propagation probability and the user value increment of the initiating user.

[0114] An activation function is a mathematical function used to quantify the impact of propagation behavior on user value. It maps propagation behavior characteristics to user characteristics as value increments, reflecting dynamic propagation effects. For example, the sigmoid function takes the type of propagation behavior (such as forwarding) and user activity as input and outputs a value increment of 0 to 1.

[0115] The behavioral features of the second propagation behavior are the deep encoded features of the second propagation behavior initiated by the user. These behavioral features are constructed using a composite feature space built with multimodal fusion coding technology, containing a fusion representation of static attribute features and dynamic behavioral features. Embedding technology is used to transform discrete features into a continuous vector space. For example, spatiotemporal features include: the time decay coefficient of the propagation behavior; semantic features include: the cross-modal similarity between the propagated content and the recommended object (e.g., text-image matching degree); topological features include: the hierarchical penetration of the behavior in the network (quantified using the PageRank algorithm); and channel features include: the weight coefficients of different propagation media (e.g., social media = 0.7, private chat = 0.3).

[0116] The user features of the initiating user are deep encoded features. User features include static attribute features, which are transformed into a continuous vector space using embedding techniques. For example, static features include one-hot encoding of basic user attributes (age, gender, region, etc.); dynamic features include time-series modeling features of user historical behavior (clicks, forwards, comments, etc.); social features include graph neural network embedding representations of user social relationships; and interest features include topic model distributions of user content preferences.

[0117] The incremental user value for the initiating user represents the dynamic change in value resulting from the user's reach through dissemination activities, quantifying the immediate impact of a single dissemination on the user node. Calculated through an activation function, it integrates dissemination behavior characteristics (such as channel type and content relevance) with user characteristics (such as historical activity and social influence) to output a numerical gain. For example, after user U3 is reached by a beauty ad, the activation function outputs a value increment of +0.02 based on their historical conversion rate (user characteristic) and ad text / image matching degree (behavioral characteristic).

[0118] Figure 6This diagram illustrates one of the probabilities in a propagation evaluation method for a recommendation object provided in an embodiment of this specification: Recommendation object 1 reaches user 1 through propagation behavior 1. User 1 may initiate propagation behavior 2 of recommendation object 2 to reach user 1 under the influence of propagation behavior 1. This influence probability is expressed as e⊙g. User 2 initiates propagation behavior of recommendation object 3 to achieve value activation, which is represented by the activation function σ.

[0119] Using an activation function, the incremental user value of the initiating user is calculated based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user. One possible approach is to directly calculate the incremental user value of the initiating user based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user using an activation function. Another possible approach is to calculate the incremental user value of the initiating user under the constraint of the forgetting decay coefficient, based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user using an activation function. No further limitation is imposed here.

[0120] For example, through activation functions Based on the second propagation behavior Behavioral characteristics and initiating user u i User characteristics are used to calculate the incremental user value of the initiating user. Based on the first propagation probability And the incremental user value of the initiating user Determine the propagation value gain of the initiating user

[0121] In the embodiments of this specification, by integrating user characteristics and propagation behavior characteristics, the potential influence of a single propagation behavior is accurately captured, solving the problem that traditional static models cannot reflect the dynamic interaction between users. By combining the first propagation probability and value increment calculation, the cascading effect of the propagation chain is explicitly quantified, enabling network value assessment to truly reflect the fission and diffusion process after user reach. Through activation functions, while ensuring calculation accuracy, the online real-time inference needs of large-scale user networks are supported.

[0122] In one optional embodiment of this specification, before calculating the user value increment of the initiating user based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user through the activation function, the following specific steps are further included: obtaining the historical time when the propagation value gain of the initiating user was determined in the last iteration; calculating the activation time interval between the current time and the historical time;

[0123] Using an activation function, the user value increment of the initiating user is calculated based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user. The specific steps include: calculating the forgetting decay coefficient based on the activation time interval; and calculating the user value increment of the initiating user based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user when the forgetting decay coefficient has not reached the decay coefficient threshold.

[0124] The historical time at which the propagation value gain of the initiating user was determined in the last iteration is the timestamp of the previous calculation of the propagation value gain for the same initiating user during the propagation queue processing. This time point is used to quantify the time-dependent decay of the propagation behavior's impact. For example, if user U2's propagation gain was calculated once at t = 10 minutes, that time point is the historical time.

[0125] The activation time interval between the current time and historical time is the time difference from the historical time to the current processing moment, used to calculate the attenuation of the propagation effect. The activation time interval uses an exponential decay model to quantify the timeliness of the propagation effect. For example, if user U2's last gain calculation was at 10:00 and the current time is 10:15, then the activation time interval is 15 minutes.

[0126] The forgetting decay coefficient is a dynamic weighting coefficient calculated based on the activation time interval, used to adjust the contribution of historical propagation behavior to the current value increment. The forgetting decay coefficient follows the Ebbinghaus forgetting curve, decaying exponentially over time. For example, it can be expressed by the formula α = exp(-λΔt), where λ is the decay rate parameter and Δt is the activation time interval. The decay coefficient threshold is a critical value used to determine whether the influence of propagation behavior has become ineffective; when the forgetting decay coefficient is below this threshold, the historical propagation effect is ignored. The decay coefficient threshold is determined through experimental data calibration and is typically set as an empirical value between 0.1 and 0.3. For example, setting the threshold α_threshold = 0.2, when α < 0.2, the value gain of the propagation chain is no longer calculated.

[0127] For example, the time when user U3's propagation gain was last calculated is t_prev = 10:00, the current time is t_now = 10:18, and the activation time interval Δt = 18 minutes is calculated. Using a decay rate of λ = 0.05 / min, the forgetting decay coefficient α = exp(-0.05 × 18) ≈ 0.41 is calculated. The threshold α_threshold is set to 0.3. Since 0.41 > 0.3, the value increment for this calculation continues.

[0128] In the embodiments of this specification, by quantifying the time-dependent decay characteristics of propagation behavior, a forgetting decay mechanism based on the activation time interval is introduced to solve the problem of value duplication calculation in continuous propagation scenarios. While ensuring the accuracy of dynamic modeling of the propagation chain, it effectively eliminates noise interference in the time dimension, making the network value gain assessment more consistent with the time-dependent laws of real propagation scenarios.

[0129] In one optional embodiment of this specification, step 402, obtaining the propagation queue, includes the following specific steps: obtaining the propagation network of the target recommendation object and the initiating user, wherein the propagation network includes a user set, a content set, and a propagation behavior set, and the initiating user is the delivery system of the target recommendation object; selecting a reachable user from the user set and selecting first content from the content set; selecting a first propagation behavior corresponding to the first content from the propagation behavior set, and calculating a first influence probability, wherein the first influence probability represents the probability that the reachable user is influenced by the first propagation behavior corresponding to the first content initiated by the initiating user; if the first influence probability is not zero, selecting a second propagation behavior corresponding to the first content from the propagation behavior set, and calculating a first propagation generation probability, wherein the first propagation generation probability is... The probability represents the likelihood that a user being influenced by a first propagation behavior will initiate a second propagation behavior corresponding to the second content. The first content and the second content correspond to the target recommendation objects. When the probability of generating the first propagation is not zero, the first propagation probability is determined based on the first influence probability and the first propagation generation probability. The first propagation probability represents the probability that the initiating user influences the user being influenced by the first propagation behavior to initiate a second propagation behavior. When the first propagation probability is greater than the propagation probability threshold, the first propagation probability is added to the propagation queue, the user being reached is updated to the initiating user, and the process returns to the steps of selecting the user being reached from the user set and selecting the first content from the content set. This process continues until all user pairs in the user set and all content in the content set have been selected, at which point the propagation queue is obtained.

[0130] The user set is the collection of all user nodes participating in the propagation process within a propagation network. The user set constitutes the basic nodes of the propagation network, and each user node can contain static attributes (such as user ID, basic profile) and / or dynamic behavioral characteristics (such as historical propagation records). For example, in the propagation network of a community content platform, the user set is represented as U = {u...} i Each user has a corresponding user value.

[0131] A content set is a collection of content objects that serve as a medium for propagation within a propagation network. Content, as the carrier of propagation behavior, connects different user nodes, and its attributes influence the probability of propagation and the gain of value. For example, a content set can be represented as T = {t}. i} represents the content propagated in the network, where t iIt can be in the form of advertising images and text, product review videos, or other specific content formats.

[0132] A set of communication behaviors is a collection of interactive behaviors among users that enable information exchange. Different communication behaviors have varying levels of efficiency and value conversion rates. For example, a set of communication behaviors can be characterized as... This refers to the dissemination of certain content (channels), such as exposure of user-generated content, reading of user-generated content, reading through comments, reading through sharing, etc.

[0133] The probability of first impact is the initial probability that a user is influenced by the initiating user's first dissemination action. The probability of first impact reflects the original influence of the first hop in the dissemination chain and is affected by the strength of the user relationship and the relevance of the content. For example, after user A shares a beauty advertisement, the probability e = 15% that their follower B sees the content.

[0134] The probability of generating a first-stage spread is the conditional probability that a user will initiate a second-stage spread after being influenced by the first-stage spread behavior. The probability of generating a first-stage spread quantifies the chain reaction potential of the spread behavior and reflects the user's willingness to reproduce content. For example, the probability g = 20% that user B will share the ad a second time after seeing it.

[0135] The first propagation probability is the joint probability of the first influence probability and the propagation generation probability, representing the likelihood of the complete propagation path being realized. For example, in the above example, the first propagation probability p = 15% × 20% = 3%.

[0136] The propagation probability threshold is a critical probability value used to determine whether a propagation path is valid. It is used to filter out inefficient propagation paths and improve computational efficiency. For example, if ε = 1%, the propagation path is ignored when the first propagation probability ε < 1%.

[0137] Figure 7 This document shows a second schematic diagram illustrating the probability in a propagation evaluation method for a recommendation object provided in one embodiment of this specification:

[0138] A basic propagation process includes:

[0139] Generation behavior g: The first user u1 is reached by the first content ω1, and then generates the second content ω2; Propagation behavior e: The second user u2 is reached by the second content ω2; Activation behavior σ: The second user u2 finally generates value activation.

[0140] The generation behavior is represented as e⊙g: after the first content ω1 is generated, the first user u1 generates the second content ω2; the activation behavior is represented as e⊙σ: after the second content ω2 is generated, the second user u2 undergoes value activation.

[0141] For example, the initial user value for multiple users Summing yields the initial network value of the propagation network for the target recommendation object. Let's define a virtual initiating user u0, representing the recommendation system, and establish an edge with the first user directly reached by the recommended object; the recommended object has only one propagation behavior ω = ad_lick, and no value or activation function is defined; its first influence probability... Among them, the first influence probability If the information is related to the document being reviewed, then the probability is 0; otherwise, it is 0. (In the first influence probability) If the probability is not zero, select the second propagation behavior corresponding to the first content from the set of propagation behaviors, and calculate the generation probability of the first propagation. The probability is 0 if and only if it relates to the recommended object, otherwise it is 0. This probability is generated during the first propagation. When the value is not zero, based on the first influence probability and the probability of first propagation generation Determine the first propagation probability if If the threshold ε is exceeded, then the binary pair will be... Add to the propagation queue Q, return to iterative execution, until the selection of each user pair in the user set and each content in the content set is completed, and the propagation queue Q is obtained.

[0142] In the embodiments described in this specification, a systematic modeling of information propagation paths is achieved by constructing a three-element network topology that includes user nodes, content carriers, and propagation behaviors. A dual probability verification mechanism is employed: first, the impact probability of the initial propagation behavior is calculated; then, the generation probability of secondary propagation behaviors is evaluated. The two mechanisms jointly determine the effectiveness of the complete propagation link. An iterative update mechanism transforms reach nodes into new propagation source nodes, constructing a cascading propagation model. This model overcomes the limitations of traditional fixed-level propagation models and can fully capture the long-term value gain of the propagation network.

[0143] In one optional embodiment of this specification, calculating the first influence probability includes the following specific steps: using a prediction model, based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user, calculating the first influence probability.

[0144] The prediction model is a machine learning model used to predict the probability of a propagation behavior occurring. The model's input includes the features of the recommended object, user attributes, and propagation behavior, and its output is a propagation probability value between 0 and 1. Optionally, the model training employs negative sampling and cross-entropy loss function optimization. For example, a deep neural network (DNN) model can be used, with the input layer taking the recommended object category (e.g., beauty), the number of followers of the initiating user (e.g., 100,000), and the target user's interest tags (e.g., "skincare"), and the output layer predicting the propagation probability using a sigmoid activation function.

[0145] During training, the behavioral information of the first propagation action ω1 and the second propagation action ω2 is clear. However, during propagation, apart from the recommendation object information in the first hop, the content information in subsequent propagation channels is completely unknown. One way to solve this problem is to ignore the content information related to the propagation channels and rely solely on the user's own information. Considering the cost of actual inference, the prediction model can be designed as a multi-tower model. Figure 8 This specification shows one of the structural schematic diagrams of a prediction model in a propagation evaluation method for recommending objects, provided in one embodiment:

[0146] The prediction model calculates the first influence probability based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user. One possible approach is to input the recommendation object information of the target recommendation object into the recommendation object tower and extract the recommendation object features; input the user information of the initiating user into the first user tower and extract the user characteristics of the initiating user; input the user information of the reaching user into the second user tower and extract the user characteristics of the reaching user; and input the recommendation object features, the user characteristics of the initiating user, and the user characteristics of the reaching user into the prediction layer of the multi-tower model for feature fusion to calculate the first influence probability e⊙g.

[0147] In the embodiments described in this specification, the calculation of the first influence probability is simplified by using a prediction model, thereby improving the accuracy and efficiency of the calculation.

[0148] In one optional embodiment of this specification, the prediction model is a multi-tower model. The prediction model calculates a first influence probability based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user. This includes the following specific steps: inputting the user information of the initiating user, the behavioral information of the second propagation behavior, and the recommendation object information of the target recommendation object into the propagation behavior tower of the multi-tower model to extract the fusion behavioral features of the second propagation behavior; inputting the user information of the initiating user into the first user tower of the multi-tower model to extract the user characteristics of the initiating user; inputting the user information of the reaching user into the second user tower of the multi-tower model to extract the user characteristics of the reaching user; and inputting the fusion behavioral features of the second propagation behavior, the user characteristics of the initiating user, and the user characteristics of the reaching user into the prediction layer of the multi-tower model for feature fusion to calculate the first influence probability.

[0149] The multi-tower model is a machine learning model architecture composed of multiple independent feature extraction towers. Each tower processes different types of input features (such as users, recommendation objects, and propagation behavior), and finally, a fusion layer jointly predicts the target value. The multi-tower model processes heterogeneous data through separate towers, avoiding feature cross-interference and improving the model's generalization ability. For example, the user tower processes user profiles, the recommendation object tower processes advertising content features, and the propagation behavior tower models interactive dynamics. For example, the user tower inputs user historical behavior embeddings, the recommendation object tower inputs recommendation object image and text vectors, and the fusion layer outputs a click-through rate prediction. The propagation behavior tower is a sub-network in the multi-tower model specifically designed to model the dynamic features of propagation behavior. It encodes features such as propagation behavior type (e.g., sharing, forwarding), content carrier (image / text / video), and spatiotemporal context into a fusion representation. For example, inputting user A's sharing behavior + beauty video features outputs a propagation potential vector. The first user tower is an independent sub-network that processes the features of the initiating user in the propagation chain. The first user tower extracts embeddings of user static attributes (age, interest tags) and dynamic behavior sequences (historical interactions). For example, the first user tower inputs the number of KOLs' followers and their activity level. The second user tower is an independent sub-network that processes the characteristics of users reached in the propagation chain. The second user tower extracts embeddings of users' static attributes (age, interest tags) and dynamic behavioral sequences (historical interactions). For example, the second user tower inputs the target user's click history. The prediction layer is a fully connected network that aggregates the outputs of each tower in the multi-tower model and generates the final predicted value. The prediction layer fuses heterogeneous features through attention mechanisms or weighted concatenation, outputting probability values ​​(such as propagation probability, value gain). For example, user tower output (0.3, 0.7) + recommendation target tower output (0.5) → after fusion, the output propagation probability is 0.6.

[0150] The user information of the initiating user is a set of attributes of user nodes that actively initiate dissemination behavior in the dissemination network. The user information of the initiating user includes features such as static profile, dynamic behavior, and social relationships, which are used to quantify their influence as a dissemination source. For example, static profile: female, 25 years old, Shanghai, 500,000 followers; dynamic profile: 12% sharing rate of beauty-related notes in the past 30 days; social profile: social influence score of 0.8 (top 10%).

[0151] The user information reaching the recipient is a set of features of the recipient user nodes directly affected by the dissemination behavior. This user information includes features such as static profiles, dynamic behaviors, and social relationships, used to quantify their sensitivity to the disseminated content and their potential for secondary dissemination. For example, response: all three previous ad clicks resulted in a purchase; interests: 35% of their reading was skincare-related content; social: the strength of their local relationship with the initiating user, A, is 0.6.

[0152] The behavioral information of the second-level dissemination behavior is metadata about the secondary dissemination actions initiated by users after being influenced by the initial dissemination. This behavioral information is used to quantify the extension capability and value-added potential of the dissemination chain. For example, the behavioral information of user B's secondary dissemination behavior includes: Type: forwarding to social media + tagging 3 friends; Content: generating a 50-word user experience; Timeliness: dissemination completed within 1.5 hours after initial exposure. For example, the recommended target information for a skincare serum includes: Content: video tutorial + ingredient analysis (text and images); Value: historical conversion rate 8%, average order value 299; Suitability: optimal for two-tier dissemination by KOLs and ordinary users.

[0153] The target recommendation object's recommendation information is a set of features of the entity (advertisement / product / content) that serves as the core object of dissemination. This recommendation information determines the appropriate dissemination path and user matching strategy.

[0154] The fusion behavior characteristics of the second propagation behavior are the deep encoding characteristics of the second propagation behavior output by the propagation behavior tower, which integrates multi-dimensional signals from users, content, and context. The propagation effect modeling is enhanced by cross features (such as user-content matching degree). For example, the sharing behavior characteristics of user C for sneaker advertisements = [user's sports preference 0.9 × advertisement's sports attribute 0.7 + timeliness factor 0.5].

[0155] The user characteristics of the initiating user are the deep coding characteristics of the user node that actively triggers the propagation behavior in the propagation network, which quantifies its potential influence as the source of propagation.

[0156] The user characteristics that reach users are the deep coding characteristics of the receiving user nodes that are directly affected by the propagation behavior, which quantifies their response potential to the propagated content.

[0157] During training, the behavioral information of the first propagation behavior ω1 and the second propagation behavior ω2 is clear. However, during propagation, apart from the recommended object information in the first hop, the content information in subsequent propagation channels is completely unknown. One way to solve this problem is to attempt to generate propagation channel information and use the generated features in the actual inference process. During training, the calculated result of the fused behavioral features of the generated second propagation behavior ω2 and the actual result loss need to be added to the total loss to help train the second propagation behavior generation tower.

[0158] Figure 9The second schematic diagram illustrates the structure of the prediction model in a propagation evaluation method for recommendation objects provided in one embodiment of this specification: The user information of the initiating user, the behavioral information of the second propagation behavior, and the recommendation object information of the target recommendation object are input into the propagation behavior tower of the multi-tower model to extract the fusion behavioral features of the second propagation behavior; the user information of the initiating user is input into the first user tower of the multi-tower model to extract the user features of the initiating user; the user information of the reached user is input into the second user tower of the multi-tower model to extract the user features of the reached user; the fusion behavioral features of the second propagation behavior, the user features of the initiating user, and the user features of the reached user are input into the prediction layer of the multi-tower model for feature fusion, and the first influence probability e⊙g is calculated.

[0159] In the embodiments described in this specification, the accuracy of the calculation is improved by dynamically refining the calculation of the first influence probability through a multi-tower model.

[0160] In one optional embodiment of this specification, calculating the first propagation generation probability includes the following specific steps: using a prediction model, based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user, calculating the first propagation generation probability.

[0161] like Figure 8 As shown, the prediction model calculates the first propagation generation probability based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user. One possible approach is to input the recommendation object information of the target recommendation object into the recommendation object tower and extract the recommendation object features; input the user information of the initiating user into the first user tower and extract the user characteristics of the initiating user; input the user information of the reaching user into the second user tower and extract the user characteristics of the reaching user; input the recommendation object features, the user characteristics of the initiating user, and the user characteristics of the reaching user into the prediction layer of the multi-tower model for feature fusion and calculate the first propagation generation probability e⊙σ.

[0162] In the embodiments described in this specification, the calculation of the first propagation generation probability is simplified by using a prediction model, thereby improving the accuracy and efficiency of the calculation.

[0163] In one optional embodiment of this specification, the prediction model is a multi-tower model. The prediction model calculates the first propagation generation probability based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user. This includes the following specific steps: inputting the user information of the initiating user, the behavioral information of the second propagation behavior, and the recommendation object information of the target recommendation object into the propagation behavior tower of the multi-tower model to extract the fusion behavioral features of the second propagation behavior; inputting the user information of the initiating user into the first user tower of the multi-tower model to extract the user characteristics of the initiating user; inputting the user information of the reaching user into the second user tower of the multi-tower model to extract the user characteristics of the reaching user; and inputting the fusion behavioral features of the second propagation behavior, the user characteristics of the initiating user, and the user characteristics of the reaching user into the prediction layer of the multi-tower model for feature fusion to calculate the first propagation generation probability.

[0164] like Figure 9 As shown, the user information of the initiating user, the behavioral information of the second propagation behavior, and the recommendation object information of the target recommendation object are input into the propagation behavior tower of the multi-tower model to extract the fusion behavioral features of the second propagation behavior; the user information of the initiating user is input into the first user tower of the multi-tower model to extract the user features of the initiating user; the user information of the reached user is input into the second user tower of the multi-tower model to extract the user features of the reached user; the fusion behavioral features of the second propagation behavior, the user features of the initiating user, and the user features of the reached user are input into the prediction layer of the multi-tower model for feature fusion, and the generation probability e⊙g of the first propagation is calculated.

[0165] In the embodiments described in this specification, the calculation of the first propagation generation probability is dynamically refined by using a multi-tower model, thereby improving the accuracy of the calculation.

[0166] In one optional embodiment of this specification, before adding the first propagation probability to the propagation queue when the first propagation probability is greater than the propagation probability threshold, the following specific steps are further included: calculating the propagation association strength between the initiating user and the reaching user based on the feature similarity between the user characteristics of the initiating user and the user characteristics of the reaching user;

[0167] If the first propagation probability is greater than the propagation probability threshold, the first propagation probability is added to the propagation queue, including the following specific steps: if the propagation association strength is greater than the strength threshold and the first propagation probability is greater than the propagation probability threshold, the first propagation probability is added to the propagation queue.

[0168] Propagation association strength is a quantitative indicator characterizing the effectiveness of propagation links between user nodes, combining user feature similarity and historical interaction frequency. For example, it can be calculated using cosine similarity or embedding distance to filter inefficient propagation paths. For instance, if users A and B have 80% overlap in their interest tags, their propagation association strength is 0.8 (with a threshold of 0.5). The strength threshold is the critical value for determining whether a propagation association is effective; edges below this threshold will be pruned. For example, setting the strength threshold to 0.6, if the association strength between users C and D is 0.4, then this propagation path will be ignored.

[0169] It should be noted that, if the influence of the first propagation action ω1 is ignored, or if the influence of the first propagation action ω1 is considered to be very weak, the propagation effect between nodes is only related to the initiating user u1 and the reaching user u2. Offline calculation g⊙e(u1,u2,.) can be used to prune some very weak propagation edges, so that the propagation network is transformed from a complete graph into a sparse graph, thereby saving a lot of time in the online inference process.

[0170] For example, the embedding of user U1 is calculated offline: emb(U1) = [0.2, 0.7, 0.1]. The Approximate Nearest Neighbor (ANN) algorithm (Hierarchical Navigable Small World (HNSW) algorithm, Inverted File with Product Quantization (IVF-PQ) algorithm, etc.) is used to recall the top 3 similar users: U2 (0.85), U3 (0.82), and U4 (0.31). The propagation edges of U2 and U3 (similarity > 0.8) are retained, and the edges of U4 are removed.

[0171] In the embodiments described in this specification, pruning is used to avoid calculating invalid propagation paths, which not only reduces computational redundancy but also preserves critical paths, thereby improving the efficiency of network enhancement planning.

[0172] In one optional embodiment of this specification, before selecting the user to be reached from the user set and selecting the first content from the content set, the method further includes the following specific steps: clustering user nodes in the propagation network based on the feature similarity between user features of users in the user set to obtain an updated propagation network, wherein the updated propagation network includes the updated user set.

[0173] The updated propagation network is a topology optimized through user clustering, where similar user nodes are merged into cluster nodes. The updated network retains the propagation characteristics of the original network, but the number of nodes is significantly reduced (e.g., millions of users are clustered into thousands of clusters). For example, in a community content platform, 100,000 beauty-interested users are clustered into 50 clusters, each representing a specific user segment. The updated user set consists of cluster nodes formed after clustering, including ordinary cluster nodes and individually retained large K nodes. Each cluster user node has aggregation characteristics (e.g., average interest vector), while large K nodes (e.g., top KOLs) remain independent. For example, the user set changes from {U1, U2, ..., U1 million} to {C1, C2, ..., C10 million, KOL1, KOL2}.

[0174] It should be noted that, considering the computational burden, clustering user nodes can reduce computational load and prediction difficulty. Clustered nodes reflect the aggregation of similar / close users. The compressed clustered nodes are treated as new user nodes and participate in the original propagation graph calculations. The results of clustering should have the following characteristics: Isotropy: Regardless of which user enters the cluster, the activation, propagation, and other effects produced within the cluster are essentially the same; Additivity: The calculation of activation and generating functions for clusters, and the calculation of propagation functions between clusters, are equivalent to the expectation of the results obtained according to the users participating in the calculation within the cluster.

[0175] Meanwhile, in order to minimize the loss of computational accuracy, user clusters will be designed as follows: some large K nodes may have a very large propagation effect in the network and will be clustered separately; the first node directly reached by the advertisement will contribute a large value and propagation effect in the network, so the prediction of the first hop and the second hop is very important. We will decompose the first node from the cluster and calculate it separately.

[0176] Figure 10 This specification illustrates a schematic diagram of user clustering in a method for evaluating the propagation of a recommendation object, as provided in one embodiment:

[0177] The original propagation network structure contains multiple dispersed user nodes (such as U1, U2, etc.). The recommendation target delivery system first reaches user node U1 and then repeatedly reaches and propagates within the propagation network.

[0178] The user nodes in the above propagation network are clustered to obtain an updated propagation network. The original user nodes are merged into multiple cluster nodes (C1-C4), and each cluster node retains the common propagation characteristics of users within the group through feature fusion. High-value individual nodes (such as U1) and recommendation object delivery nodes (AD) are retained to ensure accurate modeling of key propagation paths. Arrows indicate the optimized propagation links, and the edge weights between cluster nodes are the mean of the propagation probabilities of users within the group (e.g., the edge weight of C1→C2 = Σe(u,v) / |C1||C2|).

[0179] In the embodiments described in this specification, clustering not only reduces computational redundancy but also preserves critical paths, thereby improving the efficiency of network enhancement planning.

[0180] With the above Figures 1 to 10 The embodiments correspond to the instructions. Figure 11 A timing diagram illustrating a propagation evaluation method for a recommendation object provided in one embodiment of this specification is shown:

[0181] Initial network construction phase:

[0182] Step 1102: Load the original set of user nodes and the initial propagation link.

[0183] Step 1104: The recommended target delivery system selects the first-hop user node U1 for initial outreach.

[0184] Dynamic propagation stage:

[0185] Step 1106: User U1 triggers the first propagation behavior, influencing user U2 with a probability p1 = 15%.

[0186] Step 1108: After being triggered, user U2 generates user-generated content, which is then propagated to user U3 with a probability of p2 = 20%.

[0187] Step 1110: User U3 generates a value activation behavior, contributing an increment Δv = 0.12.

[0188] Cluster optimization phase:

[0189] Step 1112: If the feature similarity of users {U4, U5, U6} is detected to be >0.9, they are merged into cluster nodes.

[0190] Step 1114: Recalculate the inter-cluster propagation probability.

[0191] Pruning stage:

[0192] Step 1116: Filter inefficient propagation links.

[0193] Step 1118: Preserve critical propagation links.

[0194] Valuation phase:

[0195] Step 1120: Iteratively calculate the network value gain and output the target network value.

[0196] Step 1122: Update network value gain = f(propagation behavior, target network value).

[0197] See Figure 12 , Figure 12 A flowchart illustrating a propagation planning method for a recommendation object according to an embodiment of this specification is shown, including the following specific steps:

[0198] Step 1202: Obtain the network value gain of the propagation network of the target recommendation object, wherein the network value gain is the difference between the target network value and the initial network value obtained after iteratively updating the user value based on each propagation probability in the propagation queue. The propagation queue includes multiple propagation probabilities.

[0199] For example, the NetAdvv values ​​of each user node are pre-generated and cached through an offline computing module, and the network enhancement API is called in real time to dynamically calculate the values ​​from user behavior logs.

[0200] Step 1204: Based on network value gain, generate a target user reach sequence for the target recommendation object, wherein the target user reach sequence includes a sequence of user nodes arranged according to network value gain.

[0201] The target user reach sequence for the recommended object is a dynamically sorted order of user node access based on network value gain, used to maximize the propagation effect. The target user reach sequence is a user reach priority sequence generated through algorithm optimization, reflecting the strategic value of different user nodes in the propagation chain. For example, in beauty advertising, the sequence [KOL1→Cluster C3→KOL2] indicates prioritizing reaching top KOLs to activate secondary propagation.

[0202] For example, a greedy algorithm is used to sort user nodes in descending order of NetAdvv, and a long-term payoff generation sequence is optimized based on a reinforcement learning model:

[0203]

[0204] Where, d u Given the user depth, if NetAdv = [0.3, 0.5, 0.1], generate the sequence [u2, u1, u3].

[0205] In the embodiments of this specification, the dynamically calculated network value gain is used as the basis for sequence generation, which solves the problem that static sorting cannot adapt to real-time propagation changes.

[0206] See Figure 13, Figure 13 A flowchart illustrating a method for propagating a recommended object according to an embodiment of this specification is shown, including the following specific steps:

[0207] Step 1302: Obtain the target user reach sequence of the target recommended object, wherein the target user reach sequence is a sequence of user nodes reached by the target recommended object generated based on the network value gain of the propagation network, the network value gain is the difference between the target network value obtained after iteratively updating the user value of each propagation probability in the propagation queue and the initial network value, and the propagation queue includes multiple propagation probabilities.

[0208] For example, the target user reach sequence [u2, u1, u3] of the target recommendation object is obtained.

[0209] Step 1304: Based on the target user reach sequence, perform propagation operations on the content of the target recommendation object to reach users sequentially.

[0210] For example, propagation is performed on the sequence S = [u2, u1, u3]:

[0211] First round of reach to u2: Operation type: strong reminder push; Content carrier: video ad + discount code; Triggering condition: executed immediately when t=0;

[0212] Second round of reach to u1: Operation type: Social media pinning; Content carrier: Image and text review; Trigger condition: t+ΔT (ΔT=2h);

[0213] Last round of reach u3: Operation type: private message reminder; Content carrier: user-generated content collection; Trigger condition: t+2ΔT.

[0214] In the embodiments described in this specification, the optimal scheduling of propagation resources is achieved by dynamically loading the pre-calculated target user reach sequence, which solves the problem of resource waste caused by traditional fixed-order propagation, improves the propagation efficiency of recommended objects, reduces ineffective propagation, and maximizes the value of the propagation network.

[0215] The following is in conjunction with the appendix Figure 14 Taking the application of the recommendation target dissemination evaluation method provided in this manual to advertising distribution on a content community platform as an example, this paper further explains the recommendation target dissemination evaluation method. Specifically, Figure 14 This specification illustrates a flowchart of a method for evaluating the propagation of recommended targets in advertising distribution on a content community platform, according to an embodiment of this specification. The method includes the following specific steps:

[0216] Step 1402: Obtain the propagation network of the target advertisement and the initiating user. The propagation network includes user nodes corresponding to multiple users, the initial network value includes the initial user value of multiple users, the propagation network includes user set, user-generated content set and propagation behavior set, and the initiating user is the target advertisement delivery system.

[0217] For example, consider a propagation network G = (U,T,E), where the user set U = {u1,u2,...,un}, the user-generated content set T = {t1,t2,...,tm}, and the propagation behavior set E = {e1,e2,...,ek}. The initiating user is the advertising platform AD, and its user characteristics are represented as emb(AD) = [0,0,...,1].

[0218] Step 1404: Aggregate the initial user value of multiple users to obtain the initial network value of the target advertisement's propagation network.

[0219] For example, the initial network value V_init = Σv(ui), where v(u1) = 0.3 (based on 100,000 fans × 3‰ interaction rate), v(u2) = 0.15, ..., aggregated to get V_init = 2.7.

[0220] Step 1406: Select the user to reach from the user set, and select the first user-generated content from the user-generated content set.

[0221] For example, select to reach user u2 (feature emb(u2) = [0.7, 0.2, 0.1]) and user-generated content t1 (beauty review video).

[0222] Step 1408: Select the first propagation behavior corresponding to the first user-generated content from the propagation behavior set, and calculate the first influence probability, where the first influence probability represents the probability that the reaching user is affected by the first propagation behavior corresponding to the first user-generated content initiated by the initiating user.

[0223] For example, calculate e(AD→u2,t1)=σ(W1·[emb(AD)||emb(u2)||emb(t1)]+b1)=0.12, where σ is the sigmoid function and W1 is the weight matrix.

[0224] Step 1410: If the first influence probability is not zero, select the second propagation behavior corresponding to the first user-generated content from the propagation behavior set, and calculate the first propagation generation probability. The first propagation generation probability represents the probability that the reached user is influenced by the first propagation behavior to initiate the second propagation behavior corresponding to the second user-generated content. The first user-generated content and the second user-generated content correspond to the target advertisement.

[0225] For example, calculate g(u2,t1→t2)=σ(W2·[emb(u2)||Δemb(t1→t2)]+b2)=0.08, where Δemb(t1→t2) represents the content-derived feature.

[0226] Step 1412: If the first propagation generation probability is not zero, determine the first propagation probability based on the first influence probability and the first propagation generation probability, wherein the first propagation probability represents the propagation probability that the initiating user influences the reaching user to initiate the second propagation behavior through the first propagation behavior.

[0227] For example, p(AD→u2→t2)=e(AD→u2,t1)×g(u2,t1→t2)=0.12×0.08=0.0096.

[0228] Step 1414: If the first propagation probability is greater than the propagation probability threshold, add the first propagation probability to the propagation queue, update the reached user to the initiating user, and return to execute step 1404.

[0229] For example, if the threshold ε = 0.005, add (u2, 0.0096) to queue Q and update the current initiating user to u2.

[0230] Step 1416: Obtain the propagation queue after completing the selection of each user pair in the user set and each user-generated content in the user-generated content set.

[0231] For example, the final queue Q = {(u2,0.0096),(u3,0.015),...,(u5,0.007)} contains a total of k valid propagation paths.

[0232] Step 1418: Select the first propagation probability from the propagation queue.

[0233] For example, select (u2, 0.0096) from the Q header as the current processing item.

[0234] Step 1420: Using an activation function, calculate the user value increment of the initiating user based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user. Based on the first propagation probability and the user value increment of the initiating user, determine the propagation value gain of the initiating user.

[0235] For example, Δv(u2) = 0.04, and the propagation value gain ΔV = p × Δv = 0.0096 × 0.04 ≈ 0.00038.

[0236] Step 1422: Based on the initial user value and propagation value gain of the initiating user, determine the target user value of the initiating user, and return to execute step 1418.

[0237] For example, update v'(u2)=v(u2)+ΔV=0.3+0.00038=0.30038, and continue processing the next item in the queue.

[0238] Step 1424: Until the propagation queue is empty, determine the target network value of the propagation network based on the target user value of multiple users, and determine the network value gain of the propagation network based on the target network value and the initial network value.

[0239] For example, the final V_final = 3.2, and the network value gain ΔV_net = 3.2 - 2.7 = 0.5.

[0240] In the embodiments described in this specification, a systematic modeling of the advertising propagation path is achieved by constructing a three-element propagation network structure that includes user nodes, content carriers, and propagation behaviors. A dual probability verification mechanism is employed: first, the initial reach impact probability is calculated, and then the secondary propagation generation probability is evaluated, accurately quantifying the effectiveness of the complete propagation chain. Dynamic queue management enables iterative expansion of the propagation path, and combined with the cumulative calculation of value gain, a network value gain reflecting the true propagation effect is obtained. This solves the problem that traditional static models cannot capture cascading propagation, precisely and effectively improving the accuracy of advertising value assessment and providing data support for precise advertising placement.

[0241] Corresponding to the above method embodiments, this specification also provides recommended object platform embodiments. Figure 15 A schematic diagram of the structure of a recommended object platform provided in one embodiment of this specification is shown. For example... Figure 15 As shown, the recommendation platform 1500 includes an offline module 1510 and an online module 1520;

[0242] Offline module 1510 is used for: obtaining the initial network value of the propagation network of the target recommendation object and a propagation queue, wherein the propagation network includes user nodes corresponding to multiple users, the initial network value includes the initial user value of multiple users, and the propagation queue includes multiple propagation probabilities; selecting a first propagation probability from the propagation queue, wherein the first propagation probability represents the probability that the initiating user influences the reaching user to initiate a second propagation behavior through the first propagation behavior; determining the propagation value gain of the initiating user based on the first propagation probability; determining the target user value of the initiating user based on the initial user value and the propagation value gain, returning to the step of selecting the first propagation probability from the propagation queue, until the propagation queue is empty, determining the target network value of the propagation network based on the target user value of multiple users; determining the network value gain of the propagation network based on the target network value and the initial network value; and generating the target user reaching sequence of the target recommendation object based on the network value gain.

[0243] Online module 1520 is used to: perform propagation operations on the content of the target recommendation object to reach users sequentially based on the target user reach sequence.

[0244] In this embodiment, the offline module obtains the initial network value of the propagation network of the target recommendation object, as well as a propagation queue containing multiple propagation probabilities. Each propagation probability represents the probability that an initiating user will influence a reaching user to initiate a subsequent second propagation action through a first propagation behavior, thereby directly modeling the dynamic path of the propagation chain. Based on this, the module iteratively selects propagation probabilities and calculates the propagation value gain of the initiating user for a single user node. The progressively accumulated gain simulates the dynamic diffusion mechanism of the propagation behavior, reflecting its actual influence in triggering a chain reaction. The module then updates the target user value of a single user node and finally obtains the target network value of the entire propagation network. By comparing the target network value with the initial network value, the network value gain is derived. This gain quantifies the true contribution of the dynamic propagation network, making network value assessment more accurate and reliable. The network value gain dynamically calculated by the online module serves as the basis for sequence generation, solving the problem that static sorting cannot adapt to real-time propagation changes.

[0245] The above is an illustrative scheme of a recommendation platform according to this embodiment. It should be noted that the technical solution of this recommendation platform belongs to the same concept as the aforementioned technical solutions of the recommendation object propagation evaluation method, recommendation object propagation planning method, and recommendation object propagation method. Details not described in detail in the technical solution of the recommendation platform can be found in the descriptions of the aforementioned technical solutions of the recommendation object propagation evaluation method, recommendation object propagation planning method, or recommendation object propagation method.

[0246] Figure 16 A structural block diagram of a computing device according to one embodiment of this specification is shown. The components of the computing device 1600 include, but are not limited to, a memory 1610 and a processor 1620. The processor 1620 is connected to the memory 1610 via a bus 1630, and a database 1650 is used to store data.

[0247] The computing device 1600 also includes an access device 1640, which enables the computing device 1600 to communicate via one or more networks 1660. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 1640 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Controller (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0248] In one embodiment of this specification, the above-described components of the computing device 1600 and Figure 16 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 16 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0249] The computing device 1600 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 1600 can also be a mobile or stationary server.

[0250] The processor 1620 is used to execute the following computer program / instruction, which, when executed by the processor, implements the steps of the above-mentioned propagation evaluation method or propagation planning method for the recommended object.

[0251] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device belongs to the same concept as the technical solutions of the aforementioned propagation evaluation method and propagation planning method for the recommended object. Details not described in detail in the technical solution of the computing device can be found in the descriptions of the technical solutions of the aforementioned propagation evaluation method or propagation planning method for the recommended object.

[0252] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described method for evaluating the propagation of recommended objects or the method for planning the propagation of recommended objects.

[0253] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solutions of the aforementioned method for evaluating the propagation of recommended objects and the method for planning the propagation of recommended objects. Details not described in detail in the technical solution of the storage medium can be found in the descriptions of the technical solutions of the aforementioned method for evaluating the propagation of recommended objects or the method for planning the propagation of recommended objects.

[0254] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method for evaluating the propagation of recommended objects or the method for planning the propagation of recommended objects.

[0255] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product belongs to the same concept as the technical solutions of the aforementioned method for evaluating the propagation of recommended objects and the method for planning the propagation of recommended objects. Details not described in detail in the technical solution of the computer program product can be found in the descriptions of the technical solutions of the aforementioned method for evaluating the propagation of recommended objects or the method for planning the propagation of recommended objects.

[0256] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0257] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0258] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0259] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0260] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for evaluating the propagation of recommendations, characterized in that, include: Obtain the initial network value of the propagation network of the target recommendation object, and the propagation queue, wherein the propagation network includes user nodes corresponding to multiple users, the initial network value includes the initial user value of multiple users, and the propagation queue includes multiple propagation probabilities; A first propagation probability is selected from the propagation queue, wherein the first propagation probability represents the probability that the initiating user influences the reaching user to initiate a second propagation behavior through the first propagation behavior; Based on the first propagation probability, determine the propagation value gain of the initiating user; Based on the initial user value of the initiating user and the propagation value gain, the target user value of the initiating user is determined, and the step of selecting the first propagation probability from the propagation queue is returned to be executed until the propagation queue is empty. Based on the target user values ​​of the multiple users, the target network value of the propagation network is determined. Based on the target network value and the initial network value, the network value gain of the propagation network is determined.

2. The method according to claim 1, characterized in that, Before determining the propagation value gain of the initiating user based on the first propagation probability, the method further includes: The user value increment of the initiating user is calculated using an activation function based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user. The step of determining the propagation value gain of the initiating user based on the first propagation probability includes: Based on the first propagation probability and the user value increment of the initiating user, the propagation value gain of the initiating user is determined.

3. The method according to claim 2, characterized in that, Before calculating the user value increment of the initiating user based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user using an activation function, the method further includes: Obtain the historical time when the propagation value gain of the initiating user was determined in the last iteration; Calculate the activation time interval between the current time and the historical time; The step of calculating the user value increment of the initiating user based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user through an activation function includes: Calculate the forgetting decay coefficient based on the activation time interval; If the forgetting decay coefficient does not reach the decay coefficient threshold, the user value increment of the initiating user is calculated using an activation function based on the behavioral characteristics of the second propagation behavior and the user characteristics of the initiating user.

4. The method according to any one of claims 1-3, characterized in that, Obtain the propagation queue, including: Obtain the propagation network of the target recommendation object and the initiating user, wherein the propagation network includes a user set, a content set and a propagation behavior set, and the initiating user is the delivery system of the target recommendation object; Select users to reach from the user set, and select first content from the content set; Select the first propagation behavior corresponding to the first content from the set of propagation behaviors, and calculate the first influence probability, wherein the first influence probability represents the probability that the reaching user is affected by the first propagation behavior corresponding to the first content initiated by the initiating user; When the first influence probability is not zero, the second propagation behavior corresponding to the first content is selected from the propagation behavior set, and the first propagation generation probability is calculated. The first propagation generation probability represents the probability that the user is influenced by the first propagation behavior to initiate the second propagation behavior corresponding to the second content. The first content and the second content correspond to the target recommendation object. When the first propagation generation probability is not zero, a first propagation probability is determined based on the first influence probability and the first propagation generation probability, wherein the first propagation probability represents the probability that the initiating user influences the reaching user to initiate the second propagation behavior through the first propagation behavior; If the first propagation probability is greater than the propagation probability threshold, the first propagation probability is added to the propagation queue, the reached user is updated to the initiating user, and the steps of selecting the reached user from the user set and selecting the first content from the content set are returned to be executed until the selection of each user pair in the user set and each content in the content set is completed, and the propagation queue is obtained.

5. The method according to claim 4, characterized in that, The calculation of the first influence probability includes: The prediction model calculates the first probability of influence based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reached user.

6. The method according to claim 5, characterized in that, The prediction model is a multi-tower model; the calculation of the first influence probability based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reached user using the prediction model includes: The user information of the initiating user, the behavioral information of the second propagation behavior, and the recommendation object information of the target recommendation object are input into the propagation behavior tower of the multi-tower model to extract the fusion behavioral features of the second propagation behavior; Input the user information of the initiating user into the first user tower of the multi-tower model, and extract the user features of the initiating user; The user information of the reached users is input into the second user tower of the multi-tower model, and the user features of the reached users are extracted. The fusion behavior features of the second propagation behavior, the user features of the initiating user, and the user features of the reaching user are input into the prediction layer of the multi-tower model for feature fusion, and the first influence probability is calculated.

7. The method according to claim 4, characterized in that, The calculation of the first propagation generation probability includes: The prediction model calculates the first propagation generation probability based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user.

8. The method according to claim 7, characterized in that, The prediction model is a multi-tower model; the calculation of the first propagation generation probability based on the recommendation object information of the target recommendation object, the user characteristics of the initiating user, and the user characteristics of the reaching user using the prediction model includes: The user information of the initiating user, the behavioral information of the second propagation behavior, and the recommendation object information of the target recommendation object are input into the propagation behavior tower of the multi-tower model to extract the fusion behavioral features of the second propagation behavior; Input the user information of the initiating user into the first user tower of the multi-tower model, and extract the user features of the initiating user; The user information of the reached users is input into the second user tower of the multi-tower model, and the user features of the reached users are extracted. The fusion behavior features of the second propagation behavior, the user features of the initiating user, and the user features of the reaching user are input into the prediction layer of the multi-tower model for feature fusion to calculate the first propagation generation probability.

9. The method according to claim 4, characterized in that, Before adding the first propagation probability to the propagation queue when the first propagation probability is greater than the propagation probability threshold, the method further includes: Based on the feature similarity between the user characteristics of the initiating user and the user characteristics of the reaching user, the propagation association strength between the initiating user and the reaching user is calculated. If the first propagation probability is greater than the propagation probability threshold, the first propagation probability is added to the propagation queue, including: If the propagation association strength is greater than the strength threshold and the first propagation probability is greater than the propagation probability threshold, the first propagation probability is added to the propagation queue.

10. The method according to claim 4, characterized in that, Before selecting the user to reach from the user set and selecting the first content from the content set, the method further includes: Based on the feature similarity between user features in the user set, user nodes in the propagation network are clustered to obtain an updated propagation network, wherein the updated propagation network includes the updated user set.

11. A method for planning the propagation of recommendations, characterized in that, include: Obtain the network value gain of the propagation network of the target recommendation object, wherein the network value gain is the difference between the target network value obtained after iteratively updating the user value based on each propagation probability in the propagation queue and the initial network value, and the propagation queue includes multiple propagation probabilities; Based on the network value gain, a target user reach sequence for the target recommendation object is generated, wherein the target user reach sequence includes a sequence of user nodes arranged according to the network value gain.

12. A method for propagating recommendations, characterized in that, include: Obtain the target user reach sequence of the target recommendation object, wherein the target user reach sequence is a sequence of user nodes reached by the target recommendation object generated based on the network value gain of the propagation network, the network value gain is the difference between the target network value obtained after iteratively updating the user value by each propagation probability in the propagation queue and the initial network value, and the propagation queue includes multiple propagation probabilities; Based on the target user reach sequence, the content of the target recommendation object is propagated sequentially to reach users.

13. A recommendation platform, characterized in that, Includes offline modules and online modules; The offline module is used for: Obtain the initial network value of the propagation network of the target recommendation object, and the propagation queue, wherein the propagation network includes user nodes corresponding to multiple users, the initial network value includes the initial user value of multiple users, and the propagation queue includes multiple propagation probabilities; A first propagation probability is selected from the propagation queue, wherein the first propagation probability represents the probability that the initiating user influences the reaching user to initiate a second propagation behavior through the first propagation behavior; Based on the first propagation probability, determine the propagation value gain of the initiating user; Based on the initial user value of the initiating user and the propagation value gain, the target user value of the initiating user is determined, and the step of selecting the first propagation probability from the propagation queue is returned to be executed until the propagation queue is empty. Based on the target user values ​​of the multiple users, the target network value of the propagation network is determined. Based on the target network value and the initial network value, determine the network value gain of the propagation network; Based on the network value gain, a target user reach sequence for the target recommendation object is generated; The online module is used for: Based on the target user reach sequence, the content of the target recommendation object is propagated sequentially to reach users.

14. A computing device, characterized in that, include: memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, It stores a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 12.

16. A computer program product, characterized in that, Includes a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 12.

Images