Method and apparatus for determining target item, electronic device, and storage medium

Abstract

This invention discloses a method, apparatus, electronic device, and storage medium for determining target items. The method includes: determining basic behavior sub-data and tracking sub-data corresponding to each triggered item based on acquired basic behavior data and tracking data associated with a target user; obtaining attribute feature values ​​of each triggered item relative to the target user based on the basic behavior sub-data, tracking data, and a pre-trained item attribute value determination model; and determining the target item associated with the target user based on the attribute feature values. This technical solution solves the problem of treating all user trigger behaviors as equal, which leads to the inability to determine the most suitable item for the user, resulting in a low item conversion rate. It distinguishes the impact of different user trigger behaviors on item preference, thereby pushing corresponding products to the user and improving the item conversion rate.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, apparatus, electronic device and storage medium for determining a target item. Background Technology

[0002] With the development of computer technology, e-commerce has increasingly permeated users' daily lives, impacting every aspect of their lives. Currently, e-commerce platforms recommend relevant items to users primarily based on user behavior data such as clicks, browsing, and purchases of various items. This data is used to determine a user profile for each user, and then the platform recommends items based on this profile. In other words, user actions triggered on the item details page are the main way to determine user profiles.

[0003] When implementing this technical solution based on the above method, the inventors discovered the following problems:

[0004] In practical applications, there may be invalid behaviors such as accidental touches and negative clicks, such as removing items from the follow list or add-to-cart list. Data on invalid behaviors and negative clicks can impact user profiles and item recommendation features. However, currently, these triggering behaviors are treated as equivalent, leading to an error rate when using user actions to recommend items, thus creating technical issues that affect item conversion rates. Summary of the Invention

[0005] This invention provides a method, apparatus, electronic device, and storage medium for determining target items, so as to achieve the technical effect of improving item conversion rate.

[0006] In a first aspect, embodiments of the present invention provide a method for determining a target item, the method comprising:

[0007] Based on the acquired basic behavioral data and event tracking data associated with the target user, determine the basic behavioral sub-data and event tracking sub-data corresponding to each triggered item;

[0008] Based on the basic behavioral sub-data, embedded data, and pre-trained item attribute value determination model of each triggered item, the attribute feature values ​​of each triggered item relative to the target user are obtained.

[0009] Based on the attribute feature values, the target items associated with the target user are determined.

[0010] Secondly, embodiments of the present invention also provide an apparatus for determining a target article, the apparatus comprising:

[0011] The data determination module is used to determine the basic behavior sub-data and tracking sub-data corresponding to each triggered item based on the basic behavior data and tracking data associated with the target user.

[0012] The attribute feature value determination module is used to obtain the attribute feature value of each triggered item relative to the target user based on the basic behavior sub-data, the tracking point sub-data and the pre-trained item attribute value determination model of each triggered item.

[0013] The target item determination module is used to determine the target items associated with the target user based on various attribute feature values.

[0014] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0015] One or more processors;

[0016] Storage device for storing one or more programs.

[0017] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining a target item as described in any embodiment of the present invention.

[0018] Fourthly, embodiments of the present invention also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a method for determining a target item as described in any of the embodiments of the present invention.

[0019] The technical solution of this invention, by acquiring basic behavioral data and event tracking data associated with the target user, can determine the basic behavioral sub-data and event tracking sub-data corresponding to each triggered item. After inputting the basic behavioral sub-data and event tracking data into a pre-trained item attribute value determination model, the attribute feature values ​​of the triggered item relative to the target user can be obtained. Based on the attribute feature values ​​of each triggered item, the target item associated with the target user can be determined. When the user triggers the client, the target item can be displayed to the target client for the user to choose from. This solves the problem in the prior art where, when determining the items of interest to the user based on the number of times the user triggers each item and the click rate, the impact of different triggering operations on the items is not considered, resulting in a certain error rate between the items pushed to the user and the items that the user is actually interested in, thus causing a low item conversion rate. This invention achieves the technical effect of distinguishing the impact of different triggering behaviors on the user based on event tracking data, thereby improving the matching degree between the pushed items and the user, and thus improving the user experience. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of exemplary embodiments of the present invention, the accompanying drawings used in describing the embodiments are briefly introduced below. Obviously, the accompanying drawings described are only a portion of the drawings of the embodiments to be described in this invention, and not all of the drawings. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort.

[0021] Figure 1 This is a schematic flowchart of a method for determining a target item provided in Embodiment 1 of the present invention;

[0022] Figure 2 This is a schematic flowchart of a method for determining a target item provided in Embodiment 2 of the present invention;

[0023] Figure 3 This is a schematic diagram of a device for determining a target item provided in Embodiment 3 of the present invention.

[0024] Figure 4 This is a schematic diagram of an electronic device structure provided in Embodiment 4 of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0026] Example 1

[0027] Figure 1 This is a flowchart illustrating a method for determining a target item according to Embodiment 1 of the present invention. This embodiment is applicable to situations where, based on user click behavior and corresponding embedded data, an attribute evaluation value of an item relative to the user is determined, and then the user's interest in an item is determined based on the attribute evaluation value. This method can be executed by a device for determining the target item, which can be implemented in the form of software and / or hardware. The hardware can be an electronic device, such as a mobile terminal or a PC. The execution of this technical solution can be performed by a server, by a terminal device, or by a combination of both.

[0028] Before introducing this technical solution, let's illustrate the application scenario. In the e-commerce field, when a user wants to obtain an item, they can browse the item's details page. Of course, the user can also trigger corresponding actions on the item's details page. In this case, the click action can act as a bridge between client-side event tracking and search features. The user's trigger action is a relatively low-granular determination of the items the user is interested in. In practical applications, the user's trigger action may be a positive action for the item, or a negative feedback action. Positive actions may include adding items to the cart or prolonged browsing time. Negative feedback actions may include removing items from the cart or favorites list. The above behavioral data can be used as basic behavioral data. Based on the above technical solution, corresponding event tracking can be set up to detect various operational behaviors. That is, while determining the click data, it is also possible to distinguish whether the operation is positive or negative, thereby accurately pushing relevant items to the user.

[0029] like Figure 1 As shown, the method includes:

[0030] S110. Based on the basic behavior data and event tracking data associated with the target user, determine the basic behavior sub-data and event tracking sub-data corresponding to each triggered item.

[0031] In this context, if we want to identify an item that a user is interested in, we can use that user as the target user. Alternatively, when a user triggers a client-side activity, relevant items are typically pushed to them, and the user corresponding to that activity can be used as the target user. Basic behavioral data can be data corresponding to the user's trigger actions on various item detail pages. Trigger actions can include the way a user enters the item detail page. For example, there are multiple ways to enter the detail page, which can be constructed using a one-hot encoding. This means that all entry methods are represented by vectors, with different element values ​​indicating whether that method was used. For example, 0 indicates that the method was not used, and 1 indicates that the method was used. Assuming there are 5 ways to enter the item detail page, and the vector is

[01000] , we know that the second method was used to enter the item detail page. Basic behavioral data can also include browsing duration. Browsing duration is a numerical feature and can be used directly after normalization. Basic behavioral data also includes click data, such as the number of clicks. The number of clicks is a numerical feature and can also be used directly after normalization. Tracking data can be data detection tracking points pre-set by the client. The sequence of user clicks on client-side event tracking points can be viewed as a directed graph model, which visualizes the click sequence during a user's access to a specific SKU. In other words, event tracking points can be implemented based on pre-defined data to determine whether the corresponding click behavior is positive or negative.

[0032] It's important to note that users can browse multiple items within a certain timeframe, resulting in multiple sets of basic behavioral data and event tracking data associated with the user. In this case, the basic behavioral data and event tracking data correspond to multiple items. To determine the user's preference for each item, separate basic behavioral data sub-data and event tracking sub-data can be determined for each item. In other words, if a user browses multiple items within a preset timeframe, then multiple sets of basic behavioral data sub-data and event tracking data will be generated.

[0033] It should also be noted that the event tracking features themselves cannot be introduced into the deep neural network; they need to be processed into vector form. This proposal uses a Graph Embedding model for this purpose. The Node2Vec algorithm's Graph Embedding is a representative implementation that effectively represents the graph structure of user access events. After generating the embedded representation of the event tracking, it is fed into a DeepFM (i.e., a model trained to determine item attribute values) network, and processed along with other basic features (i.e., basic behavioral data, such as the number of clicks on the product details page, browsing duration, etc.) to obtain the attribute feature value of the triggered item relative to the user.

[0034] Specifically, each item triggered by a user within a preset time period can be considered a triggered item; that is, any item triggered by the target user within a certain time period can be considered a triggered item. Correspondingly, both the basic behavior sub-data and the event tracking sub-data can be data generated when a user clicks on a triggered item.

[0035] In this embodiment, before the acquisition of basic behavior data and event tracking data, the method further includes: extracting basic behavior data and event tracking data of each user for each triggered item within a preset time period from the system log; and determining the basic behavior data and event tracking data of each user for each triggered item within a preset time period according to pre-set rules, so as to determine the basic behavior sub-data and event tracking sub-data of each triggered item corresponding to each user from the basic behavior data and event tracking behavior data.

[0036] The system logs can be used to record basic behavioral data and event tracking data for each user's interactions with various items. This data is determined based on the generation time of each basic behavioral data point and event tracking data point. To determine a user's preference for each triggered item, the data can be categorized and organized according to a specific field, resulting in basic behavioral data and event tracking data for multiple triggered items associated with each user in the system logs. A pre-set rule could be to organize all data according to the content of a specific field. In this case, the data associated with each user can be used as basic behavioral data and event tracking data, and correspondingly, the data associated with each user for each triggered item can be used as basic behavioral sub-data and event tracking sub-data.

[0037] S120. Based on the basic behavior sub-data, embedded data and pre-trained item attribute value determination model of each triggered item, the attribute feature value of each triggered item relative to the target user is obtained.

[0038] The item attribute value determination model is pre-trained and used to process the basic behavioral sub-data and event tracking sub-data of each triggered item to determine the user's preference for each triggered item. Correspondingly, the attribute feature value is used to characterize the user's degree of preference for the triggered item. Optionally, the larger the attribute feature value, the more interested the user is in the triggered item, and vice versa.

[0039] Specifically, for each triggered item, the basic behavior sub-data and tracking data of the currently triggered item can be used as input parameters for the item attribute value determination model. Based on the item attribute value determination model, the input parameters are processed to obtain the attribute feature value of the currently triggered item.

[0040] In this embodiment, the step of determining the attribute feature value of each triggered item relative to the target user based on the basic behavior sub-data, tracking data, and pre-trained item attribute value determination model of each triggered item includes: determining the tracking vector based on the tracking data of each triggered item; determining the basic behavior vector corresponding to each basic behavior sub-data based on the vector construction layer in the item attribute value determination model, and determining the target behavior data vector of each triggered item based on each basic behavior vector and the corresponding tracking vector; and processing each target behavior data vector based on the item attribute value determination model to obtain the attribute feature value of each triggered item relative to the target user.

[0041] The features of the event tracking data cannot be directly incorporated into the neural network, but they can be processed into vector form. That is, for each triggered item's event tracking sub-data, it can be processed into a corresponding vector, which can then be used as the event tracking vector. In other words, the event tracking sub-data corresponding to each triggered item can be processed into an event tracking vector. The item attribute value determination model includes a vector construction layer, which processes the basic behavior sub-data into vectors. The vector obtained by integrating the vectors of the basic behavior sub-data and the event tracking data is used as the target behavior data vector. In other words, the target behavior data vector is the vector corresponding to both the basic behavior sub-data and the event tracking data.

[0042] Specifically, basic behavior sub-data and event tracking sub-data can be input into the item attribute value determination model. Based on the vector construction layer in this model, the basic behavior sub-data is processed to obtain the corresponding basic behavior vector. By integrating the basic behavior vector and the event tracking data vector, the target behavior data vector is obtained. The item attribute value determination model can then process the target behavior data vector to obtain the attribute feature values ​​of the currently triggered item.

[0043] In this embodiment, the step of processing each target behavior data vector based on the item attribute value determination model to obtain the attribute feature value of each triggered item relative to the target user includes: processing each target behavior data vector based on a first attribute value determination sub-model in the item attribute value determination model to obtain a first attribute feature value to be processed; processing each target behavior data vector based on a second attribute determination sub-model in the item attribute value determination model to obtain a second attribute value to be processed; and determining the attribute feature value of each triggered item relative to the target user based on the first attribute value to be processed and the second attribute value to be processed.

[0044] The item attribute value determination model includes two sub-models: a first attribute value determination sub-model and a second attribute value determination sub-model. The item attribute value determination model may employ a deepFM network structure. This network structure mainly consists of a first attribute value determination sub-model (FM structure) and a second attribute determination sub-model (NN structure). The value obtained after processing the target behavior vector based on the first attribute value determination sub-model can be used as the first attribute feature value to be processed. The value obtained after processing the target behavior vector based on the second attribute determination sub-model can be used as the second attribute feature value to be processed. Based on the first and second attribute value values ​​to be processed, the attribute value feature value of each triggered item relative to the target user can be determined.

[0045] It should be noted that the dynamic determination of the attribute values ​​of each triggered item primarily relies on real-time data. Optionally, real-time data is generally obtained through message middleware. By integrating real-time data from the live stream and user browsing through the Flink streaming big data processing framework, it gains the ability to process constantly changing and large volumes of data in real time. Real-time data includes user clicks, views, and order data for specific items, all of which are updated in real-time.

[0046] S130. Based on the attribute feature values, determine the target items associated with the target user.

[0047] In this embodiment, items of interest to the target user can be determined based on various attribute feature values, and then corresponding items can be pushed to the user based on the item's three-level category. The pushed items can be used as target items. Alternatively, the attribute feature values ​​of each triggered item can be used as a feature in the user profile database, so that corresponding target items can be pushed to the target user in the future based on the user profile database.

[0048] In this embodiment, determining the target item associated with the target user based on each attribute feature value includes: identifying triggered items corresponding to attribute feature values ​​higher than preset attribute values ​​as items to be determined; determining the target item associated with the item to be determined based on the item attributes of the item to be determined, so as to display the target item when the target user triggers the client.

[0049] The preset attribute values ​​are determined based on practical experience and are used to determine whether each triggered item is of interest to the user. Triggered items with attribute values ​​higher than the preset values ​​can be designated as items to be identified. The system can also determine the three-level category of the items to be identified, and items associated with that category can be designated as target items. When a target user triggers the client application, the identified target items can be displayed on the client for the user to choose from, thereby improving the conversion rate.

[0050] Specifically, items with attribute values ​​exceeding preset limits can be designated as pending items. The attributes of these pending items are obtained, and then associated items are identified based on these attributes. These associated items are then designated as target items. When a target user triggers the client application, the identified target items can be displayed on the screen for the user to select. Alternatively, the attribute information of the pending items can be used as a reference factor in the user profile, allowing for the delivery of appropriate target items to the target user based on that profile.

[0051] The technical solution of this invention, by acquiring basic behavioral data and event tracking data associated with the target user, can determine the basic behavioral sub-data and event tracking sub-data corresponding to each triggered item. After inputting the basic behavioral sub-data and event tracking data into a pre-trained item attribute value determination model, the attribute feature values ​​of the triggered item relative to the target user can be obtained. Based on the attribute feature values ​​of each triggered item, the target item associated with the target user can be determined. When the user triggers the client, the target item can be displayed to the target client for the user to choose from. This solves the problem in the prior art where, when determining the items of interest to the user based on the number of times the user triggers each item and the click rate, the impact of different triggering operations on the items is not considered, resulting in a certain error rate between the items pushed to the user and the items that the user is actually interested in, thus causing a low item conversion rate. This invention achieves the technical effect of distinguishing the impact of different triggering behaviors on the user based on event tracking data, thereby improving the matching degree between the pushed items and the user, and thus improving the user experience.

[0052] Example 2

[0053] Figure 2 This is a schematic flowchart of a method for determining a target item according to Embodiment 2 of the present invention. Based on the foregoing embodiments, an item attribute value determination model can be pre-trained. Then, based on the item attribute value determination model, the user-triggered items are processed to determine the user's preference for each triggered item. For specific implementation details, please refer to the detailed description of this technical solution. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0054] like Figure 2 As shown, the method includes:

[0055] S210. Obtain the training sample set.

[0056] The training sample set includes multiple training samples, each containing training association data and an identifier corresponding to the training association data.

[0057] It should be noted that, in order to ensure that the trained model has high accuracy, as many training samples as possible should be obtained.

[0058] Each training sample includes training association data. Optionally, the training association data may include basic user behavior data triggered by a specific item, and corresponding event tracking data. Event tracking data may include the action corresponding to the click, indicating whether the action is positive or negative. The identifier corresponding to the training association data can be an indicator of whether the action is positive or negative. For example, if a user's triggering behavior includes repeatedly clicking to browse the same item and purchasing it within a preset time, then the identifier corresponding to this item could be a positive identifier; optionally, a positive identifier can be represented by 0. Similarly, if a user's triggering behavior includes deleting an item from the add-to-cart list, then the identifier corresponding to this item could be a negative identifier; optionally, a negative identifier can be represented by 1.

[0059] As can be seen from the above, the identifier corresponding to the training associated data is used to characterize whether the training sample data is positive or negative behavior.

[0060] In this embodiment, training samples can be obtained based on a pre-built offline database. That is, by analyzing and processing offline data, the identifiers corresponding to each item are determined, and then the offline data is used as training samples.

[0061] Specifically, offline data can be obtained from Hadoop and Hive environments. Optionally, it can be based on pre-built offline data from massive datasets, user purchasing power data for different product categories, and third-level product category data. Offline data is updated using T+1 or T+n intervals, where T represents the period. After the update is complete, in order for the real-time data processing framework to promptly obtain the offline data from the Hadoop file system, the updated data needs to be pushed into the in-memory database Redis. Then, it is called through the Flink framework to train the model for determining the item attribute values.

[0062] The construction of offline foundational data aims to save data such as user clicks, browsing duration, number of clicks, and entry methods during a product detail page browsing process. This data serves as the foundation for subsequent feature processing, i.e., basic behavioral data. This data primarily comes from offline click and browsing data tables, stored on HDFS in Hive tables. We use UUID and SKU as the granularity for statistical data, taking the first click as the entry time and the last click as the end time, with the time difference representing the browsing duration. If the interval between two clicks exceeds 1 minute, it is considered two separate clicks, and the data at this point can be used as foundational data, stored in a Hive table. Based on this method, multiple training sample data can be obtained.

[0063] S220. For each training sample, input the training association data of the current training sample into the item attribute value determination model to be trained, and obtain the actual output value corresponding to the current training sample.

[0064] After obtaining the training sample set, a model can be determined based on the attribute values ​​of the items to be trained, using each training sample in the set. The newly acquired model can be used as the attribute value determination model, with its parameters set to default values. The actual output value can be understood as the result obtained after processing the samples based on the existing model parameters after inputting the current training sample data into the attribute value determination model.

[0065] It should be noted that before training the model for determining item attribute values, the training parameters in the model can be set to default values. During training, the training parameters can be adjusted based on the output of the model; that is, the model for determining item attribute values ​​can be obtained by modifying the loss function in the model.

[0066] Specifically, each training sample in the training sample set can be sequentially input into the item attribute value determination model to be trained, obtaining the actual output value corresponding to the training sample. After obtaining the actual output value, S230 can be executed. After correcting the model parameters to be corrected in the item attribute value determination model to be trained based on S230, the process is repeated, inputting other training samples different from the previous training sample into the item attribute value determination model to be trained.

[0067] S230. Based on the actual output value corresponding to the training sample and the identifier corresponding to the training sample, determine the function value of the preset loss function, and adjust the model parameters to be corrected in the model based on the attribute value of the item to be trained.

[0068] The item attribute value determination model is obtained by processing the training association information based on the item attribute value determination model to be trained. The actual output value is the evaluation value corresponding to the training association information in the training sample data. The loss function is pre-set and used to measure whether the output value of the determined item attribute value determination model to be trained is accurate.

[0069] In this embodiment, the item attribute value determination model to be trained includes a first attribute value determination sub-model and / or a second attribute value determination sub-model. The step of inputting the current training sample data into the item attribute value determination model to obtain the actual output value corresponding to the current training sample includes:

[0070] The current training sample data is input into the vector construction layer of the item attribute value determination model to obtain a sample vector corresponding to the current training sample data; the sample vector is processed based on the offset and the model parameters to be corrected in the first attribute value determination sub-model to obtain a first actual output value corresponding to the sample vector; and / or, the sample vector is processed based on the second attribute value determination sub-model to obtain a second actual output value; the actual output value is obtained according to the first actual output value and / or the second actual output value.

[0071] This can be understood as the model for determining the attribute values ​​of the item to be trained comprising two sub-models: a first attribute value determination sub-model and a second attribute value determination sub-model. The output of the first attribute value determination sub-model can be used as the first actual output value, and the output of the second attribute value determination sub-model can be used as the second actual output value. The sum of the first and second actual output values ​​can be used as the actual output value.

[0072] Specifically, the basic behavioral data can be processed by the vector construction layer in the model based on the attribute values ​​of the item to be trained, resulting in a basic vector corresponding to the basic behavioral data. Based on the basic vector and the tracking vector, the target behavioral data vector can be obtained. The target behavioral vector is input into the first attribute value determination sub-model to obtain the first actual output value. Simultaneously, the target behavioral vector is input into the second attribute value determination sub-model to obtain the actual output value. Based on the first and second actual output values, the final output value is obtained.

[0073] In this embodiment, the step of determining the offset in the sub-model and the model parameters to be corrected based on the first attribute value to process the sample vector and obtain the first actual output value corresponding to the sample vector includes: determining the product of the model parameters to be corrected and the values ​​corresponding to each feature in the sample vector to obtain a first intermediate value with the same number of features; sequentially taking the value corresponding to each feature in the sample vector as a reference value and accumulating the product of the reference value and the value corresponding to each feature to obtain a second intermediate value; and determining the first actual output value based on the first intermediate value, the second intermediate value, and the offset.

[0074] This can be understood as the first attribute value being determined in the sub-model and represented by the offset and the parameters of the model to be corrected.

[0075] Specifically, the sub-model processing of the target behavior data vector based on the first attribute value can be achieved by: calculating the product between the parameters of the model to be corrected and the numerical value corresponding to each feature in the sample vector, resulting in multiple first intermediate values. The number of first intermediate values ​​is consistent with the number of features. Further, the numerical value corresponding to each feature can be used as a baseline value, and the product between each baseline value and the numerical values ​​corresponding to other features can be calculated to obtain multiple second intermediate values. By accumulating the multiple first intermediate values, multiple second intermediate values, and the offset, the output value corresponding to the training sample vector can be obtained, which can be the first actual output value.

[0076] For example, the first item attribute value determining the sub-model can be represented by this formula:

[0077]

[0078] in n represents the number of features, i represents a specific feature in the training vector, and w i Let u represent the model parameters to be corrected, x represent the numerical value corresponding to a certain feature, and j represent a certain feature in the training vector. Used to represent the cross result of features, that is, taking the vector of a certain feature as the reference value, calculating the product of the reference value and the feature values ​​corresponding to the other features respectively, w0 represents the offset, which is a preset value.

[0079] The second item attribute value determination sub-model can also be represented by a formula: a (l+1) =σ(W (l) a (l) +b (l) Where a represents the input of the neuron, l represents a neural layer in the neural network, and w and b represent the parameters to be corrected in the second item attribute value determination sub-model.

[0080] S240. Using the convergence of the preset loss function as the training objective, train the item attribute value determination model to obtain the item attribute value determination model.

[0081] Specifically, the training error of the loss function can be used as a condition to detect whether the loss function has reached convergence. For example, this could be whether the training error is less than a preset error, whether the error trend is stable, or whether the current number of iterations equals a preset number. If convergence is detected, such as the training error of the loss function being less than the preset error or the error trend being stable, it indicates that the initial driving quality assessment model has been trained successfully, and iterative training can be stopped. If convergence has not been detected, training samples from the training data can be further obtained to train the item attribute value determination model until the training error of the loss function is within a preset range. When the training error of the loss function converges, the item attribute value determination model can be used as the driving quality assessment model. The user's preference for each triggered item can be determined based on the trained item attribute value determination model.

[0082] Based on the above technical solution, after training the item attribute value determination sub-model, the effectiveness of the item attribute value determination sub-model can be determined. Experiments show that the model used in this technical solution is effective.

[0083] This can effectively improve the accuracy, precision, and recall rate of item predictions. See Table 1 below.

[0084] Table 1

[0085] accuracy Accuracy recall F1 Xgboost 0.2347 0.2181 0.993 0.3577 DeepFM 0.6101 0.3376 0.8497 0.4833 Onehot+DeepFM 0.7476 0.449 0.7771 0.5692 Node2vec+DeepFM 0.8525 0.6649 0.6305 0.6472

[0086] As can be seen from the above, the model structure and data representation provided in this technical solution can effectively improve the accuracy of item prediction.

[0087] S250. Based on the acquired basic behavior data and event tracking data associated with the target user, determine the basic behavior sub-data and event tracking sub-data corresponding to each triggered item.

[0088] It should be noted that offline data has a certain time delay in practical applications. When faced with real-time processing tasks, it is impossible to process based on offline data. Therefore, an online system can be built based on this technical solution, that is, the model for determining the item attribute values ​​obtained through training can be used in the online data acquisition system.

[0089] Specifically, the real-time online system uses a message queue (JDQ) to obtain real-time click and browsing log data for various items. Flink is used to access the JDQ data and construct basic user browsing data. To process this basic data into the required features, the Java API of TensorFlow is introduced, specifically the interface corresponding to the basic behavioral data, to retrieve the relevant data. During this process, the results, represented by offline tracking embeddings, are stored in Redis for use by real-time tasks. This ensures that our algorithm meets timeliness requirements.

[0090] It should be noted that this technical solution utilizes the Node2Vec algorithm in graph neural networks and user browsing behavior to embed and represent the event tracking points on the product details page. The vectorized event tracking points and the graph network structure are then used to classify user click behavior on the product details page, filtering out invalid and low-quality clicks. At the task implementation level, the offline task utilizes massive amounts of offline data in HDFS and a distributed system architecture to train the parameters in the network structure. The real-time task accesses data through the Kafka message middleware and the Flink streaming data processing framework, and predicts model results by calling the TensorFlow Java interface.

[0091] S260. Based on the basic behavioral sub-data, embedded data, and pre-trained item attribute value determination model of each triggered item, the attribute feature value of each triggered item relative to the target user is obtained.

[0092] The technical solution disclosed in this invention utilizes the Node2Vec algorithm in graph neural networks and user browsing behavior to embed and represent the event tracking points on the product detail page. It then uses the vectorized event tracking points and the graph network structure to classify user click behavior on the product detail page, filtering out invalid and low-quality clicks. At the task implementation level, the offline task utilizes massive amounts of offline data in HDFS and a distributed system architecture to train the parameters in the network structure. The real-time task accesses data through the Kafka message middleware and the Flink streaming data processing framework, calls the TensorFlow Java interface to predict model results, and then delivers corresponding items to the user based on the prediction results, improving the matching degree between the pushed items and the user, thereby enhancing the user experience.

[0093] It should also be noted that the Graph Embedding model for determining item attribute values ​​uses the Node2Vec algorithm in its implementation, similar to DeepWalk. The introduction of different embedding algorithms will have a certain impact on the performance of the network model. Therefore, different embedding algorithms may achieve results similar to those described in this patent. Furthermore, different types of network structures, such as the Wide&Deep model, can replace the DeepFm algorithm used in this paper to model and analyze user clicks on product detail pages. In other words, regardless of the model used, as long as the data processed from tracking points and basic behavioral data is used to determine attribute feature values, it falls within the scope of protection of this technical solution.

[0094] S270. Based on the attribute feature values, determine the target items associated with the target user.

[0095] The technical solution of this invention, by acquiring basic behavioral data and event tracking data associated with the target user, can determine the basic behavioral sub-data and event tracking sub-data corresponding to each triggered item. After inputting the basic behavioral sub-data and event tracking data into a pre-trained item attribute value determination model, the attribute feature values ​​of the triggered item relative to the target user can be obtained. Based on the attribute feature values ​​of each triggered item, the target item associated with the target user can be determined. When the user triggers the client, the target item can be displayed to the target client for the user to choose from. This solves the problem in the prior art where, when determining the items of interest to the user based on the number of times the user triggers each item and the click rate, the impact of different triggering operations on the items is not considered, resulting in a certain error rate between the items pushed to the user and the items that the user is actually interested in, thus causing a low item conversion rate. This invention achieves the technical effect of distinguishing the impact of different triggering behaviors on the user based on event tracking data, thereby improving the matching degree between the pushed items and the user, and thus improving the user experience.

[0096] Example 3

[0097] Figure 3 This is a schematic diagram of the structure of a business data processing device provided in Embodiment 4 of the present invention, as shown below. Figure 3 The device includes: a data determination module 310, an attribute feature value determination module 320, and a target item determination module 330.

[0098] The data determination module 310 is used to determine the basic behavior sub-data and sub-data corresponding to each triggered item based on the acquired basic behavior data and tracking data associated with the target user; the attribute feature value determination module 320 is used to obtain the attribute feature value of each triggered item relative to the target user based on the basic behavior sub-data, tracking data and the pre-trained item attribute value determination model of each triggered item; and the target item determination module 330 is used to determine the target item associated with the target user based on each attribute feature value.

[0099] Based on the above technical solution, the data processing module, before processing the acquired basic behavior data and event tracking data, is further configured to: extract basic behavior data and event tracking data of each user for each triggered item within a preset time period from the system log; and determine the basic behavior data and event tracking data of each user for each triggered item within a preset time period according to pre-set rules, so as to determine the basic behavior sub-data and event tracking sub-data of each triggered item corresponding to each user from the basic behavior data and event tracking behavior data.

[0100] Based on the above technical solution, the attribute feature value determination module includes:

[0101] The embedding vector determination unit is used to determine the embedding vector based on the embedding sub-data of each triggered item;

[0102] The behavior data vector determination unit is used to determine the basic behavior vector corresponding to each basic behavior sub-data based on the vector construction layer in the item attribute value determination model, and to determine the target behavior data vector of each triggered item based on each basic behavior vector and the corresponding tracking point vector.

[0103] The attribute feature value determination unit is used to process each target behavior data vector based on the item attribute value determination model to obtain the attribute feature value of each triggered item relative to the target user.

[0104] Based on the above technical solution, the attribute feature value determination unit further includes:

[0105] The first attribute feature value determination subunit is used to process each target behavior data vector based on the first attribute value determination submodel in the item attribute value determination model to obtain the first attribute feature value to be processed.

[0106] The second attribute feature value determination subunit is used to process each target behavior data vector based on the second attribute determination submodel in the item attribute value determination model to obtain the second attribute value to be processed.

[0107] The attribute feature value determination subunit is used to determine the attribute feature value of each triggered item relative to the target user based on the first attribute value to be processed and the second attribute value to be processed.

[0108] Based on the above technical solution, the target item determination module includes:

[0109] The item determination unit is used to identify triggered items whose attribute feature values ​​are higher than preset attribute values ​​as items to be determined.

[0110] The target item determination unit is used to determine the target item associated with the item to be determined based on the item attributes of the item to be determined, so as to display the target item when the target user triggers the client.

[0111] Based on the above technical solution, the device further includes: a model training module, comprising:

[0112] A sample set acquisition unit is used to acquire a training sample set; wherein, the training sample set includes multiple training samples, and the training samples include training association data and an identifier corresponding to the training association data;

[0113] The actual output value determination unit is used to input the current training sample data into the item attribute value determination model to be trained for each training sample data, and obtain the actual output value corresponding to the current training sample.

[0114] The model parameter correction unit is used to determine the function value of the preset loss function based on the actual output value corresponding to the training sample data and the identifier corresponding to the training sample data, and to adjust the model parameters to be corrected in the model based on the calculation results of the attribute values ​​of the item to be trained.

[0115] The item attribute value determination model determination unit is used to train the item attribute value determination model to obtain the item attribute value determination model by taking the convergence of the preset loss function as the training objective.

[0116] Based on the above technical solution, the actual output value determination unit includes:

[0117] The sample vector determination subunit is used to input the current training sample data into the vector construction layer in the item attribute value determination model to be trained, so as to obtain a sample vector corresponding to the current training sample data.

[0118] The first actual output value determination subunit is used to process the sample vector based on the offset and model parameters to be corrected in the sub-model determined by the first attribute value, to obtain the first actual output value corresponding to the sample vector; and / or,

[0119] The second actual output value determination subunit is used to process the sample vector based on the second attribute value determination submodel to obtain the second actual output value;

[0120] The actual output value determination subunit is used to obtain the actual output value based on the first actual output value and / or the second actual output value.

[0121] Based on the above technical solutions, the first actual output value determining subunit includes:

[0122] The first intermediate value determination subunit is used to determine the product of the model parameter to be corrected and the value corresponding to each feature in the sample vector to obtain a first intermediate value with the same number of features; the second intermediate value determination subunit is used to take the value corresponding to each feature in the sample vector as a reference value in turn, and accumulate the product of the reference value and the value corresponding to each feature to obtain a second intermediate value; the actual output value determination subunit is used to determine the first actual output value based on the first intermediate value, the second intermediate value and the offset.

[0123] The technical solution of this invention, by acquiring basic behavioral data and event tracking data associated with the target user, can determine the basic behavioral sub-data and event tracking sub-data corresponding to each triggered item. After inputting the basic behavioral sub-data and event tracking data into a pre-trained item attribute value determination model, the attribute feature values ​​of the triggered item relative to the target user can be obtained. Based on the attribute feature values ​​of each triggered item, the target item associated with the target user can be determined. When the user triggers the client, the target item can be displayed to the target client for the user to choose from. This solves the problem in the prior art where, when determining the items of interest to the user based on the number of times the user triggers each item and the click rate, the impact of different triggering operations on the items is not considered, resulting in a certain error rate between the items pushed to the user and the items that the user is actually interested in, thus causing a low item conversion rate. This invention achieves the technical effect of distinguishing the impact of different triggering behaviors on the user based on event tracking data, thereby improving the matching degree between the pushed items and the user, and thus improving the user experience.

[0124] The apparatus for determining a target item provided in the embodiments of the present invention can execute the method for determining a target item provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0125] It is worth noting that the various units and modules included in the above-mentioned device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of the present invention.

[0126] Example 4

[0127] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. Figure 4 A block diagram is shown of an exemplary electronic device 40 suitable for implementing embodiments of the present invention. Figure 4 The electronic device 40 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0128] like Figure 4 As shown, electronic device 40 is represented in the form of a general-purpose computing device. The components of electronic device 40 may include, but are not limited to: one or more processors or processing units 401, system memory 402, and bus 403 connecting different system components (including system memory 402 and processing unit 401).

[0129] Bus 403 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0130] Electronic device 40 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 40, including volatile and non-volatile media, removable and non-removable media.

[0131] System memory 402 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 404 and / or cache memory 405. Electronic device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 406 may be used to read and write non-removable, non-volatile magnetic media (… Figure 4 Not shown; usually referred to as a "hard drive"). Although Figure 4Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 403 via one or more data media interfaces. Memory 402 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0132] A program / utility 408 having a set (at least one) of program modules 407 may be stored, for example, in memory 402. Such program modules 407 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 407 typically perform the functions and / or methods described in the embodiments of the present invention.

[0133] Electronic device 40 can also communicate with one or more external devices 409 (e.g., keyboard, pointing device, display 410, etc.), and with one or more devices that enable a user to interact with the electronic device 40, and / or with any device that enables the electronic device 40 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 411. Furthermore, electronic device 40 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 412. As shown, network adapter 412 communicates with other modules of electronic device 40 via bus 403. It should be understood that, although... Figure 4 As not shown, other hardware and / or software modules may be used in conjunction with electronic device 40, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0134] The processing unit 401 executes various functional applications and data processing by running programs stored in the system memory 402, such as implementing the method for determining a target item provided in the embodiments of the present invention.

[0135] Example 5

[0136] Embodiment 5 of the present invention also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a method for determining a target item.

[0137] The method includes:

[0138] Based on the acquired basic behavioral data and event tracking data associated with the target user, determine the basic behavioral sub-data and event tracking sub-data corresponding to each triggered item;

[0139] Based on the basic behavioral sub-data, embedded data, and pre-trained item attribute value determination model of each triggered item, the attribute feature values ​​of each triggered item relative to the target user are obtained.

[0140] Based on the attribute feature values, the target items associated with the target user are determined.

[0141] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0142] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0143] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0144] Computer program code for performing the operations of embodiments of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0145] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method for determining a target item, characterized in that, include: Based on the acquired basic behavioral data and event tracking data associated with the target user, basic behavioral sub-data and event tracking sub-data corresponding to each triggered item are determined; wherein, the basic behavioral data is constructed by real-time access of the target user's click and browsing log data of the triggered item in the Kafka message queue by Flink, and the event tracking sub-data indicates whether the basic behavioral sub-data is positive behavioral data or negative behavioral data. Based on the basic behavior sub-data, event tracking sub-data, and pre-trained item attribute value determination model of each triggered item, the attribute feature values ​​of each triggered item relative to the target user are obtained, including: determining event tracking vectors based on the event tracking sub-data of each triggered item; determining basic behavior vectors corresponding to each basic behavior sub-data based on the vector construction layer in the item attribute value determination model, and determining target behavior data vectors of each triggered item based on each basic behavior vector and the corresponding event tracking vector; processing each target behavior data vector based on the item attribute value determination model to obtain the attribute feature values ​​of each triggered item relative to the target user; wherein, the item attribute value determination model includes a first attribute value determination sub-model and a second attribute determination sub-model, and the attribute feature values ​​include a first attribute feature value to be processed and a second attribute feature value to be processed, wherein the first attribute feature value to be processed is determined based on the first attribute value determination sub-model, and the second attribute feature value to be processed is determined based on the second attribute determination sub-model; Based on the attribute feature values, the target items associated with the target user are determined.

2. The method according to claim 1, characterized in that, Before the acquisition of basic behavioral data and event tracking data, the following is also included: Extract basic behavioral data and event tracking data of each user for each triggered item within a preset time period from the system logs; Based on pre-set rules, determine the basic behavioral data and event tracking data of each user for each triggered item within a preset time period, so as to determine the basic behavioral sub-data and event tracking sub-data of each triggered item corresponding to each user from the basic behavioral data and event tracking data.

3. The method according to claim 1, characterized in that, The model based on the item attribute values ​​processes each target behavior data vector to obtain the attribute feature value of each triggered item relative to the target user, including: The first attribute value determination sub-model in the item attribute value determination model processes each target behavior data vector to obtain the first attribute feature value to be processed. The second attribute determination sub-model in the item attribute value determination model processes each target behavior data vector to obtain the feature value of the second attribute to be processed. Based on the first attribute feature value to be processed and the second attribute feature value to be processed, the attribute feature value of each triggered item relative to the target user is determined.

4. The method according to claim 1, characterized in that, The process of determining the target items associated with the target user based on each attribute feature value includes: Items that have already been triggered and whose attribute feature values ​​are higher than the preset attribute values ​​will be designated as items to be determined. Based on the item attributes of the item to be determined, a target item associated with the item to be determined is determined, so that the target item is displayed when the target user triggers the client.

5. The method according to claim 1, characterized in that, Also includes: Obtain a training sample set; wherein the training sample set includes multiple training samples, and the training samples include training association data and an identifier corresponding to the training association data; For each training sample, the training association data of the current training sample is input into the item attribute value determination model to be trained, and the actual output value corresponding to the current training sample is obtained. Based on the actual output value corresponding to the training sample and the identifier corresponding to the training sample, the function value of the preset loss function is determined, and the model parameters to be corrected in the model are adjusted according to the function value of the attribute value of the item to be trained. The convergence of the preset loss function is taken as the training objective, and the model for determining the item attribute value is trained to obtain the model for determining the item attribute value.

6. The method according to claim 5, characterized in that, The model for determining the attribute values ​​of the items to be trained includes a first attribute value determination sub-model and / or a second attribute value determination sub-model. The step of inputting the training association data of the current training sample into the model for determining the attribute values ​​of the items to be trained to obtain the actual output value corresponding to the current training sample includes: The training association data of the current training sample is input into the vector construction layer in the model for determining the attribute value of the item to be trained, so as to obtain the sample vector corresponding to the current training sample data. Based on the first attribute value, the offset in the sub-model and the parameters of the model to be corrected are determined, and the sample vector is processed to obtain a first actual output value corresponding to the sample vector; and / or, Based on the second attribute value, the sub-model is determined to process the sample vector to obtain the second actual output value; The actual output value is obtained based on the first actual output value and / or the second actual output value.

7. The method according to claim 6, characterized in that, The step of determining the offset in the sub-model and the model parameters to be corrected based on the first attribute value, and processing the sample vector to obtain the first actual output value corresponding to the sample vector, includes: The product of the parameters of the model to be corrected and the values ​​corresponding to each feature in the sample vector is determined to obtain a first intermediate value that is the same as the number of features. The value corresponding to each feature in the sample vector is taken as the baseline value in turn, and the product of the baseline value and the value corresponding to each feature is accumulated to obtain the second intermediate value. The first actual output value is determined based on the first intermediate value, the second intermediate value, and the offset.

8. A device for determining a target article, characterized in that, include: The data determination module is used to determine the basic behavior sub-data and the tracking sub-data corresponding to each triggered item based on the basic behavior data and tracking data associated with the target user. The basic behavior data is constructed by real-time access of the target user's click and browsing log data of the triggered item to the Kafka message queue via Flink. The tracking sub-data indicates whether the basic behavior sub-data is positive behavior data or negative behavior data. The attribute feature value determination module is used to obtain the attribute feature value of each triggered item relative to the target user based on the basic behavior sub-data, the tracking point sub-data, and the pre-trained item attribute value determination model of each triggered item. This includes: determining the tracking point vector based on the tracking point sub-data of each triggered item; determining the basic behavior vector corresponding to each basic behavior sub-data based on the vector construction layer in the item attribute value determination model, and determining the target behavior data vector of each triggered item based on each basic behavior vector and the corresponding tracking point vector; processing each target behavior data vector based on the item attribute value determination model to obtain the attribute feature value of each triggered item relative to the target user; wherein, the item attribute value determination model includes a first attribute value determination sub-model and a second attribute determination sub-model, and the attribute feature value includes a first attribute feature value to be processed and a second attribute feature value to be processed, wherein the first attribute feature value to be processed is determined based on the first attribute value determination sub-model, and the second attribute feature value to be processed is determined based on the second attribute determination sub-model; The target item determination module is used to determine the target items associated with the target user based on various attribute feature values.

9. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining a target item as described in any one of claims 1-7.

10. A storage medium comprising computer-executable instructions, which, when executed by a computer processor, are used to perform the method for determining a target article as described in any one of claims 1-7.

Images