Method and apparatus for determining tag weights

Abstract

The present application discloses a method and device for determining label weights, and relates to the field of computer technology. The method comprises: obtaining a portrait data set of a target population, wherein the portrait data set includes at least one original label of the target population and sample users constituting the target population; determining the weight of the original label according to the number of users in the sample users that meet the original label and the total number of sample users. The method can be used to determine the importance of the portrait label of the target population.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method and device for determining tag weights. Background Art

[0002] Crowd portrait analysis can be used to assist in commercial promotion or the development of smart products. Currently, there are many methods for analyzing crowd portrait labels. However, how to determine the accuracy or importance of crowd portrait labels in representing crowd portraits has become an urgent problem to be solved. Summary of the invention

[0003] The present disclosure provides a method, an apparatus, an electronic device, and a computer-readable storage medium for determining a tag weight.

[0004] According to a first aspect of the present disclosure, a method for determining a label weight is provided, comprising: obtaining a portrait data set of a target population, the portrait data set including at least one original label of the target population and sample users constituting the target population; determining the weight of the original label based on the number of users among the sample users who meet the original label and the total number of sample users.

[0005] In some embodiments, the weight of the original label is determined based on the number of users in the sample users who meet the original label and the total number of sample users, including: marking each original label in the portrait data set as a first label, and determining the weight of the original label through multiple rounds of iterative operations; the iterative operation includes: randomly obtaining the original label with the first label in the portrait data set, and constructing a copy portrait data set, the copy portrait data set including the copy label; using the number of users in the sample users who meet the original label and the number of users in the sample users who do not meet the original label to determine the first weight of the original label; using the number of users in the sample users who meet the copy label and the number of users in the sample users who do not meet the copy label to determine the second weight of the copy label, and determining the largest second weight among the second weights of each copy label as the benchmark weight; in response to determining that the first weight is less than the benchmark weight, before executing the next iterative operation, updating the label of the original label corresponding to the first weight to the second label; in response to determining that the iterative operation meets the stopping condition, stopping the iterative operation, and determining the weight of the original label corresponding to the first weight based on the probability that the first weight is greater than or equal to each benchmark weight determined in multiple rounds of iterative operations.

[0006] In some embodiments, in each round of the iterative operation, the first weight of the original label is determined, including: using each model of a plurality of target models to determine the first weight of the original label output by the model; the second weight of the copy label is determined, including: determining the second weight of the copy label output by the model; in response to determining that the first weight is greater than or equal to the benchmark weight, determining that the output result of the model is the first result; in response to determining that the first weight is less than the benchmark weight, before executing the next iterative operation, updating the mark of the original label corresponding to the first weight to the second mark, including: in response to determining that the probability that multiple target models output the first result and all output results does not meet the first preset condition, before executing the next iterative operation, updating the mark of the original label corresponding to the first weight to the second mark.

[0007] In some embodiments, the first preset condition is not met, including: the probability distribution function of the output probability of multiple target models outputting the first result and accounting for all output results does not meet the second preset condition.

[0008] In some embodiments, before executing the next iterative operation, the mark of the original label corresponding to the first weight is updated to a second mark, including: obtaining the probability distribution function of the output probability determined in the current iterative operation and the iterative operation before the current iterative operation; using the obtained multiple probability distribution functions to perform error discovery rate correction; in response to determining that the corrected multiple probability distribution functions do not meet the third preset condition, before executing the next iterative operation, the mark of the original label is updated to the second mark.

[0009] In some embodiments, the method further includes: after the iterative operation is completed, based on the probability distribution function of the output probabilities of multiple target models determined in the last round of iterative operation outputting the first result and accounting for all output results, determining the weight index of the original label corresponding to the first weight.

[0010] In some embodiments, the use of each of the multiple target models, which outputs a first weight of the original label, includes: using each decision tree in the random forest model, and determining the first weight based on the number of times the node in the decision tree represented by the label is determined as a split point, wherein the split point is used to divide the sample users into different subsets; and determining the second weight of the copy label output by the model, including: determining the second weight of each copy label based on the number of times the node in the decision tree represented by each copy label is determined as a split point.

[0011] In some embodiments, the method includes: obtaining a comparison portrait data set, the comparison portrait data set including at least one comparison label of a comparison population and comparison users constituting the comparison population, the comparison population being a population having opposite behaviors to the target population; determining a weight of the comparison label based on the number of comparison users that meet the comparison label and the total number of comparison users; in response to determining that the original label is the same as the comparison label, determining whether the label is a label of the target population based on the weight of the original label corresponding to the same label and the weight of the comparison label corresponding to the label.

[0012] According to a second aspect of the present disclosure, a device for determining a label weight is provided, comprising: an acquisition unit, configured to acquire a portrait data set of a target population, the portrait data set including at least one original label of the target population and sample users constituting the target population; a determination unit, configured to determine the weight of the original label based on the number of users in the sample users who meet the original label and the total number of sample users.

[0013] In some embodiments, the determination unit includes: a marking module, configured to mark each original label in the portrait data set as a first label, and determine the weight of the original label through multiple rounds of iterative operations; the iterative operation includes: a first acquisition module, configured to randomly obtain the original label with the first label in the portrait data set, and construct a copy portrait data set, the copy portrait data set including the copy label; a first determination module, configured to determine the first weight of the original label by using the number of users in the sample users who meet the original label and the number of users in the sample users who do not meet the original label; a second determination module, configured to determine the second weight of the copy label by using the number of users in the sample users who meet the copy label and the number of users in the sample users who do not meet the copy label, and determine the largest second weight among the second weights of each copy label as the benchmark weight; an update module, configured to update the label of the original label corresponding to the first weight to the second label before performing the next iterative operation in response to determining that the first weight is less than the benchmark weight; a third determination module, configured to stop the iterative operation in response to determining that the iterative operation meets the stop condition, and determine the weight of the original label corresponding to the first weight based on the probability that the first weight is greater than or equal to each benchmark weight determined in multiple rounds of iterative operations.

[0014] In some embodiments, in each round of iterative operation, the first determination module includes: a first determination submodule configured to use each model in a plurality of target models to determine the first weight of the original label output by the model; the second determination module includes: a second determination submodule configured to determine the second weight of the copy label output by the model; the output module is configured to determine that the output result of the model is the first result in response to determining that the first weight is greater than or equal to the reference weight; the update module includes: a first update submodule configured to update the mark of the original label corresponding to the first weight to the second mark in response to determining that the probability of the output of the plurality of target models as the first result and accounting for all the output results does not meet the first preset condition before performing the next iterative operation. In some embodiments, not meeting the first preset condition includes: the probability distribution function of the output probability of the plurality of target models as the first result and accounting for all the output results does not meet the second preset condition.

[0015] In some embodiments, the first update submodule includes: a second acquisition module, configured to acquire the probability distribution function of the output probability determined in the current iteration operation and the iteration operation before the current iteration operation; a correction module, configured to use the acquired multiple probability distribution functions to perform error discovery rate correction; and a second update submodule, configured to update the mark of the original label to the second mark before executing the next iteration operation in response to determining that the corrected multiple probability distribution functions do not meet the third preset condition.

[0016] In some embodiments, the device also includes: a weight index determination module, which is configured to determine the weight index of the original label corresponding to the first weight based on the probability distribution function of the output probability of multiple target models determined in the last round of iterative operations being the first result and accounting for all output results after the iterative operation is completed.

[0017] In some embodiments, a first determination submodule includes: a first classification module, configured to use each decision tree in a random forest model to determine a first weight based on the number of times a node in the decision tree represented by the original label is determined as a splitting point, wherein the splitting point is used to divide the sample users into different subsets; and a second determination submodule includes: a second classification module, configured to determine a second weight for each copy label based on the number of times a node in the decision tree represented by each copy label is determined as a splitting point, and determine the largest second weight among the second weights as a benchmark weight.

[0018] In some embodiments, the device includes: a comparison data acquisition unit, configured to acquire a comparison portrait data set, the comparison portrait data set including at least one comparison label of a comparison population and comparison users constituting the comparison population, the comparison population being a population having the opposite behavior to the target population; a comparison data determination unit, configured to determine the weight of the comparison label based on the number of comparison users that meet the comparison label and the total number of comparison users; a comparison unit, configured to determine whether the label is the label of the target population in response to determining that the original label is the same as the comparison label, based on the weight of the original label corresponding to the same label and the weight of the comparison label corresponding to the label.

[0019] According to the third aspect of the present disclosure, an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining label weights as provided in the first aspect.

[0020] According to a fourth aspect of the present disclosure, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method for determining a label weight provided in the first aspect is implemented.

[0021] The method and device for determining label weights provided by the present disclosure obtain a portrait data set of a target population, the portrait data set includes at least one original label of the target population, sample users constituting the target population, and determine the weight of the original label according to the number of users in the sample users that meet the original label and the total number of sample users. The accuracy or importance of the portrait label of the target population can be determined.

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present application.

[0024] Figure 1 is an exemplary system architecture diagram to which embodiments of the present application may be applied;

[0025] Figure 2 is a flow chart of an embodiment of a method for determining tag weights according to the present application;

[0026] Figure 3 is a flowchart of another embodiment of a method for determining a tag weight according to the present application;

[0027] Figure 4 is a structural schematic diagram of another embodiment of a method for determining a tag weight according to the present application;

[0028] Figure 5 is a structural schematic diagram of an embodiment of a device for determining a tag weight according to the present application;

[0029] Figure 6 It is a block diagram of an electronic device used to implement the method for determining tag weights in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0031] Figure 1 An exemplary system architecture 100 is shown to which an embodiment of a method for determining a tag weight or an apparatus for determining a tag weight of the present application can be applied.

[0032] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0033] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various client applications can be installed on terminal devices 101, 102, 103, such as navigation applications, recommendation applications, chat applications, shopping applications, financial applications, image applications, video applications, etc.

[0034] The terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting receiving server messages, including but not limited to smart phones, tablet computers, e-book readers, electronic players, portable computers, desktop computers, and the like.

[0035] Terminal devices 101, 102, 103 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices. When terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software or software modules (for example, multiple software modules for providing distributed services), or they can be implemented as a single software or software module. No specific limitation is made here.

[0036] Terminal devices 101, 102, and 103 can obtain a portrait data set of the target population, which includes at least one original label of the target population and sample users constituting the target population; the weight of the original label is determined based on the number of sample users that meet the original label and the total number of sample users.

[0037] It should be noted that the method for determining the tag weight provided in the embodiments of the present disclosure is generally executed by the server 105 , and accordingly, the device for determining the tag weight is generally arranged in the server 105 .

[0038] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0039] Continue to refer Figure 2 , shows a process 200 of an embodiment of a method for determining a tag weight according to the present disclosure, comprising the following steps:

[0040] Step 201, obtaining a portrait data set of a target population, wherein the portrait data set includes at least one original label of the target population and sample users constituting the target population.

[0041] In this embodiment, the execution subject of the method for determining the tag weight (eg Figure 1 The device 105 shown in the figure can obtain a portrait data set of the target population by wired or wireless means, and the portrait data set includes at least one original label of the target population and sample users constituting the target population. For example, the target population can be users who browse a shopping website and place an order within a preset time period, and the original label of the target population can be an age label, a gender label, and other population attribute / feature labels, and the sample users constituting the target population are the users who place an order. For another example, the target population can be users who purchase a certain product, and the original label of the target population can be between 20 and 30 years old, female, etc., and the sample users constituting the target population are the users who purchase the product.

[0042] Step 202: Determine the weight of the original label according to the number of sample users that match the original label and the total number of sample users.

[0043] In this embodiment, for each original label of at least one original label of the target population, the weight of the original label can be determined according to the number of users in the sample users who meet the original label and the total number of sample users. For example, if the original label of the target population is male, the total number of sample users constituting the target population is 10, of which 3 are female and 7 are male, then the weight of the original label can be determined to be 7 / 10=70%. It can be understood that the original label can be a label of a target population with a certain behavior determined according to a neural network or other method for determining a user behavior profile, and among all the sample users constituting the target population, sample users of male, female, and unknown gender are actually included. Therefore, even if the label of the target population is male, each sample user constituting the target population is not necessarily male.

[0044] The method for determining label weights provided in this embodiment obtains a portrait data set of a target population, the portrait data set includes at least one original label of the target population, sample users constituting the target population, and determines the weight of the original label according to the number of sample users that match the original label and the total number of sample users. The accuracy or importance of the portrait label of the target population can be determined.

[0045] Further references Figure 3 , shows a process 300 of another embodiment of a method for determining a tag weight according to the present disclosure, comprising the following steps:

[0046] Step 301, obtaining a portrait data set of a target population, wherein the portrait data set includes at least one original label of the target population and sample users constituting the target population.

[0047] The description of step 301 in this embodiment is consistent with the description of step 201 and will not be repeated here.

[0048] Step 302, mark each original label in the portrait data set as a first label, and determine the weight of the original label through multiple rounds of iterative operations.

[0049] In this embodiment, each original label in the target population portrait data set can be marked as a first label, such as "important" or "label A", and then the weight of the original label is determined through multiple rounds of iterative operations. The iterative operation includes:

[0050] Step 3021, randomly obtain the original label with the first mark in the portrait data set, and construct a copy portrait data set, which includes the copy label.

[0051] In this embodiment, one or more original labels can be randomly obtained from the original labels with the first mark in the portrait dataset, and a copy portrait dataset can be constructed based on the randomly obtained original labels. The labels in the copy portrait dataset are called copy labels.

[0052] Step 3022: Determine a first weight of the original label by using the number of users in the sample who match the original label and the number of users in the sample who do not match the original label.

[0053] In this embodiment, for each original label in at least one original label, the number of users in the sample users that meet the original label, the number of users in the sample users that do not meet the original label, or the total number of sample users (it can be understood that the total number is the sum of the number of users that meet the original label and the number of users that do not meet the original label) can be used to determine the first weight of the original label. For example, the proportion of users in the sample users that meet the original label can be subtracted (or divided by the same operation) from the proportion of users in the sample users that do not meet the original label to determine the first weight of the original label.

[0054] For another example, the difference or quotient between the number of sample users that match the original label and the number of sample users that do not match the original label may be determined as the first weight of the original label.

[0055] Step 3023, using the number of sample users who meet the copy label and the number of sample users who do not meet the copy label to determine the second weight of the copy label, and determine the largest second weight among the second weights of each copy label as the benchmark weight.

[0056] In this embodiment, for each replica tag in the replica portrait data set, the second weight of the replica tag can be determined according to the number of users in the sample users that meet the replica tag, the number of users in the sample users that do not meet the replica tag / or the total number of sample users. It can be understood that a corresponding second weight can be determined for each replica tag in the replica portrait data set, and multiple second weights can be determined based on multiple replica tags in the replica portrait data set. At this time, the largest second weight can be determined among the multiple second weights, and the largest second weight is determined as the reference weight.

[0057] Step 3024: In response to determining that the first weight is less than the reference weight, before performing the next iteration operation, the mark of the original label corresponding to the first weight is updated to the second mark.

[0058] In this embodiment, if it is determined that the first weight of the original label is less than the reference weight determined in the current iteration operation, the mark of the original label corresponding to the first weight can be updated to a second mark, such as "unimportant" or mark "B", before executing the next iteration operation.

[0059] Step 303, in response to determining that the iterative operation meets the stopping condition, the iterative operation is stopped, and the weight of the original label corresponding to the first weight is determined based on the probability that the first weight is greater than or equal to each reference weight determined in multiple rounds of iterative operations.

[0060] In this embodiment, if it is determined that the iterative operation meets the stopping condition, the iterative operation is stopped, and the probability that the first weight of the original label is greater than or equal to each benchmark weight determined in multiple rounds of iterative operations is determined as the final weight of the original label.

[0061] In this embodiment, since the reference weight is determined based on the replica labels randomly selected from the replica portrait data set, the replica labels in the replica portrait data set are gradually optimized in multiple rounds of iterative operations (the original labels marked as the second label are no longer selected as replica labels), so the reference weight is gradually optimized in multiple rounds of iterative operations. It can be seen that this embodiment determines the probability that the first weight of the original label is greater than or equal to the multiple reference weights determined in multiple rounds of iterative operations as the final weight of the original label through multiple rounds of iterative operations, which can improve the accuracy of determining the weight of the original label.

[0062] In addition, by constructing a replica portrait dataset based on the original labels randomly obtained from the portrait dataset, the original labels can be randomly selected within the enumeration range of the portrait dataset to ensure that the replica labels have the same definition domain as the original labels. While expanding the sample range, it ensures that there is no correlation between the randomly introduced replica portrait dataset and the original portrait dataset, which will not cause entropy reduction in the classification task.

[0063] Optionally, in each round of iterative operation, determining the first weight of the original label includes: using each model of a plurality of target models to determine the first weight of the original label output by the model; determining the second weight of the copy label includes: determining the second weight of the copy label output by the model; in response to determining that the first weight is greater than or equal to the benchmark weight, determining that the output result of the model is the first result; in response to determining that the first weight is less than the benchmark weight, before executing the next iterative operation, updating the mark of the original label corresponding to the first weight to the second mark, including: in response to determining that the probability that multiple target models output the first result and all output results does not meet the first preset condition, before executing the next iterative operation, updating the mark of the original label corresponding to the first weight to the second mark.

[0064] In this embodiment, for each round of iterative operation, the steps of determining the first weight and determining the second weight include using each model in a plurality of target models to determine the first weight of each original label and the second weight of each copy label, and determining the largest second weight in at least one second weight as the reference weight of the copy label group, wherein the copy label group is a group of copy labels obtained in each random selection operation. It can be understood that in each round of iterative operation, for the same original label and a group of copy labels randomly obtained in the round of iterative operation, multiple first weights and multiple reference weights output by multiple target models can be obtained. Among them, the multiple target models can be random forest models, logistic regression models, support vector machine models and other models.

[0065] For each model, if the first weight of the original label determined by the model is greater than or equal to the benchmark weight determined by the model, the output result of the model is determined to be the first result; if the first weight of the original label determined by the model is less than the benchmark weight determined by the model, the output result of the model is determined to be the second result.

[0066] In each round of iteration, for a certain original label, based on the output results of multiple target models, if it is determined that the probability of outputting the first result and the second result does not meet the first preset condition, the label of the original label can be updated to the second label before performing the next iteration. The first preset condition can be a preset value.

[0067] For example, in each round of iteration, 100 models are used to determine the first weight and the benchmark weight. If the first weights determined by 60 of the 100 models are greater than or equal to the benchmark weight, and the first weights determined by 40 models are less than the benchmark weight, that is, 60 of the 100 models output the first result and 40 models output the second result. At this time, if the first preset condition is that the probability of outputting the first result needs to be greater than or equal to 70%, then the probability of determining that the output is the first result in all output results does not meet the first preset condition, and before executing the next iteration, the mark of the original label is updated to the second mark.

[0068] In this embodiment, for each round of iterative operation, when determining whether to update the label of a certain original label from a first label to a second label based on the output results of multiple target models, the instability of a single model or the error of the model itself can be eliminated from affecting the determination result, thereby improving the stability and robustness of the system and improving the accuracy of determining the label of the original label in the iterative operation.

[0069] Optionally, not satisfying the first preset condition includes: a probability distribution function of the output probabilities of multiple target models outputting the first result and accounting for all output results not satisfying a second preset condition.

[0070] In this embodiment, failure to satisfy the first preset condition may be that, in each round of iterative operations, for the same original label and the copy label randomly obtained in the round of iterative operations, the probability distribution function of the output probability of multiple target models outputting the first result and all the results output by the multiple target models does not satisfy the second preset condition.

[0071] For example, if the distribution of the probability that the output of multiple target models is the first result accounts for all the outputs of the multiple target models satisfies the binomial distribution (there are only two output situations, such as the output is the first result or the output is the second result), the probability that each model outputs the first result is preset to p (p can be 0.1, 0.5, 0.9, etc.), then its probability density function can be expressed as:

[0072]

[0073] Where n represents the number of models, k represents the number of models whose output results of multiple target models are the first result when the preset probability of each model outputting the first result is p (it can be understood that since the replica labels are randomly obtained, the size of each benchmark weight determined based on the replica labels is variable, and the first result after the first weight is compared with the benchmark weight is variable, so the number changes with the randomly obtained replica labels / or the iterative process). The distribution function of the probability density function can be expressed as:

[0074]

[0075] Wherein, q represents the weight index of the original label, which is used to indicate the accuracy of the original label in describing the population portrait of the target population; Pr(X>k) represents the probability that the number of output results of multiple target models is greater than k for the first result. In this example, if the second preset condition is the preset threshold β, then if q<β, it can be determined that the probability distribution function does not meet the second preset condition.

[0076] It can be understood that not satisfying the second preset condition may also include that the probability distribution function of the output probability of multiple target models outputting the first result accounts for all results conforms to a preset distribution (such as Gaussian distribution, etc.); it may also include that the number of multiple target models outputting the first result meets a preset number, etc.

[0077] In this embodiment, it is determined based on the probability distribution function whether to update the label of a certain original label from the first label to the second label in multiple rounds of iterative operations. That is, based on the probability distribution function obtained in the current iterative operation, it is determined whether a certain original label can be selected as a copy label in the next round of iterative operations (the original label marked as the second label is not selected into the copy label group). The benchmark weight determined according to the copy label can be iteratively updated to improve the accuracy of the original label being marked as the first label.

[0078] In addition, through this embodiment, the weight index of the original label can be determined to indicate the accuracy of the original label in describing the population portrait of the target population.

[0079] Optionally, before executing the next iterative operation, the mark of the original label corresponding to the first weight is updated to a second mark, including: obtaining the probability distribution function of the output probability determined in the current iterative operation and the iterative operation before the current iterative operation; using the obtained multiple probability distribution functions to perform error discovery rate correction; in response to determining that the corrected multiple probability distribution functions do not meet the third preset condition, before executing the next iterative operation, the mark of the original label is updated to the second mark.

[0080] In this embodiment, the distribution function of the probability determined in the current iterative operation and the iterative operation before the current iterative operation can be obtained, and the probability distribution function can be corrected using the False Discovery Rate (FDR) method. If the corrected distribution function obtained based on the probability distribution function of the current iterative operation and the previous iterative operation does not meet the third preset condition, then before executing the next iterative operation, the mark of the original label is updated to the second mark.

[0081] For example, if the current iteration operation is the mth iteration operation, the weight indexes q of multiple original labels from the 1st to the mth rounds can be obtained, and the multiple weight indexes q are sorted in ascending order. After sorting, they are assigned numbers from 1 to m (the smallest q is 1 and the largest q is m). The corrected weight index q of the i-th position is i 'for:

[0082]

[0083] Among them, q m is the weight index of the original tag determined in the current iteration operation. Determine the largest i that meets the preset position threshold α and satisfies q i'<α, it is considered that there are significant differences in all the tests ranked from 1 to i. If i = m, it means that the null hypothesis holds in each round of iterative operation (the operation of marking each original label as the first label is valid. It can be understood that the original label is originally marked as the first label, rather than other labels), that is, the original label is marked as the first label. Conversely, if i < m, the mark of the original label is updated to the second label.

[0084] In this embodiment, the false discovery rate correction method is used to correct the weight index of the original label determined in each round of iterative operation, and based on the corrected weight index, it is determined whether the mark of the original label is the first label, which can improve the accuracy of determining the mark of the original label and optimize the selection range of the copy labels in the iterative operation.

[0085] Optionally, the method further includes: after the iterative operation ends, based on the probability distribution function of the output probability of the first result output by multiple target models determined in the last round of iterative operation and accounting for all the output results, determine the weight index of the original label corresponding to the first weight.

[0086] In this embodiment, after the iterative operation ends, based on the probability distribution function of the output probability of the first result output by multiple target models determined in the last round of iterative operation and accounting for all the output results, the weight index of the original label can be determined, and this weight index is used to indicate the accuracy of the population portrait described by the original label for the target population. Specifically, the weight index of the original label can be the q value corresponding to the above formula (2) obtained in the last round of iterative operation.

[0087] In this embodiment, through multiple rounds of iterative operations, the weight index for characterizing the accuracy of the portrait described by the original label for the target population can be determined, and important / accurate portrait labels can be determined among the numerous portrait labels of the target population.

[0088] Optionally, for each model in multiple target models, the model outputs the first weight of the original label, including: for each decision tree in the random forest model, based on the number of times the node in the decision tree represented by the original label is determined as a split point, determine the first weight, where the split point is used to divide the sample users into different subsets; and, determine the second weight of the copy label output by the model, including: based on the number of times the node in the decision tree represented by each copy label is determined as a split point, determine the second weight of each copy label.

[0089] In this embodiment, for each round of iterative operation, the random forest model can be used to calculate the first weight and the benchmark weight in this round of iterative operation.

[0090] Specifically, the random forest model contains multiple decision trees. For each decision tree in each round of iterative operation, the original label is input into the decision tree. The decision tree calculates the impurity of each node in the decision tree based on the Gini index. It should be noted that the more samples of different categories in a node, the more impure the node is, and the higher the Gini index is. Conversely, the fewer samples of different categories in a node, the purer the node is, and the lower the Gini index is. The Gini index of each node in the decision tree is:

[0091]

[0092] Among them, D is the total number of sample users, and j represents the number of categories under the node. For example, the categories under the node representing the original label "gender" include "male" and "female". For another example, the categories under the node representing the original label "education" can include "bachelor", "master", "doctor", etc.

[0093] Among them, C j represents the number of sample users under the jth original label, that is, the number of sample users in the target population that match the jth original label. For example, in the target population, the number of sample users with the original label "gender" of female is 10, then C j =10.

[0094] When a node is split based on an original label, the decision tree will select an original label to divide the sample users. At this time, the Gini coefficient of the node corresponding to the original label is:

[0095]

[0096] Among them, A represents the original label (for example, characteristics of the target population such as age, gender or education); d1 represents the number of sample users in the D1 subset, and d2 represents the number of sample users in the D2 subset; D1 and D2 represent the two subsets into which the sample users constituting the target population are divided by the original label A. For example, A is the original label of "gender", that is, based on the user's gender, the sample users constituting the target population (the total number is D) are divided into the male user subset D1 and the female user subset D2, and the number of male users in the target population is d1, and the number of female users is d2. At this time, the Gini coefficient of the node representing the original label of "gender" is Gini(D, A).

[0097] When a node is split, all original labels will be traversed, and the Gini coefficient Gini(D, A) of each original label will be calculated respectively. Every time the decision tree splits, the original label that maximizes (Gini(D)-Gini(D, A)) will be used as the basis for splitting the node (it can be understood that the original label can minimize the impurity of the node). If the number of times the original label A is selected as a node in the entire decision tree is T, then the first weight of the original label A is:

[0098]

[0099] The first weight p of the original label A A It represents the average contribution of the original label A to the reduction of the Gini coefficient during the decision tree establishment process, and is used to calculate the first weight of each original label.

[0100] In this embodiment, the above formulas (4) to (6) are based on the splitting of the decision tree based on the label in the label data set in the random forest model, and the weight of the label is determined based on the number of times the label is selected as a split node during the splitting process. When the above label data set is a portrait label data set containing original labels, the obtained weight is the first weight; when the above label data set is a data set including copy labels, the obtained weight is the second weight, and then the largest second weight among the obtained second weights can be determined as the reference weight.

[0101] In this embodiment, in each round of iterative operation, each decision tree in the random forest model is used to determine the first weight and the baseline weight of the original label, which can improve the accuracy of determining the first weight and the baseline weight, so as to continuously optimize the conditions for distinguishing the marking of the original label through multiple rounds of iterative operations, thereby improving the accuracy of determining the marking of the original label.

[0102] Further references Figure 4 , shows a process 400 of another embodiment of a method for determining a tag weight according to the present disclosure, comprising the following steps:

[0103] Step 401, obtaining a portrait data set of a target population, wherein the portrait data set includes at least one original label of the target population and sample users constituting the target population.

[0104] Step 402: Determine the weight of the original label according to the number of sample users that match the original label and the total number of sample users.

[0105] Step 403, obtaining a comparison portrait data set, wherein the comparison portrait data set includes at least one comparison tag of a comparison group of people and comparison users constituting the comparison group of people, wherein the comparison group of people has the opposite behavior to the target group of people.

[0106] Step 404: determine the weight of the comparison tag according to the number of the comparison users that match the comparison tag and the total number of the comparison users.

[0107] The description of step 401 and step 402 in this embodiment is consistent with the description of step 201 and step 202, and will not be repeated here.

[0108] The above-mentioned embodiments described in the present application determine the weight of each label of a certain population and the weight index of each label based on the data in the portrait data set of the population. In the steps described in the above-mentioned embodiments, the population is the target population and the portrait data set is the portrait data set of the target population, then the weight of the determined label is the weight of the original label of the target population, and the weight index of the determined label is the weight index of the original label of the target population; if the steps described in the above-mentioned embodiments, the population is the comparison population and the portrait data set is the comparison portrait data set of the comparison population, then according to the steps described in the above-mentioned embodiments, the weight of the comparison label of the comparison population and the weight index of the comparison label can be determined.

[0109] Among them, the comparison population refers to a population with opposite behaviors to the target population. For example, if the target population is users who browsed an item on a shopping website and placed an order within a preset time period, then the comparison population can be users who browsed the item on the shopping website but did not place an order within the preset time period, or the comparison population can also be users who did not browse the shopping website within the preset time period.

[0110] Step 405 , in response to determining that the original label is the same as the comparison label, determine whether the label is a label for the target population based on the weight of the original label corresponding to the same label and the weight of the comparison label corresponding to the label.

[0111] In this embodiment, if an original label of the target population is the same as a comparison label of the comparison population, it can be determined whether the label can be used as the label of the target population based on the weight of the original label corresponding to the same label and the weight of the comparison label corresponding to the same label.

[0112] For example, if the target population's original labels include the label "like traveling", and the weight of the original label is 0.5, and the weight index of the original label is 0.6, and the comparison label of the comparison population also includes the label "like traveling", and the weight of the comparison label is 0.8, and the weight index of the comparison label is 0.7, then, based on the fact that the weight of "like traveling" as the original label is less than its weight as the comparison label, the label "like traveling" can be not used as the original label of the target population (or the mark of the original label can be determined as the second mark / unimportant in the portrait data set); or, based on the fact that the weight index of "like traveling" as the original label is less than its weight index as the comparison label, the label "like traveling" can be not used as the original label of the target population; or, based on the fact that the product of the weight of "like traveling" as the original label and the weight index is less than the product of the weight index as the comparison label and the weight index, the label "like traveling" can be not used as the original label of the target population, and so on.

[0113] In this embodiment, a comparison population is introduced, and based on the relationship between the weights or weight indices of each label of the comparison population and the weights or weight indices of each label of the target population, it is determined whether a certain label is a label of the target population, that is, whether the label can be used to accurately describe the characteristics of the target population, which can improve the accuracy of determining the label of the target population.

[0114] In some application scenarios, the method includes:

[0115] Step 1: obtain a portrait data set of the target population, which includes at least one original label of the target population and sample users that constitute the target population; mark the importance of all original labels as important, and set the maximum number of iterations of the algorithm.

[0116] Perform the following iterations:

[0117] Step 2: Randomly select a random number of labels from all original label values ​​marked as important to form a replica portrait dataset. The original labels selected into the replica portrait dataset are called replica labels.

[0118] Step 3: Run a random forest classifier containing n decision trees, and collect the first weight of the original image label and the second weight of the copy label output by each tree. The first weight is calculated according to the above formulas (4) to (6), and similarly, the second weight is calculated according to the above formulas (4) to (6).

[0119] It can be understood that the above formulas (4) to (6) determine the weights of each label in the label data set based on the labels in the label data set and the sample users corresponding to the labels. When the above label data set is a data set including original labels, the obtained weight is the first weight; when the above label data set is a data set including copy labels, the obtained weight is the second weight.

[0120] Step 4: In each tree, the largest second weight in the copy label is determined as the base weight of each tree, and then each original label is compared with the base weight.

[0121] In this step, since the first weights of the original labels do not satisfy the Bernoulli distribution, the importance of the original labels cannot be directly judged based on the first weights. Therefore, in step 1, it is pre-assumed that the first weight of the original label is greater than the base weight of the copy label group, that is, all the original labels are marked as important, and the importance of the original labels is judged through the calculation of step 4. Specifically, the number "0" can be used to represent unimportant, and the number "1" can be used to represent unimportant. For each original label, on n decision trees, whether this assumption is true satisfies the binomial distribution with probability p = 0.5, and its probability density function is:

[0122]

[0123]

[0124] in,

[0125] Where n is the number of decision trees, k is the number of decision trees that meet the assumption, and Pr(X>k) represents the probability that the number of output results of multiple decision trees is 1 is greater than k. Then q is the weight index of the original label.

[0126] Step 5: After calculating in step 4, each original label is calculated to obtain its respective weight index.

[0127] Step 6: Perform FDR correction on the weight index q of each portrait label. If the corrected weight index fails to meet the preset standard, the original label is marked as unimportant.

[0128] It should be noted that, in order to ensure the accuracy of FDR correction, when the number of iterations does not meet the preset calibration number (for example, the number of iterations does not meet 3 times), the mark of the original label is not updated.

[0129] In this step, the FDR correction of the weight index of each original label can be performed according to the above formulas (1) to (3).

[0130] Step 7: Repeat steps 2 to 6 until no original labels are marked as unimportant, or the algorithm has reached the maximum number of iterations set in step 1, and then stop iterative modeling.

[0131] Step 8: Output the weight index of each original label determined in the last round of iteration, and the judgment status (important or unimportant).

[0132] In this application scenario, a comparison sample can be added, that is, in step 1, a comparison portrait data set of a comparison population is obtained, the comparison portrait data set includes at least one comparison label of the comparison population and comparison users constituting the comparison population, and the comparison population is a population with opposite behaviors to the target population;

[0133] By performing steps 2 to 8 on the comparison portrait data set, the weight index and discrimination status of each comparison label can be obtained.

[0134] Step 9, in response to determining that the original label is the same as the comparison label, determine whether the label is a label for the target population according to the weight of the original label corresponding to the label and the weight of the comparison label corresponding to the label.

[0135] Further references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a device for determining a tag weight. Figure 2 , Figure 3 or Figure 4 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0136] like Figure 5 As shown, the apparatus 500 for determining the label weight of this embodiment includes: an acquisition unit 501 and a determination unit 502. The acquisition unit is configured to acquire a portrait data set of a target population, wherein the portrait data set includes at least one original label of the target population and sample users constituting the target population; the determination unit is configured to determine the weight of the original label according to the number of users in the sample users that meet the original label and the total number of sample users.

[0137] In some embodiments, the determination unit includes: a marking module, configured to mark each original label in the portrait data set as a first label, and determine the weight of the original label through multiple rounds of iterative operations; the iterative operation includes: a first acquisition module, configured to randomly obtain the original label with the first label in the portrait data set, and construct a copy portrait data set, the copy portrait data set including the copy label; a first determination module, configured to determine the first weight of the original label by using the number of users in the sample users who meet the original label and the number of users in the sample users who do not meet the original label; a second determination module, configured to determine the second weight of the copy label by using the number of users in the sample users who meet the copy label and the number of users in the sample users who do not meet the copy label, and determine the largest second weight among the second weights of each copy label as the benchmark weight; an update module, configured to update the label of the original label corresponding to the first weight to the second label before performing the next iterative operation in response to determining that the first weight is less than the benchmark weight; a third determination module, configured to stop the iterative operation in response to determining that the iterative operation meets the stop condition, and determine the weight of the original label corresponding to the first weight based on the probability that the first weight is greater than or equal to each benchmark weight determined in multiple rounds of iterative operations.

[0138] In some embodiments, in each round of iterative operation, the first determination module includes: a first determination submodule, configured to use each model of a plurality of target models to determine the first weight of the original label output by the model; the second determination module includes: a second determination submodule, configured to determine the second weight of the copy label output by the model; the output module is configured to determine that the output result of the model is the first result in response to determining that the first weight is greater than or equal to the benchmark weight; the update module includes: a first update submodule, configured to update the mark of the original label corresponding to the first weight to the second mark before executing the next iterative operation in response to determining that the probability that the output of multiple target models is the first result and accounts for all output results does not meet the first preset condition.

[0139] In some embodiments, the first preset condition is not met, including: the probability distribution function of the output probability of multiple target models outputting the first result and accounting for all output results does not meet the second preset condition.

[0140] In some embodiments, the first update submodule includes: a second acquisition module, configured to acquire the probability distribution function of the output probability determined in the current iteration operation and the iteration operation before the current iteration operation; a correction module, configured to use the acquired multiple probability distribution functions to perform error discovery rate correction; and a second update submodule, configured to update the mark of the original label to the second mark before executing the next iteration operation in response to determining that the corrected multiple probability distribution functions do not meet the third preset condition.

[0141] In some embodiments, the device also includes: a weight index determination module, which is configured to determine the weight index of the original label corresponding to the first weight based on the probability distribution function of the output probability of multiple target models determined in the last round of iterative operations being the first result and accounting for all output results after the iterative operation is completed.

[0142] In some embodiments, a first determination submodule includes: a first classification module, configured to use each decision tree in a random forest model to determine a first weight based on the number of times a node in the decision tree represented by the original label is determined as a splitting point, wherein the splitting point is used to divide the sample users into different subsets; and a second determination submodule includes: a second classification module, configured to determine a second weight for each copy label based on the number of times a node in the decision tree represented by each copy label is determined as a splitting point, and determine the largest second weight among the second weights as a benchmark weight.

[0143] In some embodiments, the device includes: a comparison data acquisition unit, configured to acquire a comparison portrait data set, the comparison portrait data set including at least one comparison label of a comparison population and comparison users constituting the comparison population, the comparison population being a population having the opposite behavior to the target population; a comparison data determination unit, configured to determine the weight of the comparison label based on the number of comparison users that meet the comparison label and the total number of comparison users; a comparison unit, configured to determine whether the label is the label of the target population in response to determining that the original label is the same as the comparison label, based on the weight of the original label corresponding to the same label and the weight of the comparison label corresponding to the label.

[0144] Each unit in the above device 500 is similar to the reference Figure 2 , Figure 3 or Figure 4 Therefore, the operations, features and technical effects described above for the method for determining tag weights are also applicable to the device 500 and the units contained therein, and will not be described in detail here.

[0145] According to an embodiment of the present application, the present application also provides an electronic device and a readable storage medium.

[0146] like Figure 6 , is a block diagram of an electronic device 600 for determining a tag weight according to an embodiment of the present application. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or required herein.

[0147] like Figure 6 As shown, the electronic device includes: one or more processors 601, a memory 602, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected using different buses and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the electronic device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In other embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple electronic devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 601 is taken as an example.

[0148] The memory 602 is a non-transient computer-readable storage medium provided in the present application. The memory stores instructions executable by at least one processor to enable the at least one processor to perform the method for determining label weights provided in the present application. The non-transient computer-readable storage medium of the present application stores computer instructions, which are used to enable a computer to perform the method for determining label weights provided in the present application.

[0149] The memory 602 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs and modules, such as the program instructions / modules corresponding to the method for determining the tag weight in the embodiment of the present application (for example, the attached Figure 5 The processor 601 executes various functional applications and data processing of the server by running the non-transient software programs, instructions and modules stored in the memory 602, that is, the method for determining the tag weight in the above method embodiment is implemented.

[0150] The memory 602 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the electronic device for extracting video clips, etc. In addition, the memory 602 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory 602 may optionally include a memory remotely arranged relative to the processor 601, and these remote memories may be connected to the electronic device for extracting video clips via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0151] The electronic device for determining the tag weight method may further include: an input device 603, an output device 604, and a bus 605. The processor 601, the memory 602, the input device 603, and the output device 604 may be connected via the bus 605 or in other ways. Figure 6 The connection via bus 605 is taken as an example.

[0152] The input device 603 can receive input digital or character information, and generate key signal input related to user settings and function control of the electronic device for extracting video clips, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator rod, one or more mouse buttons, a trackball, a joystick, and other input devices. The output device 604 may include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0153] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0154] These computer programs (also referred to as programs, software, software applications, or code) include machine instructions for programmable processors and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or means (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0156] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0157] A computer system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship to each other.

[0158] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0159] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A method for determining a tag weight, comprising: Acquire a portrait data set of a target population, wherein the portrait data set includes at least one original label of the target population and sample users constituting the target population; Determining the weight of the original label according to the number of users in the sample users that match the original label and the total number of the sample users includes: Each original label in the portrait data set is marked as a first label, and the weight of the original label is determined through multiple rounds of iterative operations; the iterative operations include: Randomly obtain an original label with the first mark in the portrait data set, and construct a copy portrait data set, wherein the copy portrait data set includes a copy label; Determine a first weight of the original label by using the number of users among the sample users that match the original label and the number of users among the sample users that do not match the original label; Determine the second weight of the copy label by using the number of users in the sample users who match the copy label and the number of users in the sample users who do not match the copy label, and determine the largest second weight among the second weights of the copy labels as the reference weight; In response to determining that the first weight is less than the reference weight, before performing the next iteration operation, updating the mark of the original label corresponding to the first weight to a second mark; In response to determining that the iterative operation satisfies a stopping condition, the iterative operation is stopped, and a weight of an original label corresponding to the first weight is determined based on a probability that the first weight is greater than or equal to each reference weight determined in multiple rounds of iterative operations.

2. The method according to claim 1, wherein: In each round of the iterative operation, the determining the first weight of the original label includes: using each model of a plurality of target models to determine the first weight of the original label output by the model; The determining the second weight of the copy label includes: determining the second weight of the copy label output by the model; In response to determining that the first weight is greater than or equal to the reference weight, determining that the output result of the model is a first result; In response to determining that the first weight is less than the reference weight, before performing the next iteration operation, updating the mark of the original label corresponding to the first weight to a second mark, comprises: In response to determining that the probability that the multiple target models output the first result and account for all output results does not meet a first preset condition, before performing the next iteration operation, the mark of the original label corresponding to the first weight is updated to the second mark.

3. The method according to claim 2, wherein: The failure to meet the first preset condition includes: The probability distribution function of the output probabilities of the multiple target models outputting the first result and accounting for all output results does not satisfy the second preset condition.

4. The method according to claim 3, wherein: Before performing the next iteration operation, updating the mark of the original label corresponding to the first weight to the second mark includes: Obtaining a probability distribution function of the output probability determined in a current iterative operation and an iterative operation before the current iterative operation; The obtained multiple probability distribution functions are used to perform false discovery rate correction; In response to determining that the corrected multiple probability distribution functions do not satisfy the third preset condition, before performing the next iteration operation, the mark of the original label is updated to the second mark.

5. The method according to claim 3, wherein: The method further comprises: After the iteration operation is completed, the weight index of the original label corresponding to the first weight is determined based on the probability distribution function of the output probability of the multiple target models determined in the last round of iteration operation being the first result and accounting for all output results.

6. The method according to any one of claims 2 to 5, wherein: The adopting of each model of the plurality of target models, wherein the model outputs a first weight of the original label, comprises: Using each decision tree in the random forest model, determining the first weight based on the number of times a node in the decision tree represented by the original label is determined as a split point, wherein the split point is used to divide the sample users into different subsets; and The determining a second weight of the copy label output by the model includes: Based on the number of times that the nodes in the decision tree represented by the respective copy labels are determined as split points, the second weights of the respective copy labels are determined.

7. The method according to claim 1, wherein: The method comprises: Acquire a comparison portrait data set, wherein the comparison portrait data set includes at least one comparison label of a comparison group of people and comparison users constituting the comparison group of people, wherein the comparison group of people has the opposite behavior to the target group of people; Determining a weight of the comparison tag according to the number of the comparison users that match the comparison tag and the total number of the comparison users; In response to determining that the original label is the same as the comparison label, it is determined whether the label is a label of the target population according to the weight of the original label corresponding to the same label and the weight of the comparison label corresponding to the label.

8. A device for determining a tag weight, comprising: An acquisition unit is configured to acquire a portrait data set of a target population, wherein the portrait data set includes at least one original label of the target population and sample users constituting the target population; A determination unit is configured to determine the weight of the original label according to the number of users in the sample users that match the original label and the total number of the sample users, including: The marking module is configured to mark each original label in the portrait data set as a first label, and determine the weight of the original label through multiple rounds of iterative operations; the iterative operations include: A first acquisition module is configured to randomly acquire an original label having the first mark in the portrait data set, and construct a copy portrait data set, wherein the copy portrait data set includes a copy label; A first determination module is configured to determine a first weight of the original label by using the number of users in the sample users that match the original label and the number of users in the sample users that do not match the original label; A second determination module is configured to determine a second weight of the copy label by using the number of users in the sample users that match the copy label and the number of users in the sample users that do not match the copy label, and determine the largest second weight among the second weights of the copy labels as a reference weight; an updating module, configured to update the mark of the original label corresponding to the first weight to a second mark before performing the next iteration operation in response to determining that the first weight is less than the reference weight; The third determination module is configured to stop the iterative operation in response to determining that the iterative operation satisfies the stopping condition, and determine the weight of the original label corresponding to the first weight based on the probability that the first weight is greater than or equal to each benchmark weight determined in multiple rounds of iterative operations.

9. The device according to claim 8, wherein: In each round of the iterative operation, the first determination module includes: a first determination submodule configured to use each model of a plurality of target models to determine a first weight of the original label output by the model; The second determination module includes: a second determination submodule configured to determine a second weight of the replica label output by the model; an output module, configured to, in response to determining that the first weight is greater than or equal to the reference weight, determine that an output result of the model is a first result; The update module comprises: The first updating submodule is configured to update the mark of the original label corresponding to the first weight to the second mark before executing the next iteration operation in response to determining that the probability that the multiple target models output the first result and account for all the output results does not meet the first preset condition.

10. The device according to claim 9, wherein: The failure to meet the first preset condition includes: The probability distribution function of the output probabilities of the multiple target models outputting the first result and accounting for all output results does not satisfy the second preset condition.

11. The device according to claim 10, wherein: The first updating submodule includes: A second acquisition module is configured to acquire a probability distribution function of the output probability determined in a current iteration operation and an iteration operation before the current iteration operation; A correction module is configured to perform false discovery rate correction using the obtained multiple probability distribution functions; The second updating submodule is configured to update the mark of the original label to the second mark before performing the next iteration operation in response to determining that the corrected multiple probability distribution functions do not meet the third preset condition.

12. The device according to claim 10, wherein: The device also includes: The weight index determination module is configured to determine the weight index of the original label corresponding to the first weight after the iterative operation is completed based on the probability distribution function of the output probability of the multiple target models determined in the last round of iterative operation being the first result and accounting for all output results.

13. The device according to any one of claims 9 to 12, wherein: The first determining submodule includes: a first classification module configured to use each decision tree in a random forest model to determine the first weight based on the number of times a node in the decision tree represented by the original label is determined as a split point, wherein the split point is used to divide the sample users into different subsets; and The second determining submodule includes: The second classification module is configured to determine the second weight of each copy label based on the number of times the node in the decision tree represented by each copy label is determined as a split point.

14. The device according to claim 8, wherein: The device comprises: A comparison data acquisition unit is configured to acquire a comparison portrait data set, wherein the comparison portrait data set includes at least one comparison tag of a comparison group of people and comparison users constituting the comparison group of people, wherein the comparison group of people has the opposite behavior to the target group of people; a comparison data determination unit, configured to determine a weight of the comparison tag according to the number of the comparison users that meet the comparison tag and the total number of the comparison users; The comparison unit is configured to, in response to determining that the original label is the same as the comparison label, determine whether the label is the label of the target population according to the weight of the original label corresponding to the same label and the weight of the comparison label corresponding to the label.

15. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 7.

Images