Message push methods, devices, computer equipment, and storage media

Abstract

This application relates to a message push method, apparatus, computer device, and storage medium. The method includes: acquiring an initial sample set, comprising positive samples with qualified activity participation and negative samples with unqualified activity participation; converting negative samples that meet label conversion conditions into positive samples to obtain an intermediate sample set; inputting the intermediate sample set into a pre-trained causal model, identifying common and individual factors among the samples through the causal model, and labeling the basic information containing the common and individual factors in the samples with common and individual factor labels to obtain a labeled final sample set; inputting the final sample set into a multi-label multi-classification model, which predicts the target activity type corresponding to each sample, and pushing activity messages corresponding to the target activity type to each sample. The above method can solve problems such as insignificant individual features and the sparsity of training due to insufficient positive samples, preventing effective scaling.

Description

Technical Field

[0001] This application relates to the field of activity message push technology, and in particular to a message push method, apparatus, computer device and storage medium. Background Technology

[0002] In digital marketing, an existing seed audience can be used to find a group of similar or most likely conversion-oriented expansion audiences, thereby expanding the reach of ad placements or targeted marketing. Existing technologies use feature approximation and preference models to deliver campaigns or advertisements, and then use marketing results to tag users to help with customer acquisition. During customer acquisition, business personnel analyze the tag distribution of the respondent audience and use experience to select similar groups for expansion. Preference models use methods such as Propensity Score Matching (PSM), Decision Tree (DT), or GBDT to build classification models to classify non-seed users and identify potential similar groups.

[0003] Existing model methods, which use business features and reclassify labels based on result data, suffer from the problem that the same seed user's label is repeatedly hit in different activities due to the large number of common features among different activities. This leads to the seed user's label weight being increased when relabeling the seed user, thus increasing the seed's hit rate. Furthermore, in complex multi-activity push environments, the excessive sharing of features between activities makes it impossible to distinguish the features of different types of activities, resulting in low accuracy of activity push. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of targeted marketing to similar customers, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a message push method. The method includes:

[0006] Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0007] Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set;

[0008] The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. Common factors are the common features among different activities; individual factors are the individual features that distinguish different types of activities.

[0009] The final sample set is input into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0010] In one embodiment, negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set, including:

[0011] The positive samples in the initial sample set are used as the initial positive sample set. Cluster analysis is performed on the label distribution of the initial positive sample set to obtain the trained classifier.

[0012] Each negative sample in the initial sample set is input into the classifier, and the classifier calculates the probability value of each negative sample being labeled as a positive sample.

[0013] Negative samples with a probability value greater than a preset value are used as target negative samples that meet the label conversion conditions. The second label of the target negative sample is modified to represent that the activity participation of the target negative sample meets the standard, so as to convert the target negative sample into a positive sample.

[0014] Add the target negative sample to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process;

[0015] The positive sample expansion process is iterated multiple times until the number of positive samples in the final initial positive sample set reaches the required level. Then the iteration stops, and an intermediate sample set is obtained.

[0016] In one embodiment, each negative sample is input into a classifier, which calculates the probability value of each negative sample being labeled as a positive sample, including:

[0017] Each negative sample is input into the classifier, which predicts the first probability value of each negative sample being labeled and the second probability value of each positive sample being labeled, and calculates the mean probability of the positive sample being labeled.

[0018] The ratio of the square of the probability mean to the first probability value is used to determine the probability that a negative sample is labeled as a positive sample.

[0019] In one embodiment, the final sample set is input into a multi-label multi-classification model. The model predicts the target activity type for each sample based on a first label, a second label, a third label, a common factor label, and a unique factor label. It then pushes activity messages corresponding to the target activity type to each sample, including:

[0020] The first label, second label, third label, common factor label, and individual factor label of each sample are transformed into dummy variables to obtain multiple single-label binary classification sub-models;

[0021] Multiple single-label binary classification sub-models are used to predict the activity type of each sample. The activity types predicted by the multiple single-label binary classification sub-models are classified, and the activity type with the most classification results is taken as the target activity type corresponding to the sample.

[0022] In one embodiment, the method further includes:

[0023] Get the activity messages to be pushed;

[0024] In the initial sample set, select target samples whose target activity type is the same as the type of the activity message to be pushed, and prioritize pushing the activity message to the target samples.

[0025] In one embodiment, the method further includes:

[0026] Obtain the activity data of the user to be pushed. If the activity data of the user to be pushed belongs to the initial sample set, determine the target activity type corresponding to the user to be pushed in the initial sample set, and push the activity message corresponding to the target activity type to the user to be pushed.

[0027] Secondly, this application also provides a message push device. The device includes:

[0028] The acquisition module is used to acquire an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0029] The sample expansion module is used to convert negative samples that meet the label conversion conditions into positive samples to obtain an intermediate sample set;

[0030] The causal identification module is used to input the intermediate sample set into the pre-trained causal model. The causal model identifies the common and individual factors among the samples, and labels the basic information containing the common factors with common factor labels, and labels the basic information containing the individual factors with individual factor labels, to obtain the labeled final sample set. The common factors are the common features among different activities; the individual factors are the individual features that distinguish different types of activities.

[0031] The multi-classification module is used to input the final sample set into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0032] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0033] Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0034] Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set;

[0035] The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. Common factors are the common features among different activities; individual factors are the individual features that distinguish different types of activities.

[0036] The final sample set is input into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0037] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0038] Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0039] Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set;

[0040] The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. Common factors are the common features among different activities; individual factors are the individual features that distinguish different types of activities.

[0041] The final sample set is input into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0042] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0043] Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0044] Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set;

[0045] The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. Common factors are the common features among different activities; individual factors are the individual features that distinguish different types of activities.

[0046] The final sample set is input into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0047] The aforementioned message push method, device, computer equipment, and storage medium convert negative samples that meet the label conversion conditions into positive samples, thereby obtaining a sufficient number of positive samples. This addresses the problems of insignificant individual characteristics and insufficient positive samples leading to sparse training and ineffective scaling. The intermediate sample set after positive sample expansion is used as input to the causal model. By enhancing and expanding the positive samples, combined with the causal model to identify common and individual factors, and weakening the identified common factors, individual factors between different activity types can be strengthened. This effectively distinguishes common and individual factors between different types of activities. Furthermore, a multi-label, multi-classification model is used to predict the target activity type corresponding to the sample, solving the problem of insignificant individual factors and improving the accuracy of targeting different activity groups. Attached Figure Description

[0048] Figure 1 This is an application environment diagram of the message push method in one embodiment;

[0049] Figure 2 This is a flowchart illustrating a message push method in one embodiment;

[0050] Figure 3 This is a schematic diagram of the process for obtaining an intermediate sample set in one embodiment;

[0051] Figure 4 This is a schematic diagram of the training process of a semi-supervised framework model in another embodiment;

[0052] Figure 5 This is a schematic diagram of a cause-effect graph in one embodiment;

[0053] Figure 6 This is a flowchart illustrating the process of a multi-label, multi-classification model predicting the target activity type for each sample in one embodiment.

[0054] Figure 7 This is a structural block diagram of a message push device in one embodiment;

[0055] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] The message push method provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown. The process involves the terminal 102 acquiring an initial sample set, which includes multiple samples. Each sample contains activity data of users participating in different activity types, a first label indicating the activity type, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples with the standard activity participation level are designated as positive samples, and samples with the standard activity participation level are designated as negative samples. Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set. The intermediate sample set is then input into a pre-trained causal model, which identifies common and individual factors among the samples. Common factor labels are assigned to the basic information containing common factors, and individual factor labels are assigned to the basic information containing individual factors, resulting in a labeled final sample set. Common factors are the common features among different activities, and individual factors are the individual features that distinguish different activity types. The final sample set is then input into a multi-label multi-classification model, which predicts the target activity type for each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample. Terminal 102 communicates with server 104 via a network. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart TVs, etc. Portable wearable devices can include head-mounted devices, etc. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0058] In one embodiment, such as Figure 2 As shown, a message push method is provided, which can be applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0059] Step 202: Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating the basic information in the activity data. Samples with the activity participation level meeting the standard are taken as positive samples, and samples with the activity participation level not meeting the standard are taken as negative samples.

[0060] One sample corresponds to one user. The activity data includes basic information about the user's participation history in different types of activities, activity category, customer number, whether clicked, and whether the target was met, etc. The basic information can be used to distinguish users and different activities. For example, the basic information can be activity level, activity preference, and activity sensitivity. Activity level can be determined by the ratio of the number of times a user participates in an activity to the total number of activities. Activity preference can be determined by mathematical statistics to determine whether a user prefers a certain type or several types of activities.

[0061] Activity engagement refers to a user's response to a particular type of activity. For example, engagement can be measured by factors such as the duration and frequency of viewing activity information, whether the user clicks on the activity link, downloads the video, or registers as a member. Engagement levels can be assessed from multiple dimensions; for instance, whether a user clicks on the activity link can determine if engagement has met the target.

[0062] Data from several major campaign categories over the past year was extracted from the marketing system, including campaign type, clicks, and achievement markers. Basic information was extracted from the customer relationship system. Activity-related data was processed by combining user login data from multiple channels. Data exploration and processing were performed, removing null and outlier values, standardizing the time frame for seasonal data, and discretizing continuous variables. Data distribution was identified, and standardization and normalization were applied. Basic statistical features utilized tagged data from the customer relationship system; activity levels were normalized based on channel response counts; asset level data was tagged; and a gain model was used for feature processing of certain campaign-sensitive factors.

[0063] Optionally, the terminal acquires activity data of different users participating in different types of historical activities. The activity data includes the user's basic information. One user corresponds to one sample. A first label is used to mark the activity type to which multiple activity data in each sample belong. A second label is used to mark whether the activity participation level of each sample meets the standard. A third label is used to mark the basic information of each activity data in each sample. Samples with unmet activity participation levels are used as negative samples, and samples with met activity participation levels are used as positive samples. The positive and negative samples form an initial sample set.

[0064] Step 204: Convert negative samples that meet the label conversion conditions into positive samples to obtain an intermediate sample set.

[0065] Among them, negative samples that meet the label conversion conditions are the similar groups of positive samples, and are denoted as target negative samples. The intermediate sample set includes the original positive samples in the initial sample set, the target negative samples, and the remaining negative samples after removing the target negative samples from the original negative sample set.

[0066] In practical applications, a large number of users fail to meet the activity participation standards due to reasons such as not being reached or not meeting the criteria. Because the sample size meeting the participation standards is small, the modeling conditions are not met, and it is impossible to identify true potential users. Generally, to ensure accuracy, different activity groups are further segmented, further diluting the effective seeds (i.e., positive samples), making effective scaling difficult. Therefore, to solve the above problems, this embodiment uses a semi-supervised model to identify reliable negative samples from the initial sample set. These reliable negative samples are approximate groups of positive samples. Training is performed on the positive samples and the supervised learning reliable negative samples until the number of positive samples in the initial sample set reaches a preset number. This embodiment uses real qualified users as positive samples and uses enhancement methods to mix positive samples into reliable negative samples for modeling. This enhances and expands the positive samples, thereby solving problems such as insignificant individual characteristics and the sparsity of training due to insufficient positive samples, which prevents effective scaling.

[0067] Optionally, the terminal identifies reliable negative samples, i.e. target negative samples, from the initial sample set through a semi-supervised model, and trains on the reliable negative samples and the original positive samples until the number of positive samples in the initial sample set reaches a preset number, thereby obtaining an intermediate sample set including the original positive samples in the initial sample set, the target negative samples, and the remaining negative samples after removing the target negative samples from the original negative samples in the initial sample set.

[0068] Step 206: Input the intermediate sample set into the pre-trained causal model, identify the common factors and individual factors among the samples through the causal model, and label the basic information containing the common factors in the samples with common factor labels, and label the basic information containing the individual factors in the samples with individual factor labels, to obtain the labeled final sample set; the common factors are the common features among different activities; the individual factors are the individual features that distinguish different types of activities.

[0069] The activity data includes multiple basic information segments, which can be categorized into common factors and individual factors. If a user's basic information includes multiple common and individual factors, samples containing common factors are labeled with common factor tags. For example, different common factor tags can be X1, X2, X3, etc. The final sample set includes a first tag, a second tag, a third tag, a common factor tag, and an individual factor tag for each sample. The more common factors there are, the more common characteristics exist between different activity types, making it more difficult to distinguish user preferences for different activity types.

[0070] Because different activities share many common features, the same seed user's tags are repeatedly hit in different activities. This leads to increased tag weight for the same seed user when re-tagging, increasing the hit rate. Furthermore, in complex multi-activity push environments, excessive sharing of features between activities makes it difficult to distinguish between different activity types, resulting in low accuracy in activity pushes. Therefore, to address these issues, this embodiment uses historical activity data as the object, establishes a causal model based on existing model results and business assumptions, uses causal relationship identification and effect analysis, and employs robustness verification to determine common and unique factors among different types of activities. The intermediate sample set, after positive sample expansion through a semi-supervised model, is used as input to the causal model. By enhancing and expanding the positive samples through a semi-supervised model, combined with the causal model's identification of common and unique factors, and by weakening the identified common factors, the unique factors between different activity types can be strengthened. Therefore, for complex and frequent activities, the causal model can effectively distinguish common and unique factors among different types of activities, solving the problem of insignificant unique factors and improving the accuracy of customer targeting for different activities.

[0071] Optionally, the terminal uses historical activity data as the object, establishes a causal model based on existing model result data and business assumptions, uses causal relationship identification and effect analysis, and uses robustness verification to determine the common and unique factors among different types of activities, thus obtaining a trained causal model. The terminal inputs the intermediate sample set into the pre-trained causal model, identifies the common and unique factors among the samples through the causal model, and labels the basic information containing common factors in the samples with common factor labels, and labels the basic information containing unique factors in the samples with unique factor labels, thus obtaining the labeled final sample set.

[0072] Step 208: Input the final sample set into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes the activity message corresponding to the target activity type to each sample.

[0073] The target activity type is the type of activity that users prefer or are most interested in. After steps 204 and 206 above, on the one hand, we obtained expanded data sensitive to the activities, and on the other hand, we also marked the activity influencing factors. Step 208 analyzes the sensitivity factors of each item to obtain the sensitivity characteristics of each activity, so as to make a reference for ranking when launching the sub-activities.

[0074] It is important to note that when analyzing various personality factors, the multi-label multi-classification model uses a uniform minimum weighting method to manually reduce the weight of common factors in order to increase the weight of non-confounding factors, improve the sensitivity of personality factors, and prevent the same seed user's label from being repeatedly hit in different activities.

[0075] This embodiment deploys offline and real-time feature processing. Basic information is stored locally, with causal observation data primarily processed offline to generate relevant common features. Real-time features (mainly behavioral and path-related data) are calculated from the marketing system using its own engine and stored in Redis.

[0076] For example, a sample in the final sample set includes activity data of user a participating in activities A, B, C, and D, as well as first label, second label, third label, common factor label, and individual factor label. The final sample set is input into a multi-label multi-classification model, which will output the encoded value corresponding to each label. A threshold is set for each label as the basis for judging whether it belongs to that label. The multi-label multi-classification model outputs the user's preference for the activities, and takes the maximum preference value as the target activity type corresponding to the sample.

[0077] Optionally, the terminal inputs the final sample set into the multi-label multi-classification model. The multi-label multi-classification model predicts the preference of each sample for different activity types based on the first label, second label, third label, common factor label, and individual factor label, and takes the activity type corresponding to the maximum preference as the target activity type for the sample, and pushes the activity message corresponding to the target activity type to each sample.

[0078] In one embodiment, after determining the target activity type most preferred by each sample, when pushing activity messages, priority is given to samples whose target activity type is the same as the pushed activity type. The specific steps are as follows:

[0079] Step 1: Obtain the activity messages to be pushed.

[0080] The activity message to be pushed can be a historical activity message or an activity message of a similar type to a historical activity message.

[0081] Step 2: Select target samples in the initial sample set whose target activity type is the same as the type of the activity message to be pushed, and push the activity message to be pushed to the target samples first.

[0082] The above steps determine the preferences of each user in the user pool for different activity types, as well as the most preferred activity type for each user, i.e., the target activity type. When pushing a message for an activity to be pushed, target samples with the same target activity type as the message to be pushed are selected from the initial sample set, and the message to be pushed is prioritized to these target samples.

[0083] In one embodiment, after determining the most preferred target activity type for each sample, when analyzing the activity data of users in the user pool, it is first determined whether the user to be pushed to belongs to the initial sample set, i.e., whether it belongs to the user pool. If the user to be pushed does not belong to the initial sample set, i.e., does not belong to the user pool, then after the activity data of the user to be pushed reaches a preset requirement, the user to be pushed is included in the user pool, and the target activity type corresponding to the user to be pushed is determined according to the method described in the above embodiment. If the user to be pushed belongs to the initial sample set, i.e., belongs to the user pool, then the target activity type corresponding to the user to be pushed is determined in the initial sample set, and the activity message corresponding to the target activity type is pushed to the user to be pushed.

[0084] In this embodiment, when pushing out an activity message, a target sample with the same type as the activity message to be pushed is selected from the initial sample set, and the activity message to be pushed is pushed to the target sample first, which can improve the accuracy of activity message delivery.

[0085] In the aforementioned message push method, negative samples that meet the label conversion conditions are converted into positive samples to obtain a sufficient number of positive samples, thus solving the problems of insignificant individual characteristics and insufficient positive samples leading to sparse training and ineffective scaling. The intermediate sample set after positive sample expansion is used as the input of the causal model. By enhancing and expanding the positive samples, combined with the causal model to identify common and individual factors, and weakening the identified common factors, the individual factors between different activity types can be strengthened, effectively distinguishing the common and individual factors between different types of activities. A multi-label multi-classification model is used to predict the target activity type corresponding to the sample, solving the problem of insignificant individual factors and improving the accuracy of customer targeting for different activity groups.

[0086] In one embodiment, such as Figure 3 As shown, negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set, including:

[0087] Step 302: Take the positive samples in the initial sample set as the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, and obtain the trained classifier.

[0088] Optionally, the terminal uses the positive samples in the initial sample set as the initial positive sample set and the negative samples as the initial negative sample set, and uses the expectation-maximization model to perform cluster analysis on the label distribution of the initial positive sample set to train a classifier, wherein the classifier can use a gradient boosting model.

[0089] Step 304: Input each negative sample in the initial sample set into the classifier, and calculate the probability value of each negative sample being labeled as a positive sample through the classifier.

[0090] Optionally, the terminal inputs each negative sample from the initial negative sample set into the classifier, which classifies the negative samples and calculates the probability value of each negative sample being labeled as a positive sample.

[0091] In some embodiments, step 304 specifically includes the following steps:

[0092] Step 1: Input each negative sample into the classifier, predict the first probability value of each negative sample being labeled and the second probability value of each positive sample being labeled through the classifier, and calculate the mean probability of the positive sample being labeled.

[0093] Optionally, the terminal will use a classifier to classify positive and negative samples, using positive sample values ​​as targets for training, and predict the first probability value P(y) of a negative sample x being labeled. x =1|x); Use a classifier to predict the second probability value P(y) of positive samples being labeled in the initial sample set. x =1|y=1|x), based on the second probability value, calculate the mean probability P(y=1|x) of the positive sample being labeled. x =1|y=1).

[0094] Step 2: The ratio of the square of the probability mean to the first probability value is determined as the probability value of a negative sample being labeled as a positive sample.

[0095] Optionally, the probability of a negative sample being labeled as a positive sample can be expressed mathematically as follows:

[0096]

[0097] Step 306: Take negative samples with probability values ​​greater than preset values ​​as target negative samples that meet the label conversion conditions, and modify the second label of the target negative samples to represent the activity participation level of the target negative samples so as to convert the target negative samples into positive samples.

[0098] Among them, such as Figure 4 As shown, the terminal arranges the probability values ​​of negative samples being labeled as positive samples in descending order, selects negative samples with probability values ​​greater than preset values ​​as target negative samples that meet the label conversion conditions, and adds the target negative samples to the initial positive sample set.

[0099] Step 308: Add the target negative sample to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process.

[0100] Optionally, the terminal removes the target negative sample from the initial negative sample set and adds the target negative sample to the initial positive sample set, and continues to train the classifier according to the method in step 302 to obtain a new classifier. At the same time, one iteration can complete one positive sample expansion process.

[0101] Step 310: Iterate the positive sample expansion process multiple times until the number of positive samples in the final initial positive sample set reaches the required level, then stop the iteration and obtain the intermediate sample set.

[0102] In this embodiment, the classifier trained on the positive sample set is used to identify negative samples in the negative sample set that meet the label conversion conditions, i.e., the approximate population, and the approximate population is included in the positive sample set, thereby expanding the number of positive samples and solving the problems of insignificant individual characteristics and the lack of training sparsity and effective expansion caused by too few positive samples.

[0103] In one embodiment, a causal model is used to distinguish common factors of activities from individual factors of activity types within a limited sample. Through repeated activity analysis and model review, some understanding and evidence regarding the relationship between customer groups and responses have been accumulated; therefore, this is first utilized...

[0104] Prior knowledge is used to model causal relationships. A graphical model approach for causal inference is employed. Causal learning cannot exhaustively cover confounding factors and requires input from business experience and the accumulation of variable relationships through historical analysis. Based on these predefined common factors, instrumental variables, and outcome variables, a causal graph is constructed. In this embodiment, the common factors are those that affect both the intervention (activity) and the outcome (response), such as customer segmentation and activity level. The instrumental variable is the activity push notification, and the outcome variable is whether or not a response is made.

[0105] The established cause-effect graph is as follows Figure 5 As shown. The statistical significance of the above effect estimates was tested using a permutation test. Where W represents the common factor, Z is the instrumental variable, Y is the outcome variable (achievement), X is the effect modification variable, and V is the intervention variable (activity).

[0106] Based on the initial design using a causal graph, the backdoor criterion, frontdoor criterion, and instrumental variable (IV) were tested. The initial sample set was divided into two groups: an intervention group affected by the intervention and a control group not affected by the same intervention. The difference between the intervention and control groups was calculated twice, before and after the intervention, representing the relative relationship between the two groups. A second difference was performed to eliminate the original difference between the intervention and control groups, ultimately yielding the average intervention effect. Priority was given to predicting the intervention effect of whether or not the intervention was performed. Robustness was checked by adding confounding factors, including statistical and behavioral statistics such as mobile banking login frequency and risk assessment status. This process was repeated, adjusting the parameters until the test requirements were met.

[0107] The identified confounding factors are labeled X1, X2, Xn, etc. The common factor types identified by the causal model include categories such as activity level, activity sensitivity, and product preference. As common influencing factors of activities, when the multi-label multi-classification model distinguishes activities, the factor features are downweighted to enhance other factors.

[0108] In one embodiment, such as Figure 6 As shown, the final sample set is input into a multi-label multi-classification model. Based on the first label, second label, third label, common factor label, and individual factor label, the multi-label multi-classification model predicts the target activity type for each sample and pushes activity messages corresponding to the target activity type to each sample, including:

[0109] Step 602: Perform dummy variable transformation on the first label, second label, third label, common factor label and individual factor label of each sample to obtain multiple single-label binary sub-models.

[0110] In some samples, certain feature attributes have parallel relationships and cannot be simply assigned meaning by numerical values ​​or characters. This can be addressed by constructing dummy variables. For example, a feature with three factors can be transformed into a vector with three columns, each consisting only of 0-1 values; such a vector is a dummy variable. The dummy variable transformation process converts a multi-category variable into a binary variable.

[0111] The single-label binary classification sub-model combines pairwise data from K categories, trains a model using the combined data, resulting in K(K-1) / 2 classifiers, merges the results of these classifiers, and outputs the final prediction value using a majority vote. The single-label binary classification sub-model uses the ML-Knn classification algorithm for prediction.

[0112] Optionally, the terminal converts the first label, second label, third label, common factor label, and individual factor label of each sample into binary variables, combines the pairwise data from multiple categories, and then uses the combined data to train multiple single-label binary sub-models.

[0113] Step 604: Use multiple single-label binary classification sub-models to predict the activity type of each sample, classify the activity types predicted by the multiple single-label binary classification sub-models, and take the activity type with the most classification results as the target activity type corresponding to the sample.

[0114] The sample includes user activity data for different activities. Multiple single-label binary classification sub-models are used to predict the activity type of the sample. A majority voting method is used to classify the activity types predicted by the multiple single-label binary classification sub-models, obtaining the number of prediction results for each activity type. The ranking of the prediction results for each activity type determines the ranking reference for the deployment of sub-activities. For example, if sample a user participates in activities A, B, C, and D, the activity types predicted by multiple single-label binary classification sub-models are classified and ranked from highest to lowest, resulting in the ranking of activities C, A, D, and B. Therefore, the target activity type for sample a is activity C, and the activity deployment ranking for sample a is activity C, activity A, activity D, and activity B.

[0115] In this embodiment, the first label, second label, third label, common factor label, and individual factor label of each sample are transformed into dummy variables to obtain multiple single-label binary classification sub-models. Multiple single-label binary classification sub-models are used to predict the activity type of each sample. Based on the prediction results, the target activity type corresponding to the sample is determined. When recommending activities in a large volume based on activity type, the activity message corresponding to the target activity type is recommended first to improve the accuracy of customer acquisition.

[0116] In one embodiment, this embodiment provides the most detailed steps of a message push method, specifically including the following steps:

[0117] Step 1: Obtain the initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating the basic information in the activity data. Samples with the activity participation level meeting the standard are used as positive samples, and samples with the activity participation level not meeting the standard are used as negative samples.

[0118] Step 2: Use the positive samples in the initial sample set as the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, and obtain the trained classifier.

[0119] Step 3: Input each negative sample into the classifier, predict the first probability value of each negative sample being labeled and the second probability value of each positive sample being labeled through the classifier, and calculate the mean probability of the positive sample being labeled.

[0120] Step 4: The ratio of the square of the probability mean to the first probability value is determined as the probability value of a negative sample being labeled as a positive sample.

[0121] Step 5: Take negative samples with probability values ​​greater than preset values ​​as target negative samples that meet the label conversion conditions, and modify the second label of the target negative samples to represent the activity participation level of the target negative samples so as to convert the target negative samples into positive samples.

[0122] Step 6: Add the target negative samples to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process.

[0123] Step 7: Iterate the positive sample expansion process multiple times until the number of positive samples in the final initial positive sample set reaches the required level, then stop the iteration and obtain the intermediate sample set.

[0124] Step 8: Input the intermediate sample set into the pre-trained causal model. The causal model identifies the common factors and individual factors among the samples. Label the basic information containing the common factors in the samples with common factor labels and label the basic information containing the individual factors with individual factor labels to obtain the labeled final sample set. The common factors are the common features among different activities. The individual factors are the individual features that distinguish different types of activities.

[0125] Step 9: Perform dummy variable transformation on the first label, second label, third label, common factor label and individual factor label of each sample to obtain multiple single-label binary sub-models.

[0126] Step 10: Use multiple single-label binary classification sub-models to predict the activity type of each sample, classify the activity types predicted by the multiple single-label binary classification sub-models, and take the activity type with the most classification results as the target activity type corresponding to the sample.

[0127] In this embodiment, a semi-supervised model framework is applied to effectively expand the sample size even with a relatively small number of positive samples (due to unreachable responses), without reducing the return on investment. A causal model identifies common factors across activities, ensuring this expansion while providing feature-based recommendations for subsequent classification and prioritization. A multi-classification model bridges the gap, ranking and differentiating activities in specific scenarios to maximize marketing effectiveness. Based on a business analysis framework, a multi-layered, joint modeling approach reduces the impact of interfering factors, effectively guiding business decisions.

[0128] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0129] Based on the same inventive concept, this application also provides a message push device for implementing the message push method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more message push device embodiments provided below can be found in the limitations of the message push method described above, and will not be repeated here.

[0130] In one embodiment, such as Figure 7 As shown, a message push device is provided, including: an acquisition module 100, a sample expansion module 200, a causal recognition module 300, and a multi-classification module 400, wherein:

[0131] The acquisition module 100 is used to acquire an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating the basic information in the activity data. Samples with the activity participation level meeting the standard are used as positive samples, and samples with the activity participation level not meeting the standard are used as negative samples.

[0132] The sample expansion module 200 is used to convert negative samples that meet the label conversion conditions into positive samples to obtain an intermediate sample set.

[0133] The causal identification module 300 is used to input the intermediate sample set into the pre-trained causal model, identify the common factors and individual factors among the samples through the causal model, and label the basic information containing the common factors in the samples with common factor labels and label the basic information containing the individual factors in the samples with individual factor labels to obtain the labeled final sample set; the common factors are the common features among different activities; the individual factors are the individual features that distinguish different types of activities.

[0134] The multi-classification module 400 is used to input the final sample set into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0135] In one embodiment, the sample expansion module 200 is further configured to use the positive samples in the initial sample set as the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, and obtain a trained classifier.

[0136] Each negative sample in the initial sample set is input into the classifier, and the classifier calculates the probability value of each negative sample being labeled as a positive sample.

[0137] Negative samples with a probability value greater than a preset value are used as target negative samples that meet the label conversion conditions. The second label of the target negative sample is modified to represent that the activity participation of the target negative sample meets the standard, so as to convert the target negative sample into a positive sample.

[0138] Add the target negative sample to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process;

[0139] The positive sample expansion process is iterated multiple times until the number of positive samples in the final initial positive sample set reaches the required level. Then the iteration stops, and an intermediate sample set is obtained.

[0140] In one embodiment, the sample expansion module 200 is further configured to input each negative sample into a classifier, predict a first probability value for each negative sample to be labeled and a second probability value for each positive sample to be labeled through the classifier, and calculate the average probability of positive samples being labeled.

[0141] The ratio of the square of the probability mean to the first probability value is used to determine the probability that a negative sample is labeled as a positive sample.

[0142] In one embodiment, the multi-classification module 400 is also used to perform dummy variable transformation on the first label, second label, third label, common factor label and individual factor label of each sample to obtain multiple single-label binary classification sub-models;

[0143] Multiple single-label binary classification sub-models are used to predict the activity type of each sample. The activity types predicted by the multiple single-label binary classification sub-models are classified, and the activity type with the most classification results is taken as the target activity type corresponding to the sample.

[0144] In one embodiment, the multi-classification module 400 is also used to obtain the activity message to be pushed;

[0145] In the initial sample set, select target samples whose target activity type is the same as the type of the activity message to be pushed, and prioritize pushing the activity message to the target samples.

[0146] In one embodiment, the multi-classification module 400 is further configured to obtain the activity data of the user to be pushed to, and if the activity data of the user to be pushed to belongs to the initial sample set, determine the target activity type corresponding to the user to be pushed to in the initial sample set, and push the activity message corresponding to the target activity type to the user to be pushed to.

[0147] Each module in the aforementioned message push device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0148] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a message push method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0149] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0151] Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0152] Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set;

[0153] The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. Common factors are the common features among different activities; individual factors are the individual features that distinguish different types of activities.

[0154] The final sample set is input into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0155] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0156] The positive samples in the initial sample set are used as the initial positive sample set. Cluster analysis is performed on the label distribution of the initial positive sample set to obtain the trained classifier.

[0157] Each negative sample in the initial sample set is input into the classifier, and the classifier calculates the probability value of each negative sample being labeled as a positive sample.

[0158] Negative samples with a probability value greater than a preset value are used as target negative samples that meet the label conversion conditions. The second label of the target negative sample is modified to represent that the activity participation of the target negative sample meets the standard, so as to convert the target negative sample into a positive sample.

[0159] Add the target negative sample to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process;

[0160] The positive sample expansion process is iterated multiple times until the number of positive samples in the final initial positive sample set reaches the required level. Then the iteration stops, and an intermediate sample set is obtained.

[0161] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0162] Each negative sample is input into the classifier, which predicts the first probability value of each negative sample being labeled and the second probability value of each positive sample being labeled, and calculates the mean probability of the positive sample being labeled.

[0163] The ratio of the square of the probability mean to the first probability value is used to determine the probability that a negative sample is labeled as a positive sample.

[0164] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0165] The first label, second label, third label, common factor label, and individual factor label of each sample are transformed into dummy variables to obtain multiple single-label binary classification sub-models;

[0166] Multiple single-label binary classification sub-models are used to predict the activity type of each sample. The activity types predicted by the multiple single-label binary classification sub-models are classified, and the activity type with the most classification results is taken as the target activity type corresponding to the sample.

[0167] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0168] Get the activity messages to be pushed;

[0169] In the initial sample set, select target samples whose target activity type is the same as the type of the activity message to be pushed, and prioritize pushing the activity message to the target samples.

[0170] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0171] Obtain the activity data of the user to be pushed. If the activity data of the user to be pushed belongs to the initial sample set, determine the target activity type corresponding to the user to be pushed in the initial sample set, and push the activity message corresponding to the target activity type to the user to be pushed.

[0172] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0173] Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0174] Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set;

[0175] The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. Common factors are the common features among different activities; individual factors are the individual features that distinguish different types of activities.

[0176] The final sample set is input into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0177] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: taking the positive samples in the initial sample set as the initial positive sample set, performing cluster analysis on the label distribution of the initial positive sample set, and obtaining a trained classifier;

[0178] Each negative sample in the initial sample set is input into the classifier, and the classifier calculates the probability value of each negative sample being labeled as a positive sample.

[0179] Negative samples with a probability value greater than a preset value are used as target negative samples that meet the label conversion conditions. The second label of the target negative sample is modified to represent that the activity participation of the target negative sample meets the standard, so as to convert the target negative sample into a positive sample.

[0180] Add the target negative sample to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process;

[0181] The positive sample expansion process is iterated multiple times until the number of positive samples in the final initial positive sample set reaches the required level. Then the iteration stops, and an intermediate sample set is obtained.

[0182] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0183] Each negative sample is input into the classifier, which predicts the first probability value of each negative sample being labeled and the second probability value of each positive sample being labeled, and calculates the mean probability of the positive sample being labeled.

[0184] The ratio of the square of the probability mean to the first probability value is used to determine the probability that a negative sample is labeled as a positive sample.

[0185] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0186] The first label, second label, third label, common factor label, and individual factor label of each sample are transformed into dummy variables to obtain multiple single-label binary classification sub-models;

[0187] Multiple single-label binary classification sub-models are used to predict the activity type of each sample. The activity types predicted by the multiple single-label binary classification sub-models are classified, and the activity type with the most classification results is taken as the target activity type corresponding to the sample.

[0188] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0189] Get the activity messages to be pushed;

[0190] In the initial sample set, select target samples whose target activity type is the same as the type of the activity message to be pushed, and prioritize pushing the activity message to the target samples.

[0191] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0192] Obtain the activity data of the user to be pushed. If the activity data of the user to be pushed belongs to the initial sample set, determine the target activity type corresponding to the user to be pushed in the initial sample set, and push the activity message corresponding to the target activity type to the user to be pushed.

[0193] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0194] Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, as well as a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples.

[0195] Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set;

[0196] The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. Common factors are the common features among different activities; individual factors are the individual features that distinguish different types of activities.

[0197] The final sample set is input into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

[0198] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: taking the positive samples in the initial sample set as the initial positive sample set, performing cluster analysis on the label distribution of the initial positive sample set, and obtaining a trained classifier;

[0199] Each negative sample in the initial sample set is input into the classifier, and the classifier calculates the probability value of each negative sample being labeled as a positive sample.

[0200] Negative samples with a probability value greater than a preset value are used as target negative samples that meet the label conversion conditions. The second label of the target negative sample is modified to represent that the activity participation of the target negative sample meets the standard, so as to convert the target negative sample into a positive sample.

[0201] Add the target negative sample to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process;

[0202] The positive sample expansion process is iterated multiple times until the number of positive samples in the final initial positive sample set reaches the required level. Then the iteration stops, and an intermediate sample set is obtained.

[0203] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0204] Each negative sample is input into the classifier, which predicts the first probability value of each negative sample being labeled and the second probability value of each positive sample being labeled, and calculates the mean probability of the positive sample being labeled.

[0205] The ratio of the square of the probability mean to the first probability value is used to determine the probability that a negative sample is labeled as a positive sample.

[0206] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0207] The first label, second label, third label, common factor label, and individual factor label of each sample are transformed into dummy variables to obtain multiple single-label binary classification sub-models;

[0208] Multiple single-label binary classification sub-models are used to predict the activity type of each sample. The activity types predicted by the multiple single-label binary classification sub-models are classified, and the activity type with the most classification results is taken as the target activity type corresponding to the sample.

[0209] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0210] Get the activity messages to be pushed;

[0211] In the initial sample set, select target samples whose target activity type is the same as the type of the activity message to be pushed, and prioritize pushing the activity message to the target samples.

[0212] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0213] Obtain the activity data of the user to be pushed. If the activity data of the user to be pushed belongs to the initial sample set, determine the target activity type corresponding to the user to be pushed in the initial sample set, and push the activity message corresponding to the target activity type to the user to be pushed.

[0214] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0215] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0216] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0217] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A message pushing method, characterized by, The method includes: Obtain an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different types of activities, a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating the basic information in the activity data. Samples that meet the activity participation standard are taken as positive samples, and samples that do not meet the activity participation standard are taken as negative samples. One sample corresponds to one user. The activity data includes basic information about the user's participation history in different types of activities, activity category, customer number, whether clicked or not, and whether the target was met. The basic information is used to distinguish users and different activities, including activity level, activity preference, and activity sensitivity. Activity level is determined by the ratio of the number of times the user participates in an activity to the total number of activities. Activity preference is determined by mathematical statistics to determine whether the user prefers a certain type or several types of activities. Activity engagement refers to a user's response to a particular activity. It is determined by factors such as the duration and frequency of viewing activity information, whether the user clicks on the activity link, whether the user downloads and watches the video, and whether the user registers for membership. The activity engagement is assessed from multiple dimensions, including whether the user clicks on the activity link. We extracted data from several major categories of activities over the past year from the marketing system, including activity type, clicks, and achievement markers; we extracted basic information from the customer relationship system; we processed activity-related data by combining user login information from multiple channels; we conducted data exploration and processing, removing null and outlier values, standardizing the time for seasonal data, and discretizing continuous variables. Negative samples that meet the label conversion conditions are converted into positive samples to obtain an intermediate sample set; The intermediate sample set is input into a pre-trained causal model. The causal model identifies common and individual factors among the samples. Common factor labels are used to label the basic information of samples containing common factors, and individual factor labels are used to label the basic information of samples containing individual factors, thus obtaining the labeled final sample set. The common factors are common features among different activities; the individual factors are individual features that distinguish different types of activities. The final sample set is input into a multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample.

2. The method of claim 1, wherein, The step of converting negative samples that meet the label conversion conditions into positive samples to obtain an intermediate sample set includes: The positive samples in the initial sample set are used as the initial positive sample set. The label distribution of the initial positive sample set is analyzed by clustering to obtain the trained classifier. Each negative sample in the initial sample set is input into the classifier, and the classifier calculates the probability value of each negative sample being labeled as a positive sample. Negative samples with a probability value greater than a preset value are used as target negative samples that meet the label conversion conditions. The second label of the target negative sample is modified to represent that the activity participation of the target negative sample meets the standard, so as to convert the target negative sample into a positive sample. Add the target negative sample to the initial positive sample set, perform cluster analysis on the label distribution of the initial positive sample set, obtain a new classifier, and complete one positive sample expansion process; The positive sample expansion process is iterated multiple times until the number of positive samples in the final initial positive sample set reaches the required level. Then the iteration stops, and an intermediate sample set is obtained.

3. The method of claim 2, wherein, The step of inputting each negative sample in the initial sample set into the classifier and calculating the probability value of each negative sample being labeled as a positive sample by the classifier includes: Each negative sample in the initial sample set is input into the classifier. The classifier predicts a first probability value for each negative sample to be labeled and a second probability value for each positive sample to be labeled. The mean probability of positive samples being labeled is calculated. The ratio of the square of the probability mean to the first probability value is determined as the probability value of a negative sample being labeled as a positive sample.

4. The method of claim 1, wherein, The final sample set is input into a multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type for each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample, including: The first label, second label, third label, common factor label, and individual factor label of each sample are transformed into dummy variables to obtain multiple single-label binary classification sub-models; Multiple single-label binary classification sub-models are used to predict the activity type of each sample. The activity types predicted by the multiple single-label binary classification sub-models are classified, and the activity type with the most classification results is taken as the target activity type corresponding to the sample.

5. The method of claim 1, wherein, The method further includes: Get the activity messages to be pushed; Select target samples in the initial sample set whose target activity type is the same as the type of the activity message to be pushed, and push the activity message to be pushed to the target samples first.

6. The method of claim 1, wherein, The method further includes: Obtain the activity data of the user to be pushed. If the activity data of the user to be pushed belongs to the initial sample set, determine the target activity type corresponding to the user to be pushed in the initial sample set, and push the activity message corresponding to the target activity type to the user to be pushed.

7. A message pushing apparatus characterized by comprising: The device includes: The acquisition module is used to acquire an initial sample set, which includes multiple samples. Each sample includes activity data of users participating in different activity types, a first label indicating the activity type to which the activity data belongs, a second label indicating whether the activity participation level of the sample meets the standard, and multiple third labels indicating various basic information in the activity data. Samples with the activity participation level meeting the standard are designated as positive samples, and samples with the activity participation level not meeting the standard are designated as negative samples. Each sample corresponds to one user, and the activity data includes basic information about the user's historical participation in different activity types, activity category, customer number, whether clicked, and whether the standard was met. The basic information is used to distinguish users and different activities, including activity level, activity preference, and activity sensitivity. Activity level is determined by user participation... The ratio of the number of activities to the total number of activities is determined; activity preferences are determined through mathematical statistics to identify user preferences for certain types or categories of activities; activity participation refers to a user's response to an activity, measured by the duration and frequency of viewing activity information, whether they clicked on the activity link, downloaded and watched the video, and registered as a member; activity participation is assessed from multiple dimensions to determine if the target is met, including whether users clicked on the activity link; data from several major activity categories over the past year is extracted from the marketing system, including activity type, clicks, and target achievement markers; basic information is extracted from the customer relationship system; activity-related data is processed by combining user login information from multiple channels; data exploration and processing are performed, removing null and outlier values, standardizing the time frame for seasonal data, and discretizing continuous variables; The sample expansion module is used to convert negative samples that meet the label conversion conditions into positive samples to obtain an intermediate sample set; The causal identification module is used to input the intermediate sample set into a pre-trained causal model, identify common and individual factors among the samples through the causal model, and label the basic information containing common factors in the samples with common factor labels and label the basic information containing individual factors in the samples with individual factor labels to obtain the labeled final sample set; the common factors are common features among different activities; the individual factors are individual features that distinguish different types of activities; The multi-classification module is used to input the final sample set into the multi-label multi-classification model. The multi-label multi-classification model predicts the target activity type corresponding to each sample based on the first label, second label, third label, common factor label, and individual factor label, and pushes activity messages corresponding to the target activity type to each sample. 8.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-7. When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images