Data processing method and device based on knowledge graph

Abstract

The application provides a data processing method and device based on a knowledge graph. The method can determine an average interest score of interest scores corresponding to identity information of each potential customer, determine whether the average interest score is lower than a second set score, execute step 14 if the average interest score is lower than the second set score, and execute step 15 if the average interest score is not lower than the second set score. Step 14: updating a selection mode of a second user operation feature and a cross mode with the selected second user operation feature based on the average interest score by using a reinforcement learning model, and returning to execute a cross operation on each first user operation feature and a matched second user operation feature to obtain an extended user operation feature set. Step 15: storing a final obtained knowledge graph set. Based on the above, the number of the final obtained knowledge graph set is large, the quality of the extended user operation feature in the knowledge graph set is high, and the time cost is less than one month.

Description

Technical Field

[0001] This application relates to the field of knowledge graph technology, and in particular to a data processing method and apparatus based on knowledge graphs. Background Technology

[0002] With the rapid development of mobile internet and e-commerce, merchants have adopted a mainstream sales model of marketing their products by sending purchase links and product descriptions of target products to users' devices.

[0003] Currently, user operation characteristics related to the target product can be collected from users who have purchased the target product, establishing a set of user operation characteristics of purchased users. If any user has user operation characteristics belonging to the set of user operation characteristics of purchased users, and that user has not purchased the target product, then a purchase link and product introduction information of the target product to be promoted will be sent to that user's user terminal.

[0004] However, collecting a sufficient number of unique user behavior profiles of users who have purchased the target product requires a very long time (e.g., a year or even longer). This wastes a significant amount of marketing time, leading to sluggish sales of the target product, or even causing it to become obsolete (like clothing going out of season or outdated, or certain financial products expiring and being removed from shelves). Summary of the Invention

[0005] This application provides a knowledge graph-based data processing method and apparatus to solve the problem that the existing method of marketing products to users based on the user operation feature set that has been purchased will waste a lot of marketing time, resulting in the unsaleability of the target product, or even the obsolescence of the target product.

[0006] Firstly, this application provides a knowledge graph-based data processing method applied to a business server. The method provided by this application includes: Step 1: Retrieve the set of target user identity information that has purchased the target product within a first preset time period from the preset database, wherein the first preset time period is less than 1 month; Step 2: Extract the user operation features associated with the target product for each target user identity information in the target user identity information set to obtain a user operation feature set. Each user operation feature includes at least four dimensions of operation elements, which include at least user profile information, operation event type, operation time window, and operation number. Step 3: Divide the user operation feature set into a first user operation feature subset and a second user operation feature subset; Step 4: For each first user operation feature in the first user operation feature subset, select a second user operation feature that matches the first user operation feature from the second user operation feature subset; Step 5: Perform a cross operation on each first user operation feature and its matching second user operation feature to obtain an expanded set of user operation features; Step 6: Retrieve the identity information of potential customers who have performed user actions corresponding to any extended user action feature in the extended user action feature set that has previously performed user actions on the target product from the preset database; Step 7: Send the product introduction and purchase link to the target product to the user terminals associated with the identity information of each potential customer; Step 8: Wait for the second preset time period, and obtain the type of operation performed by the potential customer's identity information on the purchase link of the target product sent by each user terminal; Step 9: Based on the operation type corresponding to each potential customer's identity information, retrieve the interest score for the target product from the preset mapping table; Step 10: Extract the extended user operation features corresponding to the interest scores that are higher than the set first score, and add them to the knowledge graph set; Step 11: Determine whether the number of extended user operation features corresponding to interest scores higher than the set first score in the knowledge graph set is less than the set number threshold. If yes, proceed to step 12; otherwise, proceed to step 15. Step 12: Determine the average interest score corresponding to the identity information of each potential customer; Step 13: Determine whether the average interest score is lower than the set second score. If yes, proceed to step 14; otherwise, proceed to step 15. Step 14: Using the reinforcement learning model based on the average interest score, update the selection method of the second user operation feature and the interaction method with the selected second user operation feature, and return to execute step 3, where the first score is greater than the second score; Step 15: Store the final knowledge graph set; Step 16: Retrieve the latest batch of candidate user identity information from the database at preset intervals; Step 17: If the user operation characteristics of any candidate user identity information in the candidate user identity information set that are associated with the target product are included in the final knowledge graph set, then send the introduction and purchase link of the target product to the user terminal associated with the candidate user identity information.

[0007] In some implementations, a crossover operation is performed on each first user operation feature and its matching second user operation feature to obtain an extended set of user operation features, including: For each first user operation feature and its corresponding second user operation feature, a weighted average is calculated for the operation time window or the number of operations. The weighted average is then multiplied by a preset first coefficient, the sums are then multiplied by a preset second coefficient, or the sums are then multiplied by a preset third coefficient to obtain an extended set of user operation features.

[0008] In some implementations, a crossover operation is performed on each first user operation feature and its matching second user operation feature to obtain an extended set of user operation features, including: Swap the operation elements with the same dimension in each first user operation feature and its corresponding second user operation feature.

[0009] In some implementations, the selection method of the second user operation and the crossover method of the crossover operation are the state of the reinforcement learning model, the update of the selection method of the second user operation and the crossover method of the crossover operation are the actions of the reinforcement learning model, and the average interest score is the reward of the reinforcement learning model.

[0010] In some implementations, the latest set of candidate user identity information is retrieved from the database at preset intervals, including: Every preset period, a set of users who have viewed product promotion web pages or live streams related to the target product in the previous period is obtained as the latest batch of candidate user identity information.

[0011] Secondly, this application also provides a knowledge graph-based data processing device applied to a business server. The device provided by this application includes: The identity information acquisition unit is used to retrieve a set of target user identity information that has purchased the target product within a first preset time period from a preset database, where the first preset time period is less than one month. The feature extraction unit is used to extract the user operation features associated with the target product for each target user identity information in the target user identity information set, to obtain a user operation feature set. Each user operation feature includes at least four dimensions of operation elements, and the at least four dimensions of operation elements include at least user profile information, operation event type, operation time window, and operation number. A set partitioning unit is used to divide the user operation feature set into a first user operation feature subset and a second user operation feature subset; The feature selection unit is used to select, for each first user operation feature in the first user operation feature subset, a second user operation feature that matches the first user operation feature from the second user operation feature subset. The feature cross-operation unit is used to perform cross-operation on each first user operation feature and the matching second user operation feature to obtain an extended user operation feature set. The identity information acquisition unit is also used to obtain the identity information of potential customers who have performed user operations corresponding to any extended user operation feature in the extended user operation feature set that has previously performed extended user operation operations on the target product from a preset database. The information sending unit is used to send the introduction of the target product and the purchase link of the target product to the user terminals associated with the identity information of each potential customer. The operation type acquisition unit is used to wait for a second preset time and acquire the operation type of the potential customer's identity information sent by each user terminal on the purchase link of the target product. The score lookup unit is used to look up the interest score of the target product from a preset mapping table based on the operation type corresponding to the identity information of each potential customer. The feature extraction unit is also used to extract extended user operation features corresponding to interest scores higher than the set first score, and add them to the knowledge graph set; The data judgment unit is used to determine whether the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set is less than a set number threshold. The average score determination unit is used to determine the average interest score corresponding to the identity information of each potential customer if the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set is less than a set number threshold. The data judgment unit is also used to determine whether the average interest score is lower than the set second score; The data update unit updates the selection method of the second user operation feature and the interaction method with the selected second user operation feature based on the average interest score using a reinforcement learning model, provided that the first score is greater than the second score. The knowledge graph storage unit is used to store the final knowledge graph set if the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set reaches a set threshold, or the average interest score reaches a set second score. The identity information acquisition unit is also used to retrieve the latest batch of candidate user identity information from the database every preset period. The information sending unit is also used to send the introduction and purchase link of the target product to the user terminal associated with the candidate user identity information if the user operation characteristics of any candidate user identity information in the candidate user identity information set that are associated with the target product are included in the final knowledge graph set.

[0012] In some implementations, the feature crossover unit is specifically used to perform a weighted average of the operation time window or operation number in each first user operation feature and its corresponding second user operation feature, multiply the weighted average by a preset first coefficient, add them and multiply by a preset second coefficient, or subtract them and multiply by a preset third coefficient to obtain an extended user operation feature set.

[0013] In some implementations, the feature crossing unit is specifically used to exchange the operation elements with the same dimension in each first user operation feature and its corresponding second user operation feature.

[0014] In some implementations, the selection method of the second user operation and the crossover method of the crossover operation are the state of the reinforcement learning model, the update of the selection method of the second user operation and the crossover method of the crossover operation are the actions of the reinforcement learning model, and the average interest score is the reward of the reinforcement learning model.

[0015] In some implementations, the identity information acquisition unit is also specifically used to acquire, at preset intervals, a set of users who have browsed product promotion web pages or live broadcast rooms associated with the target product in the previous period, as the latest batch of candidate user identity information sets.

[0016] This application provides a data processing method and apparatus based on a knowledge graph, comprising: Step 3: dividing a set of user operation features into a first subset of user operation features and a second subset of user operation features; Step 4: for each first user operation feature in the first subset of user operation features, selecting a second user operation feature from the second subset of user operation features that matches the first user operation feature. By selecting a matching second user operation feature, it is more beneficial to obtain two extended user operation features with high interest scores, which can reduce the subsequent iteration process and save time in obtaining the final knowledge graph set.

[0017] Step 5: Perform a cross operation on each first user action feature and its matching second user action feature to obtain an extended user action feature set. Understandably, the extended user action feature set can be viewed as a set of "newly created" user action features that did not exist before, i.e., a set of user action features obtained through knowledge reasoning.

[0018] Step 6: Retrieve the identity information of potential customers who have performed user actions corresponding to any extended user action feature in the extended user action feature set that has previously performed user actions on the target product from the preset database; Step 7: Send the target product introduction and purchase link to the user terminals associated with each potential customer's identity information. Step 8: Wait for a second preset time and obtain the action type performed by each potential customer on the purchase link. Step 9: Based on the action type corresponding to each potential customer's identity information, retrieve the interest score for the target product from a preset mapping table. The interest score represents the likelihood of a potential customer purchasing the target product; the higher the interest score, the higher the conversion rate for promoting the target product.

[0019] Step 10: Extract extended user action features corresponding to interest scores higher than the set first score and add them to the knowledge graph set. This ensures that the quality of extended user action features in the knowledge graph set is high (i.e., the corresponding users are more likely to purchase the target product). Step 11: Determine if the number of extended user action features corresponding to interest scores higher than the set first score in the knowledge graph set is less than the set threshold. If so, the knowledge graph set needs to be further expanded to cover more potential customers who will purchase the target product; therefore, proceed to Step 12. If not, the knowledge graph set is already large enough to cover more potential customers who will purchase the target product; therefore, proceed to Step 15.

[0020] Step 12: Determine the average interest score corresponding to the identity information of each potential customer; Step 13: Determine whether the average interest score is lower than the set second score. If so, it means that the overall quality of the extended user operation feature set obtained this time is not high, and the number of extended user operation features corresponding to interest scores higher than the set first score is small. Therefore, proceed to step 14. If not, it means that it is no longer possible to expand the size of the knowledge graph set to the set threshold. However, the size of the knowledge graph set obtained at this time is much larger than the initially obtained user operation feature set. Therefore, proceed to step 15.

[0021] Step 14: Using a reinforcement learning model based on the average interest score, update the selection method for the second user action feature and the interaction method with the selected second user action feature, and return to execute Step 4, with the first score greater than the second score. This ensures that the proportion of extended user action features with interest scores higher than the set first score in each iteration is higher than the previous iteration, thus further reducing the number of iterations and allowing the knowledge graph set to reach the set threshold more quickly.

[0022] Step 15: Store the final knowledge graph set. Based on the above, the final knowledge graph set is much larger than the initially obtained user operation feature set, and the quality of the extended user operation features in the knowledge graph set is high (i.e., the corresponding users are more likely to purchase the target product), and the time spent is within the first preset time period and less than one month.

[0023] This allows for the retrieval of the latest batch of candidate user identity information from the database at preset intervals within a month. If any user operation feature of any candidate user identity information in the candidate user identity information set is included in the final knowledge graph set, then an introduction to the target product and a purchase link are sent to the user terminal associated with the candidate user identity information. Because the extended user operation features in the knowledge graph set are of high quality, the conversion rate for promoting the target product is also higher. Furthermore, since the size of the knowledge graph set is much larger than the initially obtained set of user operation features, it covers a large number of potential customers who wish to purchase the target product. Moreover, the time frame for obtaining the knowledge graph set saves a significant amount of time, preventing the target product from becoming unsaleable or outdated, thus protecting the interests of the merchant. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart illustrating a knowledge graph-based data processing method provided in an embodiment of this application; Figure 2 A functional block diagram of a knowledge graph-based data processing device provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments made by those skilled in the art under the guidance of these embodiments are within the scope of protection of this application.

[0027] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] This application provides a knowledge graph-based data processing method applied to a business server. The method provided in this application includes: S101: Retrieve from the preset database a set of target user identity information that has purchased the target product within a first preset time period, wherein the first preset time period is less than 1 month.

[0029] For example, the target product may be, but is not limited to, insurance, financial products, clothing, or electronic products. The first preset duration may be 15 days, 20 days, or 25 days, etc.

[0030] S102: Extract the user operation features associated with the target product for each target user identity information in the target user identity information set to obtain a user operation feature set. Each user operation feature includes at least four operational elements, and these elements must include at least user profile information, operation event type, operation time window, and number of operations.

[0031] It should be noted that the number of user operation features in the user operation feature set may be 500, 800, or 1000, etc., and is not limited here.

[0032] For example, one user action characteristic could be that a user of user profile information A (i.e., user profile information) liked (i.e., action event type) 5 times (i.e., number of actions) the promotional live stream of the target product within 2 days (i.e., action time window). User profile information A can include the user's age group, physical information, income range, occupation, etc.

[0033] For example, another user action characteristic could be that a user of user profile information B (i.e., user profile information) commented on the promotional webpage of the target product 3 times within 5 days (i.e., action event type) (i.e., number of actions). User profile information B can include the user's age group, physical information, income range, occupation, etc.

[0034] S103: Divide the user operation feature set into a first user operation feature subset and a second user operation feature subset.

[0035] For example, the number of user operation features in the first and second user operation feature subsets after partitioning can be equal or unequal.

[0036] S104: For each first user operation feature in the first user operation feature subset, select a second user operation feature that matches the first user operation feature from the second user operation feature subset.

[0037] S105: Perform a cross operation on each first user operation feature and the matching second user operation feature to obtain an extended user operation feature set.

[0038] Specifically, the implementation of S105 includes, but is not limited to, the following two methods: The first method involves taking a weighted average of the operation time window or number of operations for each first user operation feature and its corresponding second user operation feature, multiplying the weighted average by a preset first coefficient (such as 0.5, 2, or 3), adding them together and multiplying by a preset second coefficient (such as 0.5, 1.5, or 2), or subtracting them and multiplying by a preset third coefficient (such as 1.5, 2, or 3), to obtain an extended set of user operation features.

[0039] For example, if the first user action feature is user profile A, and the user likes the target product's promotional live stream 4 times within 2 days; and the matching first user action feature is user profile B, and the user likes the target product's promotional live stream 2 times within 6 days, with the weighting coefficient of the first user action feature being 0.5 and the weighting coefficient of the second user action feature being 0.5, the resulting extended user action features can include two: one is user profile A, and the other is user profile B, who liked the target product's promotional live stream 3 times within 4 days.

[0040] The second method involves swapping the operation elements with the same dimension in each first user operation feature and its corresponding second user operation feature.

[0041] For example, a user with the first user action characteristic, user profile information A, liked the target product's promotional live stream 4 times within 2 days; a user with the matching first user action characteristic, user profile information B, commented on the target product's promotional webpage 2 times within 6 days. If the action time window dimension and the action event type dimension are swapped, the resulting extended user action characteristics can include two: one is user profile information A liking the target product's promotional webpage 4 times within 6 days; the other is user profile information B commenting on the target product's promotional live stream 2 times within 2 days.

[0042] S106: Retrieve the identity information of potential customers who have performed user actions corresponding to any extended user action feature in the set of extended user action features that have previously performed extended user action features on the target product from the preset database.

[0043] For example, if user profile A's extended user action characteristics show that user A liked the promotional webpage for the target product 4 times within 6 days, then the identity information of potential customers who liked the promotional webpage for the target product 4 times within 6 days can be obtained.

[0044] S107: Send the introduction of the target product and the purchase link of the target product to the user terminal associated with the identity information of each potential customer.

[0045] S108: Wait for the second preset time, obtain the type of operation performed by the potential customer's identity information sent by each user terminal on the purchase link of the target product.

[0046] For example, the second preset duration can be 1 day, 2 days or 3 days.

[0047] S109: Based on the operation type corresponding to the identity information of each potential customer, retrieve the interest score for the target product from the preset mapping relationship table.

[0048] For example, the interest score is 5 points for browsing the webpage corresponding to the purchase link; 6 points for forwarding the webpage corresponding to the purchase link; 7 points for commenting on the webpage corresponding to the purchase link; 8 points for adding the target product to the shopping cart on the webpage corresponding to the purchase link; 10 points for purchasing the target product on the webpage corresponding to the purchase link; and less than 5 points for not taking any action on the purchase link.

[0049] S110: Extract the extended user operation features corresponding to the interest scores that are higher than the set first score, and add them to the knowledge graph set.

[0050] For example, the first score could be 8 points.

[0051] S111: Determine whether the number of extended user operation features corresponding to interest scores higher than the set first score in the knowledge graph set is less than the set number threshold. If yes, execute S112; otherwise, execute S115.

[0052] S112: Determine the average interest score corresponding to the identity information of each potential customer.

[0053] S113: Determine whether the average interest score is lower than the set second score. If yes, execute S114; otherwise, execute S115.

[0054] For example, the first score is greater than the second score, which can be 6 points.

[0055] S114: Using a reinforcement learning model based on the average interest score, update the selection method for the second user operation feature and the interaction method with the selected second user operation feature, and return to execute S104.

[0056] It should be noted that the selection method for the second user operation and the cross operation method are the states of the reinforcement learning model, the update of the selection method for the second user operation and the cross operation method are the actions of the reinforcement learning model, and the average interest score is the reward of the reinforcement learning model.

[0057] It should be noted that the reinforcement learning model described above can be a deep learning-based Q-learning network, i.e., a Deep Q-network (DQN) network. Unlike Q-learning, the DQN network does not require building a complete Q-matrix; instead, it uses a neural network to estimate the value of the Q-function. The neural network responsible for estimating the Q-function is called the main network. This represents the parameter set of the main network, while the target network outputs target values, which are used to update the parameters of the main network. The estimated values ​​of the main network and the target values ​​of the target network can form a loss function. The Adam optimization algorithm can be used to perform stochastic gradient descent to update the parameters of the main network. The parameters of the main network can be continuously updated based on the gradient of the loss function, allowing the loss function value to continuously decrease, thus making the estimated values ​​of the main network more accurate. Both the main network and the target network contain input layers, convolutional layers, fully connected layers, and output layers. The neurons in the input layer are responsible for feeding data to the neurons in the convolutional layers, which are responsible for extracting local features from the input data. The fully connected layers integrate these local features into global features, and the output layer is responsible for outputting the estimated value of the Q-function corresponding to each action in the current state.

[0058] S115: Store the final knowledge graph set.

[0059] S116: Retrieve the latest batch of candidate user identity information from the database at preset intervals.

[0060] Specifically, at preset intervals, a set of users who have viewed product promotion web pages or live streams related to the target product in the previous period can be obtained as the latest batch of candidate user identity information. For example, the preset interval can be 1 day, 3 days, or 7 days, etc., and is not limited here.

[0061] S117: If the user operation characteristics of any candidate user identity information in the candidate user identity information set that are associated with the target product are included in the final knowledge graph set, then the introduction and purchase link of the target product will be sent to the user terminal associated with the candidate user identity information.

[0062] In summary, the knowledge graph-based data processing method provided in this application includes: S103: dividing the user operation feature set into a first user operation feature subset and a second user operation feature subset; S104: for each first user operation feature in the first user operation feature subset, selecting a second user operation feature that matches the first user operation feature from the second user operation feature subset. By selecting matching second user operation features, it is more beneficial to obtain two extended user operation features with high interest scores, which can reduce the subsequent iteration process and save time in obtaining the final knowledge graph set.

[0063] S105: Perform a cross operation on each first user operation feature and its matching second user operation feature to obtain an extended user operation feature set. Understandably, the extended user operation feature set can be viewed as a set of "newly created" user operation features that did not exist before, i.e., a set of user operation features obtained through knowledge reasoning.

[0064] S106: Retrieve the identity information of potential customers who have performed user operations corresponding to any extended user operation feature in the extended user operation feature set that has previously performed extended user operation on the target product from the preset database; S107: Send an introduction to the target product and a purchase link to the target product to the user terminals associated with the identity information of each potential customer. S108: Wait for a second preset time and obtain the action type performed by each potential customer on the purchase link of the target product using their identity information. S109: Based on the action type corresponding to each potential customer's identity information, retrieve the interest score for the target product from a preset mapping table. The interest score represents the likelihood of a potential customer purchasing the target product; the higher the interest score, the higher the conversion rate for promoting the target product.

[0065] S110: Extract extended user action features corresponding to interest scores higher than the set first score and add them to the knowledge graph set. This ensures that the quality of extended user action features in the knowledge graph set is high (i.e., the corresponding users are more likely to purchase the target product). S111: Determine if the number of extended user action features corresponding to interest scores higher than the set first score in the knowledge graph set is less than a set threshold. If so, the size of the knowledge graph set needs to be further expanded to cover more potential customers who will purchase the target product, thus proceeding to S112; otherwise, the size of the knowledge graph set is already large enough to cover more potential customers who will purchase the target product, thus proceeding to S115.

[0066] S112: Determine the average interest score corresponding to the identity information of each potential customer; S113: Determine whether the average interest score is lower than the set second score. If so, it means that the overall quality of the extended user operation feature set obtained this time is not high, and the number of extended user operation features corresponding to interest scores higher than the set first score is small. Therefore, execute S114. If not, it means that it is no longer possible to expand the size of the knowledge graph set to the set threshold. However, the size of the knowledge graph set obtained at this time is much larger than the initially obtained user operation feature set. Therefore, execute S115.

[0067] S114: Using a reinforcement learning model based on the average interest score, update the selection method for the second user operation feature and the interaction method with the selected second user operation feature, and return to execute S103, where the first score is greater than the second score. This ensures that the proportion of extended user operation features with interest scores higher than the set first score in each iteration is higher than the previous iteration, thus further reducing the number of iterations and allowing the knowledge graph set to reach the set threshold more quickly.

[0068] S115: Store the final knowledge graph set. Based on the above, the final knowledge graph set is much larger than the initially obtained user operation feature set, and the quality of the extended user operation features in the knowledge graph set is high (i.e., the corresponding users are more likely to purchase the target product), and the time spent is within the first preset time period and less than one month.

[0069] This allows for the retrieval of the latest batch of candidate user identity information from the database at preset intervals within a month. If any user operation feature of any candidate user identity information in the candidate user identity information set is included in the final knowledge graph set, then an introduction to the target product and a purchase link are sent to the user terminal associated with the candidate user identity information. Because the extended user operation features in the knowledge graph set are of high quality, the conversion rate for promoting the target product is also higher. Furthermore, since the size of the knowledge graph set is much larger than the initially obtained set of user operation features, it covers a large number of potential customers who wish to purchase the target product. Moreover, the time frame for obtaining the knowledge graph set saves a significant amount of time, preventing the target product from becoming unsaleable or outdated, thus protecting the interests of the merchant.

[0070] Additionally, please see Figure 2 This application also provides a knowledge graph-based data processing device applied to a business server. It should be noted that the basic principles and technical effects of the knowledge graph-based data processing provided in this application are the same as those in the above embodiments. For the sake of brevity, any parts not mentioned in this application can be referred to the corresponding content in the above embodiments. The device provided in this application includes an identity information acquisition unit, a feature extraction unit, a set partitioning unit, a feature selection unit, a feature cross-referencing unit, an information sending unit, an operation type acquisition unit, a score lookup unit, a data judgment unit, an average score determination unit, a data update unit, and a knowledge graph storage unit. The identity information acquisition unit is used to retrieve a set of target user identity information that has purchased the target product within a first preset time period from a preset database, where the first preset time period is less than one month. The feature extraction unit is used to extract the user operation features associated with the target product for each target user identity information in the target user identity information set, to obtain a user operation feature set. Each user operation feature includes at least four dimensions of operation elements, and the at least four dimensions of operation elements include at least user profile information, operation event type, operation time window, and operation number. A set partitioning unit is used to divide the user operation feature set into a first user operation feature subset and a second user operation feature subset; The feature selection unit is used to select, for each first user operation feature in the first user operation feature subset, a second user operation feature that matches the first user operation feature from the second user operation feature subset. The feature cross-operation unit is used to perform cross-operation on each first user operation feature and the matching second user operation feature to obtain an extended user operation feature set. The identity information acquisition unit is also used to obtain the identity information of potential customers who have performed user operations corresponding to any extended user operation feature in the extended user operation feature set that has previously performed extended user operation operations on the target product from a preset database. The information sending unit is used to send the introduction of the target product and the purchase link of the target product to the user terminals associated with the identity information of each potential customer. The operation type acquisition unit is used to wait for a second preset time and acquire the operation type of the potential customer's identity information sent by each user terminal on the purchase link of the target product. The score lookup unit is used to look up the interest score of the target product from a preset mapping table based on the operation type corresponding to the identity information of each potential customer. The feature extraction unit is also used to extract extended user operation features corresponding to interest scores higher than the set first score, and add them to the knowledge graph set; The data judgment unit is used to determine whether the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set is less than a set number threshold. The average score determination unit is used to determine the average interest score corresponding to the identity information of each potential customer if the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set is less than a set number threshold. The data judgment unit is also used to determine whether the average interest score is lower than the set second score; The data update unit updates the selection method of the second user operation feature and the interaction method with the selected second user operation feature based on the average interest score using a reinforcement learning model, provided that the first score is greater than the second score. The knowledge graph storage unit is used to store the final knowledge graph set if the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set reaches a set threshold, or the average interest score reaches a set second score. The identity information acquisition unit is also used to retrieve the latest batch of candidate user identity information from the database every preset period. The information sending unit is also used to send the introduction and purchase link of the target product to the user terminal associated with the candidate user identity information if the user operation characteristics of any candidate user identity information in the candidate user identity information set that are associated with the target product are included in the final knowledge graph set.

[0071] In some implementations, the feature crossover unit is specifically used to perform a weighted average of the operation time window or operation number in each first user operation feature and its corresponding second user operation feature, multiply the weighted average by a preset first coefficient, add them and multiply by a preset second coefficient, or subtract them and multiply by a preset third coefficient to obtain an extended user operation feature set.

[0072] In some implementations, the feature crossing unit is specifically used to exchange the operation elements with the same dimension in each first user operation feature and its corresponding second user operation feature.

[0073] In some implementations, the selection method of the second user operation and the crossover method of the crossover operation are the state of the reinforcement learning model, the update of the selection method of the second user operation and the crossover method of the crossover operation are the actions of the reinforcement learning model, and the average interest score is the reward of the reinforcement learning model.

[0074] In some implementations, the identity information acquisition unit is also specifically used to acquire, at preset intervals, a set of users who have browsed product promotion web pages or live broadcast rooms associated with the target product in the previous period, as the latest batch of candidate user identity information set.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A data processing method based on knowledge graphs, characterized in that, Applied to a business server, the method includes: Step 1: Retrieve a set of target user identity information that has purchased the target product within a first preset time period from a preset database, wherein the first preset time period is less than 1 month; Step 2: Extract the user operation features associated with the target product for each target user identity information in the target user identity information set to obtain a user operation feature set. Each user operation feature includes at least four dimensions of operation elements, and the at least four dimensions of operation elements include at least user profile information, operation event type, operation time window, and operation number. Step 3: Divide the user operation feature set into a first user operation feature subset and a second user operation feature subset; Step 4: For each first user operation feature in the first user operation feature subset, select a second user operation feature that matches the first user operation feature from the second user operation feature subset; Step 5: Perform a cross operation on each of the first user operation features and the matching second user operation features to obtain an expanded set of user operation features; Step 6: Retrieve the identity information of potential customers who have performed user operations corresponding to any extended user operation feature in the extended user operation feature set that have previously performed such operations on the target product from the preset database; Step 7: Send an introduction to the target product and a purchase link for the target product to the user terminals associated with the identity information of each potential customer; Step 8: Wait for the second preset time period, and obtain the operation type of the potential customer's identity information sent by each user terminal on the purchase link of the target product; Step 9: Based on the operation type corresponding to the identity information of each potential customer, find the interest score for the target product from the preset mapping table; Step 10: Extract the extended user operation features corresponding to the interest scores that are higher than the set first score, and add them to the knowledge graph set; Step 11: Determine whether the number of extended user operation features corresponding to interest scores higher than the set first score in the knowledge graph set is less than the set number threshold. If yes, proceed to step 12; otherwise, proceed to step 15. Step 12: Determine the average interest score corresponding to the identity information of each potential customer; Step 13: Determine whether the average interest score is lower than the set second score. If yes, proceed to step 14; otherwise, proceed to step 15. Step 14: Using the reinforcement learning model based on the average interest score, update the selection method of the second user operation feature and the interaction method with the selected second user operation feature, and return to execute step 4, where the first score is greater than the second score; Step 15: Store the final knowledge graph set; Step 16: Retrieve the latest batch of candidate user identity information from the database at preset intervals; Step 17: If the user operation features associated with the target product for any candidate user identity information in the candidate user identity information set are included in the final knowledge graph set, then send the introduction and purchase link of the target product to the user terminal associated with the candidate user identity information.

2. The method according to claim 1, characterized in that, The step of performing a cross operation on each of the first user operation features and the matching second user operation features to obtain an extended user operation feature set includes: For each first user operation feature and its corresponding second user operation feature, a weighted average is calculated for the operation time window or the number of operations. The weighted average is then multiplied by a preset first coefficient, the sums are then multiplied by a preset second coefficient, or the subtractions are then multiplied by a preset third coefficient to obtain an extended set of user operation features.

3. The method according to claim 1, characterized in that, The step of performing a cross operation on each of the first user operation features and the matching second user operation features to obtain an extended user operation feature set includes: Swap the operation elements with the same dimension in each of the first user operation features and its corresponding second user operation features.

4. The method according to claim 1, characterized in that, The selection method for the second user operation and the crossover method of the crossover operation are the state of the reinforcement learning model. Updating the selection method for the second user operation and the crossover method of the crossover operation are the actions of the reinforcement learning model. The average interest score is the reward of the reinforcement learning model.

5. The method according to claim 1, characterized in that, The step of retrieving the latest batch of candidate user identity information from the database at preset intervals includes: Every preset period, a set of users who have viewed product promotion web pages or live streams associated with the target product in the previous period is obtained as the latest batch of candidate user identity information set.

6. A data processing device based on knowledge graphs, characterized in that, The device, applied to a business server, includes: The identity information acquisition unit is used to acquire a set of target user identity information that has purchased the target product within a first preset time period from a preset database, and the first preset time period is less than one month. The feature extraction unit is used to extract user operation features associated with the target product for each target user identity information in the target user identity information set to obtain a user operation feature set. Each user operation feature includes at least four dimensions of operation elements, and the at least four dimensions of operation elements include at least user profile information, operation event type, operation time window, and operation number. A set partitioning unit is used to divide the user operation feature set into a first user operation feature subset and a second user operation feature subset; A feature selection unit is configured to, for each first user operation feature in the first user operation feature subset, select a second user operation feature from the second user operation feature subset that matches the first user operation feature; The feature crossover unit is used to perform crossover operations on each of the first user operation feature and the matched second user operation feature to obtain an extended user operation feature set. The identity information acquisition unit is also used to acquire the identity information of potential customers who have performed user operations corresponding to any extended user operation feature in the extended user operation feature set on the target product from a preset database. An information sending unit is used to send an introduction to the target product and a purchase link for the target product to the user terminals associated with the identity information of each potential customer. An operation type acquisition unit is used to wait for a second preset time and acquire the operation type of the potential customer's identity information sent by each user terminal on the purchase link of the target product. The score lookup unit is used to look up the interest score for the target product from a preset mapping table based on the operation type corresponding to the identity information of each potential customer. The feature extraction unit is also used to extract extended user operation features corresponding to interest scores higher than a set first score, and add them to the knowledge graph set; The data judgment unit is used to determine whether the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set is less than a set number threshold. The average score determination unit is used to determine the average interest score corresponding to the identity information of each potential customer if the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set is less than a set number threshold. The data judgment unit is also used to determine whether the average interest score is lower than a set second score; The data update unit, if the average interest score is lower than a set second score, uses a reinforcement learning model to update the selection method of the second user operation feature and the interaction method with the selected second user operation feature based on the average interest score, and the first score is greater than the second score. The knowledge graph storage unit is used to store the final knowledge graph set if the number of extended user operation features corresponding to interest scores higher than a set first score in the knowledge graph set reaches a set threshold, or the average interest score reaches a set second score. The identity information acquisition unit is also used to acquire the latest batch of candidate user identity information from the database every preset period. The information sending unit is further configured to send an introduction and purchase link of the target product to the user terminal associated with the candidate user identity information if the user operation features associated with the target product of any candidate user identity information in the candidate user identity information set are included in the final knowledge graph set.

7. The apparatus according to claim 6, characterized in that, The feature cross-interaction unit is specifically used to perform a weighted average of the operation time window or operation number in each of the first user operation features and its corresponding second user operation features, multiply the weighted average by a preset first coefficient, add them and multiply by a preset second coefficient, or subtract them and multiply by a preset third coefficient to obtain an extended user operation feature set.

8. The apparatus according to claim 6, characterized in that, The feature crossing unit is specifically used to exchange the operation elements with the same dimension in each of the first user operation features and its corresponding second user operation features.

9. The apparatus according to claim 6, characterized in that, The selection method for the second user operation and the crossover method of the crossover operation are the state of the reinforcement learning model. Updating the selection method for the second user operation and the crossover method of the crossover operation are the actions of the reinforcement learning model. The average interest score is the reward of the reinforcement learning model.

10. The apparatus according to claim 6, characterized in that, The identity information acquisition unit is also specifically used to acquire, at preset intervals, a set of users who have browsed product promotion web pages or live broadcast rooms associated with the target product in the previous period, as the latest batch of candidate user identity information sets.

Images