Self-service user group expansion method and device, computer device and storage medium

Abstract

The application relates to a self-service user group expansion method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: obtaining customer labels of a seed population; sorting the customer labels according to importance parameters to obtain sorted features; screening the sorted features according to input business information; training a classification model corresponding to the business information based on the screened sorted features to obtain an expansion model of the seed population; and expanding the user group in the full population through the expansion model to obtain a target group matched with the seed population. The method can sort the characteristic variables of the seed group according to the importance, and then the business personnel can select the characteristic variables and select different algorithm models according to the seed population and the importance of the characteristic variables of the seed population, so that efficient modeling is realized, and then the highly similar population to the seed population is automatically found from the population to be expanded for the business personnel to analyze and expand.

Description

Technical Field

[0001] This application relates to the field of data modeling technology, and in particular to a self-service user group expansion method, apparatus, computer equipment, storage medium, and computer program product. Background Technology

[0002] The user base expansion model is based on the customer base that responded to airline marketing campaigns. By analyzing the characteristics of the responding and non-responding groups of certain historical advertising / marketing campaigns (which we define as the "seed group") and expanding user similarity modeling, it can quickly identify potential customers similar to those who have responded in the seed group. This effectively helps airlines tap into potential customers and expand their business.

[0003] In the traditional user group expansion algorithm model building process, algorithm engineers perform manual feature engineering and customized machine learning algorithm implementation for different marketing scenarios. For any similar user group expansion scenario, from seed user feature analysis and feature extraction, data preprocessing, to machine learning algorithm optimization and model results, algorithm engineers and business personnel need to communicate repeatedly. This results in a long modeling cycle, numerous business scenarios requiring customization of multiple models, and low modeling efficiency for user group expansion. Summary of the Invention

[0004] Therefore, it is necessary to provide a self-service user group expansion method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can efficiently expand the user group, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a self-service user group expansion method. The method includes:

[0006] Obtain customer tags from the seed audience;

[0007] The customer tags are sorted according to the importance parameter to obtain the sorted features;

[0008] The sorted features are then filtered according to the input business information;

[0009] Based on the selected sorted features, the classification model corresponding to the business information is trained to obtain the extended model of the seed population.

[0010] The user group is expanded in the full population through the extended model to obtain the target group that matches the seed population.

[0011] In one embodiment, obtaining the customer tags of the seed population includes:

[0012] Identify the seed audience based on at least one of the following elements: customer groups, marketing activities, and user characteristics;

[0013] The seed population is divided according to the participation information of the marketing activity to obtain an initial positive sample group and an initial negative sample group;

[0014] Determine whether the sample sizes are balanced between the initial positive sample group and the initial negative sample group;

[0015] If balanced, the initial positive sample group and the initial negative sample group are regarded as the positive sample group and negative sample group with difference balance;

[0016] If the sample size is not balanced, at least one group is selected from the initial positive sample group and the initial negative sample group for sampling, based on the sample size, to obtain a positive sample group and a negative sample group with balanced differences.

[0017] Customer tags are obtained based on the positive sample group and the negative sample group.

[0018] In one embodiment, the seed population is a subset of the entire population; the step of selecting at least one population from the initial positive sample population and the initial negative sample population for sampling based on the sample size to obtain a balanced positive sample population and a balanced negative sample population includes:

[0019] When the difference in the number of samples between the initial positive sample group and the full population is within the downsampling interval, the initial positive sample group is downsampled based on the difference to obtain a positive sample group and a negative sample group with balanced differences.

[0020] When the initial negative sample group is missing, random sampling is performed from the full population, and the missing initial negative sample group is filled by the sample group obtained by sampling, so as to obtain the difference-balanced positive sample group and negative sample group.

[0021] In one embodiment, the customer tags include customer tags in a wide table file and customer tags in a mirror file; obtaining the customer tags of the seed population includes:

[0022] Based on the time point and elements corresponding to the audience selection instruction, filter the customer tags in the image file;

[0023] Based on the seed audience corresponding to the customer tags filtered from the image file, select the customer tags from the wide table file;

[0024] The customer tags selected in the wide table file are used to generate the sorted features in the display.

[0025] In one embodiment, the importance parameter includes an information value and an evidence weight; sorting the customer tags according to the importance parameter to obtain the sorted features includes:

[0026] The customer tags are encoded and converted according to evidence weights to obtain the converted feature variables;

[0027] Based on the evidence weights, the information values ​​of the feature variables are calculated;

[0028] The feature variables are sorted according to their information values ​​to obtain sorted features.

[0029] In one embodiment, training the classification model corresponding to the business information based on the selected ranked features to obtain the extended model of the seed population includes:

[0030] Based on the aforementioned business information, obtain at least one model selection parameter from modeling speed, prediction accuracy, and model applicability characteristics;

[0031] Based on the parameters of the model, at least one model is selected from logistic regression model, gradient boosting decision tree model and multilayer perceptron model.

[0032] Based on the sorted features, the selected model is trained for classification to obtain an extended model of the seed population.

[0033] In one embodiment, the step of expanding the user base across the entire population using the extended model to obtain a target group matching the seed population includes:

[0034] The similarity between the entire population and the seed population is calculated using the extended model.

[0035] The entire population is sorted based on the similarity to obtain the sorted population.

[0036] When a filtered or uploaded file is received, the target group is matched with the full population to obtain a similarity score for the target group.

[0037] Select from the population to be expanded according to the similarity score to obtain the target population that matches the seed population.

[0038] Secondly, this application also provides a self-service user group expansion device. The device includes:

[0039] The tag acquisition module is used to acquire customer tags from the seed audience.

[0040] The feature generation module is used to sort the customer tags according to the importance parameter to obtain sorted features;

[0041] The self-service filtering module is used to filter the sorted features according to the input business information;

[0042] The model training module is used to train the classification model corresponding to the business information based on the selected and ranked features, so as to obtain the extended model of the seed population.

[0043] The audience expansion module is used to expand the user group in the full population through the expansion model to obtain the target group that matches the seed audience.

[0044] 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 implement the user group expansion steps in any of the above embodiments.

[0045] 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, implements the user group expansion steps in any of the above embodiments.

[0046] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the user group expansion steps in any of the above embodiments.

[0047] The aforementioned self-service user group expansion method, apparatus, computer equipment, storage medium, and computer program product acquire customer tags of the seed population; sort the customer tags according to importance parameters to obtain sorted features; filter the sorted features according to input business information; and train the classification model corresponding to the business information based on the filtered sorted features. This allows business personnel to independently select feature variables and corresponding algorithm models for training based on their understanding of the business of the seed population to be expanded and the feature importance of the seed population, thereby obtaining an expansion model for the seed population. Based on this expansion model, user group expansion is performed on the entire population to obtain target groups matching the seed population. These similar groups can be tagged for application and analysis by business personnel, saving manpower, time, and management costs, and greatly improving the modeling efficiency of user group expansion. Attached Figure Description

[0048] Figure 1 This is a diagram illustrating the application environment of a self-service user group expansion method in one embodiment.

[0049] Figure 2This is a flowchart illustrating a self-service user group expansion method in one embodiment;

[0050] Figure 3 This is a flowchart illustrating a self-service user group expansion method in one embodiment;

[0051] Figure 4 This is a structural block diagram of a self-service user group expansion device in one embodiment;

[0052] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0053] 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.

[0054] The self-service user group expansion method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Terminal 102 acquires customer tags from the seed population, sorts the customer tags according to importance parameters to obtain sorted features, filters these features based on input business information, trains a classification model corresponding to the filtered sorted features to obtain an expanded model for the seed population, and expands the user group across the entire population using this expanded model to obtain a target group matching the seed population.

[0055] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices that form part of the extended system. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0056] In one embodiment, such as Figure 2 As shown, a self-service user group expansion method is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps:

[0057] Step 202: Obtain customer tags from the seed population.

[0058] The seed audience is a group defined by operational staff, such as business personnel, who are not proficient in programming. This group is obtained through methods such as uploading seed audience files or comprehensively filtering customer tags. When comprehensively filtering customer tags, the system provides business personnel with business-specific scenario-based filtering capabilities, efficiently achieving personalized audience selection. Filtering elements can involve user characteristic data such as customer tags, or business information such as key business nodes within specific business scenarios. For example, filtering could identify a high-frequency group of middle-aged men who cancel tickets on a specific date; or, filtering could identify a high-frequency and active group of young women whose destination is Beijing within the next five days.

[0059] In one embodiment, candidate customer tags are stored in a customer tag management system, employing both persistent data storage and snapshot data storage strategies. The persistent data storage strategy stores candidate customer tags in wide table files partitioned by day, by placing relevant fields within the same wide table, as shown in Table 1.

[0060] Table 1

[0061] Cst_id Age sex Flt_Times_0_365 ... A00000000000462348 23 0 (Male) 2 (2 trips) ... A00000000000462321 35 1 (Female) 3 (3 trips) ... A00000000000462336 46 0 (Male) 0 (0 trips) ...

[0062] Compared to persistent data storage strategies, snapshot data storage strategies obtain mirror files from wide table files at a certain point in time. The mirror files combine the characteristic data of tags with key business nodes from multiple business scenarios. The mirror files use a distributed, multi-user full-text search engine (Elasticsearch) to build an index to achieve millisecond tag response.

[0063] Correspondingly, obtaining customer tags for the seed audience includes: filtering customer tags from the mirror file based on the time point and elements corresponding to the audience selection instruction; selecting customer tags from the wide table file according to the seed audience corresponding to the customer tags filtered from the mirror file; wherein, the customer tags selected from the wide table file are used to generate the sorted features in the display.

[0064] The audience selection command is an instruction entered by business personnel based on their understanding of the business. This command specifies a mirror file at a given time point and the elements to be filtered, and specifies the corresponding elements to filter customer tags in the mirror file. The customer tags filtered from the mirror file are used to determine the corresponding seed audience, and then customer tags in the wide table file are selected based on the seed audience.

[0065] Wide table files reduce the number of table joins, improve the efficiency of iterative calculations during data mining model training, increase query speed and response efficiency, and are easier to use, providing a more intuitive interface for business users. Meanwhile, mirrored files offer business users the ability to filter based on specific scenarios, efficiently enabling personalized audience segmentation and identifying target demographics.

[0066] Specifically, the terminal's extended system can display customer groups saved by the comprehensive filtering module through a drop-down box for a specific group field, allowing users to select seed groups. Alternatively, business personnel can upload a list of users for a marketing campaign or a certain characteristic to the user group extended system. This list typically includes a customer ID (Cst_id) and a positive / negative sample identifier column to indicate whether the customer ID is a positive sample. The customer ID (Cst_id) and the positive / negative sample identifier column can then be used to train a classification-supervised learning model.

[0067] In one embodiment, when some marketing campaigns typically only record the respondent population or the positive and negative sample ratios differ significantly, it is necessary to address the class imbalance problem. Class imbalance refers to a situation in a classification problem where the number of training samples for different classes differs greatly. Correspondingly, the customer tags for obtaining the seed audience include:

[0068] A seed population is obtained based on at least one of the following elements: customer groups, marketing activities, and user characteristics. The seed population is then divided according to participation information from the marketing activities to obtain an initial positive sample group and an initial negative sample group. It is determined whether the sample sizes of the initial positive and negative sample groups are balanced. If balanced, the initial positive and negative sample groups are used as balanced positive and negative sample groups. If unbalanced, based on the sample size, at least one group is selected from the initial positive and negative sample groups for sampling to obtain balanced positive and negative sample groups. Customer tags are obtained based on the positive and negative sample groups.

[0069] In one embodiment, based on the sample size, at least one population is selected from the initial positive sample population and the initial negative sample population for sampling to obtain a positive sample population and a negative sample population with balanced differences, including:

[0070] When the difference in sample size between the initial positive sample group and the full population is within the downsampling interval, the initial positive sample group is downsampled based on the difference value to obtain a positive sample group and a negative sample group with balanced differences.

[0071] Specifically, the proportion of the initial positive sample group in the entire population is the difference in sample size. When the difference is less than n% or greater than m% in the downsampling interval, the initial positive sample group is downsampled based on this difference, randomly deleting some samples from the majority class. During downsampling, the difference between the number of positive and negative examples is controlled to be reduced to N times, while ensuring that the number of minority class samples in both the positive and negative sample groups is not less than the corresponding threshold, thus obtaining a balanced positive and negative sample group. The balanced positive and negative sample groups are used to preserve the complete features of the original seed population as much as possible, and to avoid training bias caused by class imbalance in the model.

[0072] When the initial negative sample population is missing, random sampling is performed from the full population, and the sample population obtained from the sampling fills the missing initial negative sample population, resulting in a balanced positive and negative sample population.

[0073] When a marketing campaign only records the initial positive sample group of the participants, and the number of the initial positive sample group is not less than the corresponding threshold, the negative sample group that did not participate in the marketing campaign is not recorded. In this scenario, a random sampling method is used to directly randomly select a number of sample data equal to the initial positive sample group from the full user tag table as the initial negative sample group of the seed population. This method is simple and fast and is suitable for situations where the sample user dataset is relatively large.

[0074] Step 204: Sort the customer tags according to the importance parameter to obtain the sorted features.

[0075] The reason for sorting customer tags by importance is that there are hundreds of candidate customer tags, and the number is constantly increasing. Too many tags or tags of poor quality will affect the accuracy of the model. Therefore, it is often necessary to select meaningful tag features for model training.

[0076] In one embodiment, the importance parameter includes information value and evidence weight; sorting customer tags according to the importance parameter to obtain sorted features includes: encoding and transforming customer tags according to evidence weight to obtain transformed feature variables, calculating the information value of customer tags based on evidence weight, and sorting the feature variables according to the information value to obtain sorted features.

[0077] The weight of evidence (WOE) is the difference between the proportion of positive samples in a given feature interval and the proportion of negative samples in a given feature interval. It is calculated by encoding the original feature variables using methods such as binning. Since the WOE values ​​for each feature are similar in magnitude, the model needs to handle fewer independent variables, reducing modeling complexity. This is especially beneficial when the model is sensitive to feature values, leading to higher accuracy.

[0078] Specifically, the weight of evidence (WOE) has several advantages. Firstly, it reflects the contribution of independent variables because the fluctuations in the WEE within each independent variable are combined with the coefficients fitted by the model. These coefficients are positively correlated with the variance of the WEE and the contribution rate of each independent variable. Secondly, the WEE standardizes features and improves the model's generalization ability. This is because feature binning is performed during encoding, reducing the number of feature values ​​and allowing direct comparisons between values ​​within each independent variable. Furthermore, WEE can also be used to directly compare values ​​between different independent variables. In addition, the WEE is insensitive to outliers, allowing extreme values ​​to be transformed into non-outliers through WEE.

[0079] Information Value (IV) is a weighted sum of evidence weights. It represents the degree to which a feature contributes to similarity and is positively correlated with the probability of identifying highly similar individuals for that feature variable.

[0080] Furthermore, in actual production environments, in any grouping interval of feature variables, when the extreme case of initial positive sample count = 0 or initial negative sample count = 0 occurs, the value of evidence weight being positive or negative infinity is meaningless. The information value cannot automatically handle the situation where the proportion of initial positive samples in the variable group is 0 or 100%. When such extreme cases occur, this extended system automatically adjusts the number of initial positive samples and initial negative samples in the group: adjusting the original number of initial positive samples from 0 to 1; adjusting the original number of initial negative samples from 0 to 1.

[0081] Step 206: Filter the sorted features according to the input business information.

[0082] Business information is generated based on the business understanding of business personnel. This information is input by business personnel, and can be entered through methods such as selecting and inputting from sorted features displayed in the extended system interface, or through language or other media. By inputting business information to filter business tags, business personnel directly access the customer tag library in the tag management module. Through the diverse tag features in the customer tag library, rich analytical materials are provided for precise marketing, reducing the frequency of communication between business personnel and algorithm engineers. This allows for the filtering of sorted features according to the input business information to select feature variables that meet business needs.

[0083] In one embodiment, the sorted features are filtered according to the input business information. Specifically, the extended system sorts the feature variables according to their information values ​​to obtain the sorted features, which are then dispatched to the front end for display to business personnel. Business personnel use the quick information value range selection and manual feature variable selection methods provided by the extended system to select at least two labels as two features for training. The quick information value range selection method is based on quickly selecting feature variables within a range of information values. At the same time, business personnel can provide customer labels for training the model based on their understanding of the seed population's business and the information values ​​of the features.

[0084] Step 208: Based on the selected ranked features, train the classification model corresponding to the business information to obtain the extended model of the seed population.

[0085] In one embodiment, the classification model corresponding to the business information can be any type of model, or multiple models. When multiple models exist, the model with the highest similarity is selected based on its performance on the dataset. The multiple models can be at least one of learned logistic regression, gradient boosting decision tree model, and multilayer perceptron model.

[0086] In one embodiment, based on the selected ranked features, a classification model corresponding to the business information is trained to obtain an extended model for the seed population. This includes: obtaining at least one model selection parameter among modeling speed, prediction accuracy, and model applicability features according to the business information; selecting at least one model from logistic regression, gradient boosting decision tree, and multilayer perceptron models according to the model selection parameters; and performing classification training on the selected model based on the selected ranked features to obtain an extended model for the seed population.

[0087] In one embodiment, the process of selecting a model based on model selection parameters is described. When modeling speed is used as the model selection parameter, a logistic regression model can be selected; when prediction accuracy is used as the model selection parameter, a gradient boosting decision tree (GBDT) model or a multilayer perceptron (MLP) model can be selected; when the applicable features of the model are high-dimensional sparse features, a logistic regression model or a multilayer perceptron model can be selected.

[0088] After understanding the selection process, the advantages of the candidate models are discussed in turn. Logistic regression is a classic generalized linear model, while our problem belongs to the binomial logistic regression model. The output value falls between 0 and 1 and has probabilistic significance. The model parameters represent the influence of each feature on the output, making it highly interpretable, simple to implement, very efficient, and suitable for large amounts of data.

[0089] Gradient boosting decision tree models are additive models that use a linear combination of basis functions to classify data by continuously reducing the residuals generated during training. During classification, the terminal performs multiple iterations, each generating a weak classifier. Each weak classifier is trained based on the residuals of the previous classifier. Weak classifiers are typically chosen as classification regression trees (CART TREEs). Due to the requirements of high bias and simplicity, the depth of each classification regression tree is not very deep. The final overall classifier is obtained by weighted summation of the weak classifiers obtained in each training round. For binary classification prediction problems in marketing, the classification model trained by gradient boosting decision tree models has good predictive performance. In this case, basic customer attributes, travel attributes, additional service attributes, and value attributes are usually used as features to train and predict the dependent variable. The accuracy of the prediction mainly depends on the sufficiency of feature variable extraction.

[0090] The Multilayer Perceptron (MLP) model is a feedforward-structured artificial neural network classifier. MLPs can handle non-linearly separable problems. An MLP model is a directed graph that maps a set of input vectors to a set of output vectors. The directed graph consists of multiple nodes, with each layer's nodes fully connected to the next layer. Besides the input nodes, each node has a neuron with a non-linear activation function to fit the non-linear model. It is trained using a supervised learning method with backpropagation (BP) to perform continuous mappings across different dimensions. Classification models trained with MLPs possess high parallel processing capabilities, non-linear global effects, good fault tolerance, and strong self-learning ability, thus exhibiting excellent predictive power.

[0091] Optionally, before classifying and training the selected model based on the sorted features, the method further includes a preprocessing process, wherein: the sorted features selected from the seed population are used as training data, the training data is preprocessed to obtain vectorized training data, the vectorized training data is randomly sampled, 70% of the vectorized training data is used as the training set, and the remaining 30% of the vectorized training data is used as the test set; wherein the training set is used for model training, and the test set is used for model evaluation.

[0092] Based on the selected ranked features, the chosen model is trained for classification, including: encoding the selected model, importing the encoded model into a multi-fold cross-validation training set for modeling and training; fine-tuning the trained model in a specified parameter space using a grid search method to select the optimal model parameters, resulting in the trained classification model. After the model parameters are fitted, a model score is generated on the test set for model evaluation. The system uses model evaluation metrics such as confusion matrix, accuracy, precision, recall, and AUC value for visualization.

[0093] Optionally, in this system, the threshold for classifying based on the model's output probability is set as an adaptive parameter: threshold = number of positive training samples / number of negative training samples + number of positive training samples. In this case, the model will be more sensitive to fewer classes. Multiple evaluation metrics are visualized based on the threshold shift parameter to assess the model's performance.

[0094] Step 210: Expand the user group in the full population by extending the model to obtain the target group that matches the seed population.

[0095] In one embodiment, expanding the user group in the full population using an extended model to obtain a target group matching the seed population includes: calculating the similarity between the full population and the seed group using the extended model; sorting the full population based on the similarity to obtain a sorted full population; when a filtered or uploaded group to be expanded is received, associating and matching the group to be expanded with the full population to obtain a similarity score for the group to be expanded; and selecting from the group to be expanded according to the similarity score to obtain a target group matching the seed population.

[0096] Specifically, the similarity score between the full population and the seed population is calculated and stored as basic data. When a group to be expanded is received through comprehensive filtering or file upload, the similarity score of the group to be expanded is matched with the full population. From the group to be expanded, the target groups with high similarity scores (Top 1%, Top 10%, ..., Top 100%) selected by the user on the front end are combined, and custom tags are assigned to the target groups. These tags are then imported into the tag management module.

[0097] In the aforementioned self-service user group expansion method, customer tags of the seed population are obtained; these tags are sorted according to importance parameters to obtain sorted features; the sorted features are then filtered according to the input business information; based on the filtered sorted features, a classification model corresponding to the business information is trained. This allows business personnel to independently select feature variables and corresponding algorithm models for training based on their understanding of the business of the seed population to be expanded and the feature importance of that seed population, thus obtaining an expansion model for the seed population. Based on this expansion model, user group expansion is performed on the entire population to obtain target groups matching the seed population, and these similar groups can be tagged for application and analysis by business personnel. Therefore, the expansion system can trigger the completion of the entire closed-loop management from seed population import, population feature filtering, population business feature filtering, acquisition of highly similar groups, evaluation of highly familiar groups, and group tagging, greatly improving the modeling efficiency of user group expansion.

[0098] In one embodiment, such as Figure 3 As shown, the application of this solution is discussed in general. The solution includes: uploading seed users according to the file upload or customer comprehensive screening module; determining the passenger tag wide table of the passenger tag association module according to the uploaded seed users; obtaining feature variables from the passenger tag wide table; sorting the obtained feature variables from high to low according to information value (IV value) and visualizing them; the displayed feature variables are screened by business personnel; the screened features are encoded and transformed according to weighted evidence; the transformed features are substituted into the classification model selected by business personnel for training; based on the trained classification model, the similarity scores of each user group relative to the seed users in the full passenger pool are calculated; the similarity scores are sorted in descending order and stored; when obtaining the target group matching the seed users in the model application, a preset proportion of high similarity user IDs are selected from the group to be expanded; the high similarity user IDs are labeled to obtain the expanded group.

[0099] In this embodiment, business personnel can independently select feature variables and choose a model for training based on the seed population they need to expand. The trained model automatically identifies highly similar individuals from the entire user base and labels them. Business personnel do not need to be experts in algorithms; the system can complete the entire closed loop from seed population import and feature selection to highly similar individual acquisition and labeling within a day, greatly improving the modeling efficiency of user base expansion. This solves the problems of traditional user base expansion models, which involve long development cycles, high communication and management costs, and the need for hundreds or thousands of customized algorithm models for various business scenarios.

[0100] 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.

[0101] Based on the same inventive concept, this application also provides a self-service user group expansion device for implementing the self-service user group expansion 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 embodiments of the self-service user group expansion device provided below can be found in the limitations of the self-service user group expansion method described above, and will not be repeated here.

[0102] In one embodiment, such as Figure 4 As shown, a self-service user group expansion device is provided, including: a tag acquisition module 402, a feature generation module 404, a self-service screening module 406, a model training module 408, and a user group expansion module 410, wherein:

[0103] Tag acquisition module 402 is used to acquire customer tags of the seed audience;

[0104] The feature generation module 404 is used to sort the customer tags according to the importance parameter to obtain sorted features;

[0105] The self-service filtering module 406 is used to filter the sorted features according to the input business information;

[0106] The model training module 408 is used to train the classification model corresponding to the business information based on the selected sorted features to obtain the extended model of the seed population.

[0107] The audience expansion module 410 is used to expand the user group in the full population through the expansion model to obtain the target group that matches the seed audience.

[0108] In one embodiment, the tag acquisition module 402 includes:

[0109] The seed audience acquisition unit is used to acquire seed audiences based on at least one of the following elements: customer groups, marketing activities, and user characteristics.

[0110] The sample conversion unit is used to divide the seed population according to the participation information of the marketing activity to obtain an initial positive sample group and an initial negative sample group;

[0111] A quantity balance judgment unit is used to determine whether the sample quantity is balanced between the initial positive sample group and the initial negative sample group;

[0112] The first sample balancing processing unit is used to, if balanced, treat the initial positive sample group and the initial negative sample group as the difference-balanced positive sample group and negative sample group.

[0113] The second sample balancing processing unit is used to, if unbalanced, select at least one group from the initial positive sample group and the initial negative sample group for sampling based on the sample quantity, so as to obtain a positive sample group and a negative sample group with balanced differences.

[0114] The tag acquisition unit is used to acquire customer tags based on the positive sample group and the negative sample group.

[0115] In one embodiment, the seed population is a subset of the entire population; the tag acquisition unit includes:

[0116] The downsampling subunit is used to downsample the initial positive sample group based on the difference value when the difference in the number of samples between the initial positive sample group and the full population is within the downsampling interval, so as to obtain a positive sample group and a negative sample group with a balanced difference.

[0117] The sample completion subunit is used to randomly sample from the full population when the initial negative sample group is missing, and fill the missing initial negative sample group with the sampled sample group to obtain the difference-balanced positive sample group and negative sample group.

[0118] In one embodiment, the customer tags include customer tags in a wide table file and customer tags in a mirror file; the tag acquisition module 402 includes:

[0119] The seed audience acquisition unit is used to filter customer tags in the image file based on the time point and elements corresponding to the audience selection instruction.

[0120] The tag acquisition unit is used to select customer tags from the wide table file according to the seed population corresponding to the customer tags filtered in the image file.

[0121] The customer tags selected in the wide table file are used to generate the sorted features in the display.

[0122] In one embodiment, the importance parameter includes an information value and an evidence weight; the feature generation module 404 includes:

[0123] The encoding unit is used to encode and transform the customer tags according to evidence weights to obtain the transformed feature variables;

[0124] An information value calculation unit is used to calculate the information value of the feature variable based on the evidence weight;

[0125] The sorting unit is used to sort the feature variables according to the information values ​​to obtain sorted features.

[0126] In one embodiment, the model training module 408 includes:

[0127] The model cognition unit is used to obtain at least one model selection parameter from modeling speed, prediction accuracy, and model applicability features according to the business information.

[0128] The model selection unit is used to select at least one model from logistic regression model, gradient boosting decision tree model and multilayer perceptron model according to the model selection parameters.

[0129] The model training unit is used to perform classification training on the selected model based on the sorted features to obtain the extended model of the seed population.

[0130] In one embodiment, the crowd expansion module 410 includes:

[0131] The similarity calculation unit is used to calculate the similarity between the entire population and the seed population through the extended model;

[0132] A similarity ranking unit is used to rank the entire population based on the similarity to obtain the ranked entire population.

[0133] The similarity score calculation unit is used to match the target group with the full population when a filtered or uploaded file is received, and to obtain the similarity score of the target group.

[0134] The target population selection unit is used to select from the population to be expanded according to the similarity score to obtain a target population that matches the seed population.

[0135] Each module in the aforementioned self-service user group expansion device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0136] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As 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 self-service user group expansion method. The display unit of the computer device is used to form a visually visible image. It 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.

[0137] Those skilled in the art will understand that Figure 5 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.

[0138] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0139] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0140] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0141] 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.

[0142] 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.

[0143] 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.

[0144] 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 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 self-service user group expansion method, characterized in that, The method includes: Identify the seed audience based on at least one of the following elements: customer groups, marketing activities, and user characteristics; Based on the participation information of the seed population in the marketing campaign, an initial positive sample group and an initial negative sample group are obtained; The difference value is determined based on the proportion of the initial positive sample group in the full population. The initial positive sample group and the initial negative sample group are then balanced to obtain the positive sample group and the negative sample group used for model training. Based on the positive sample group and the negative sample group, customer tags are obtained; The customer tags are sorted according to the importance parameter to obtain the sorted features; Display the sorted features and obtain business information based on the displayed sorted features; The sorted features are then filtered according to the business information; Based on the selected sorted features, the classification model corresponding to the business information is trained to obtain the extended model of the seed population; the classification threshold of the classification model is determined based on the ratio of the number of positive sample groups to the number of negative sample groups. The user group is expanded in the full population through the extended model to obtain the target group that matches the seed population.

2. The method according to claim 1, characterized in that, The seed population is a subset of the full population; the process of determining the difference value based on the proportion of the initial positive sample population in the full population, and balancing the initial positive sample population and the initial negative sample population to obtain the positive sample population and negative sample population used for model training includes: If the difference in sample size between the initial positive sample group and the full population is greater than or equal to n% and less than or equal to m%, the initial positive sample group and the initial negative sample group are regarded as a positive sample group and a negative sample group with difference balance. If the difference value is less than n% or greater than m%, downsampling is performed on the largest sample group among the initial positive sample group and the initial negative sample group based on the difference value until the difference between the proportion of the initial positive sample group and the initial negative sample group is reduced to within N times, thus obtaining a positive sample group and a negative sample group with balanced differences. When the initial negative sample group is missing, random sampling is performed from the full population, and the missing initial negative sample group is filled by the sample group obtained by sampling, so as to obtain the difference-balanced positive sample group and negative sample group.

3. The method according to claim 1, characterized in that, The customer tags include customer tags in the wide table file and customer tags in the mirror file; the customer tags for obtaining the seed population include: Based on the time point and elements corresponding to the audience selection instruction, filter the customer tags in the image file; Based on the seed audience corresponding to the customer tags filtered from the image file, select the customer tags from the wide table file; The customer tags selected in the wide table file are used to generate the sorted features in the display.

4. The method according to claim 1, characterized in that, The importance parameters include information value and evidence weight; The step of sorting the customer tags according to an importance parameter to obtain sorted features includes: The data obtained from binning the customer tags is encoded and transformed according to the evidence weight to obtain the transformed feature variables. When calculating the evidence weight value, if the number of positive or negative samples in the bin is zero, the number of zero samples is adjusted to a preset non-zero value. The variance of the evidence weight value is positively correlated with the coefficients fitted by the model containing each independent variable, and the coefficients are positively correlated with the contribution rate of each independent variable. Based on the evidence weights, the information values ​​of the feature variables are calculated; The feature variables are sorted according to their information values ​​to obtain sorted features.

5. The method according to claim 1, characterized in that, The step of training a classification model corresponding to the business information based on the selected ranked features to obtain an extended model for the seed population includes: Based on the aforementioned business information, obtain at least one model selection parameter from modeling speed, prediction accuracy, and model applicability characteristics; Based on the parameters of the model, at least one model is selected from logistic regression model, gradient boosting decision tree model and multilayer perceptron model. Based on the sorted features, the selected model is trained for classification to obtain an extended model of the seed population.

6. The method according to claim 1, characterized in that, The step of expanding the user base across the entire population using the extended model to obtain a target group matching the seed population includes: The similarity between the full population and the seed population is calculated using the extended model. The entire population is sorted based on the similarity to obtain the sorted population. When a filtered or uploaded file is received, the target group is matched with the full population to obtain a similarity score for the target group. Select from the population to be expanded according to the similarity score to obtain the target population that matches the seed population.

7. The method according to claim 1, characterized in that, The expression for the classification threshold is: Classification threshold = Number of positive samples / (Number of negative samples + Number of positive samples).

8. A self-service user group expansion device, characterized in that, The device includes: The tag acquisition module is used to acquire a seed population based on at least one of the following elements: customer groups, marketing activities, and user characteristics; divide the seed population into an initial positive sample group and an initial negative sample group based on their participation information in the marketing activities; determine the difference value based on the proportion of the initial positive sample group in the full population; balance the initial positive sample group and the initial negative sample group to obtain a positive sample group and a negative sample group for model training; and acquire customer tags based on the positive sample group and the negative sample group. The feature generation module is used to sort the customer tags according to the importance parameter to obtain sorted features; A self-service filtering module is used to display the sorted features and obtain business information based on the displayed sorted features; and to filter the sorted features according to the business information. The model training module is used to train the classification model corresponding to the business information based on the selected and ranked features to obtain the extended model of the seed population; the classification threshold of the classification model is determined based on the ratio of the number of positive sample groups to the number of negative sample groups. The audience expansion module is used to expand the user group in the full population through the expansion model to obtain the target group that matches the seed audience.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, 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 7.

Images