User label evaluation method, device, equipment and computer storage medium

Abstract

The present application discloses a method, device, equipment and computer storage medium for user label evaluation, and relates to the field of data edge computing technology. The method includes: obtaining label features to be evaluated, associated label features and label association relationships to be evaluated, the label features to be evaluated include label vectors to be evaluated, and the associated label features include associated label vectors and associated label scores; constructing label graph data to be evaluated according to the label features to be evaluated, the associated label features and the label association relationships to be evaluated, the label graph data to be evaluated is composed of multiple nodes and edges of nodes, the nodes are label features to be evaluated or associated label features, and the edges of nodes are label association relationships to be evaluated; inputting the label graph data to be evaluated into the label evaluation model, encoding the label features to be evaluated and the target label features and performing weighted summation to obtain the predicted score of the label to be evaluated. The embodiment of the present application can automatically evaluate the label, thereby improving the accuracy of user label evaluation.

Description

Technical Field

[0001] The present application belongs to the field of data edge computing technology, and in particular, relates to a method, device, equipment and computer storage medium for user label evaluation. Background Art

[0002] In the field of data edge computing technology, user portraits play an important role in personalized recommendations, and user tags are the basic elements for building user portraits. User tags can be used as an important indicator for analyzing user behavior, and user tags can be used to achieve accurate user positioning.

[0003] In the prior art, when evaluating user tags, relevant experts often manually evaluate user tags based on their own professional knowledge and experience. However, when evaluating user tags manually, it depends on the subjective judgment and experience of the evaluator, and the evaluation results are often highly subjective. Different evaluators may give different evaluation results. Therefore, the accuracy of user tag evaluation by the existing methods is low. Summary of the invention

[0004] The embodiments of the present application provide a method, apparatus, device and computer storage medium for evaluating user labels to solve the problem of low accuracy of user label evaluation in existing methods.

[0005] In a first aspect, an embodiment of the present application provides a method for evaluating a user tag, the method comprising:

[0006] Obtaining the tag features to be evaluated, the associated tag features and the associated relationship of the tag to be evaluated, wherein the tag features to be evaluated include the tag vector to be evaluated, and the associated tag features include the associated tag vector and the associated tag score;

[0007] Constructing label graph data to be evaluated according to the label features to be evaluated, the associated label features and the label association relationships to be evaluated, wherein the label graph data to be evaluated is composed of a plurality of nodes and edges of nodes, wherein the nodes are label features to be evaluated or associated label features, and the edges of the nodes are label association relationships to be evaluated;

[0008] The label graph data to be evaluated is input into the label evaluation model, the label features to be evaluated and the target label features are encoded to obtain the encoding result, and the predicted score of the label to be evaluated is obtained according to the encoding result. The target label feature is the associated label feature associated with the label feature to be evaluated.

[0009] In a second aspect, an embodiment of the present application provides a device for evaluating user tags, the device comprising:

[0010] An acquisition module is used to acquire the label features to be evaluated, the associated label features and the label association relationship of the label to be evaluated, the label features to be evaluated include the label vector to be evaluated, and the associated label features include the associated label vector and the associated label score;

[0011] A construction module is used to construct the label graph data to be evaluated according to the label features to be evaluated, the associated label features and the label association relationship, wherein the label graph data to be evaluated is composed of multiple nodes and edges of nodes, wherein the nodes are the label features to be evaluated or the associated label features, and the edges of the nodes are the label association relationships;

[0012] The evaluation module is used to input the label graph data to be evaluated into the label evaluation model, encode the label features to be evaluated and the target label features to obtain the encoding results, and obtain the predicted score of the label to be evaluated based on the encoding results. The target label features are the associated label features associated with the label features to be evaluated.

[0013] In a third aspect, an embodiment of the present application provides a terminal device, comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for user tag evaluation as in the first aspect is implemented.

[0014] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the method for user tag evaluation according to the first aspect is implemented.

[0015] In a fifth aspect, an embodiment of the present application provides a computer program product. When instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method for user tag evaluation as in the first aspect.

[0016] The embodiment of the present application provides a method, device, equipment and computer storage medium for user label evaluation. The method first obtains the label features to be evaluated, the associated label features and the associated relationship of the label to be evaluated, the label features to be evaluated include the label vector to be evaluated, and the associated label features include the associated label vector and the associated label score; the connection between the labels is established through the label features and the associated relationship. The label graph data to be evaluated is constructed according to the label features to be evaluated, the associated label features and the associated relationship of the label to be evaluated. The label graph data to be evaluated is composed of multiple nodes and edges of nodes, wherein the nodes are the label features to be evaluated or the associated label features, and the edges of the nodes are the associated relationships of the labels to be evaluated; the associated relationships and feature information between the labels can be more clearly displayed, which is conducive to modeling and analyzing the complex relationships between the labels. The label graph data to be evaluated is input into the label evaluation model, the label features to be evaluated and the target label features are encoded and weighted summed to obtain the predicted score of the label to be evaluated; the label graph data is constructed by using the collected data, and the complex label relationship is converted into an easy-to-process graphic structure and the model is trained. By using the label vector and label score in the label feature, a more comprehensive and comprehensive node feature representation can be provided, the associated information of the training label can be more effectively utilized, the usage value of the label can be automatically scored, and the accuracy of user label evaluation can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 It is a flowchart of a method for evaluating user tags provided in an embodiment of the present application;

[0019] Figure 2 It is a flowchart of the label encoding method provided in the embodiment of the present application;

[0020] Figure 3 It is a flowchart of the method for training a label encoding model provided in an embodiment of the present application;

[0021] Figure 4 It is a flowchart of a method for training a label evaluation model provided in an embodiment of the present application;

[0022] Figure 5 It is a flowchart of the method for training a discriminator provided in an embodiment of the present application;

[0023] Figure 6 This is a schematic diagram of dividing the number of times a tag is used by days provided in an embodiment of the present application;

[0024] Figure 7 This is a schematic diagram of the division of the tag response rate by day provided in the embodiment of the present application;

[0025] Figure 8 It is a structural schematic diagram of a device for evaluating user tags provided in an embodiment of the present application;

[0026] Fig. 9 It is a schematic diagram of the structure of the terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0028] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0029] In the prior art, when evaluating user tags, it is often done manually based on professional knowledge and experience. However, when evaluating user tags manually, it depends on the subjective judgment and experience of the evaluator, and the evaluation results are often highly subjective. Different evaluators may give different evaluation results. Therefore, the accuracy of user tag evaluation by the prior art is low.

[0030] In order to solve the existing technology, the embodiment of the present application provides a method, device, equipment and computer storage medium for user label evaluation. The method first obtains the label features to be evaluated, the associated label features and the associated relationship of the label to be evaluated, the label features to be evaluated include the label vector to be evaluated, and the associated label features include the associated label vector and the associated label score; the connection between the labels is established through the label features and the associated relationship. According to the label features to be evaluated, the associated label features and the associated relationship of the label to be evaluated, the label graph data to be evaluated is constructed, and the label graph data to be evaluated is composed of multiple nodes and edges of nodes, wherein the nodes are the label features to be evaluated or the associated label features, and the edges of the nodes are the associated relationships of the labels to be evaluated; it can more clearly show the associated relationships and feature information between the labels, which is conducive to modeling and analyzing the complex relationships between the labels. The label graph data to be evaluated is input into the label evaluation model, the label features to be evaluated and the target label features are encoded and weighted summed to obtain the predicted score of the label to be evaluated; by using the collected data to construct the label graph data, the complex label relationship is converted into an easy-to-handle graphic structure and the model is trained. By using the label vector and label score in the label feature, a more comprehensive and comprehensive node feature representation can be provided, the associated information of the training label can be more effectively utilized, the usage value of the label can be automatically scored, and the accuracy of user label evaluation can be improved.

[0031] The following first introduces the method for evaluating user tags provided in the embodiments of the present application in conjunction with the accompanying drawings.

[0032] Figure 1 FIG. 1 is a flow chart of a method for evaluating user tags provided by an embodiment of the present application. Figure 1 As shown, the method may include the following steps: S101 to S103.

[0033] S101, obtaining a tag feature to be evaluated, an associated tag feature, and an associated relationship of the tag to be evaluated, wherein the tag feature to be evaluated includes a tag vector to be evaluated, and the associated tag feature includes an associated tag vector and an associated tag score.

[0034] Among them, the features of the label to be evaluated are the relevant features of the label that needs to be evaluated; the associated label features are the relevant features of the associated labels associated with the label to be evaluated; the association relationship of the label to be evaluated is the association relationship between the label to be evaluated and its associated labels; the label vector is the embedded representation obtained after encoding the label, and the label score is the score obtained based on the relevant data performance of the label.

[0035] In some embodiments, the tag association relationship can be defined as a tag pair that appears more than 7 times at the same time is associated with each other, and if the number of simultaneous appearances is less than or equal to 7 times or has never appeared at the same time, they are not associated with each other.

[0036] In some embodiments, the tag score may include a social influence score, a data quality score, a user feedback score, and a usage effect score.

[0037] The social influence score is the prevalence of user tags on the Internet, the degree to which their semantics match user needs, and the degree to which they resonate with users on an emotional level; it can include the number of platforms covered by the tag, the tag semantic relevance score, and the tag attractiveness score.

[0038] The data quality score is the quality level of the user tag in the mathematical properties of the data itself; it can include the score of the number of users covered by the tag, the score of the proportion of users covered by the tag, the score of the amount of information in the tag, and the score of the timeliness of the tag.

[0039] The user feedback score is various forms of feedback from system users about the tag, which characterizes the system users' reliance on the tag, the frequency of tag usage, and the popularity of the tag. It may include a score for the number of times the tag is used, a score for the number of times the tag is searched, a score for the number of times the tag is viewed, and a score for the number of times the tag is collected.

[0040] The usage effect score is the degree of improvement in the response rate of marketing activities and the return on investment of marketing activities after using tags for marketing activities; it can include the score of the improvement in the response rate of tag marketing activities and the score of the improvement in the return on investment (ROI).

[0041] In some embodiments, the label vector can be obtained by encoding the label to be evaluated through a pre-trained model, and the pre-trained encoding model can be a bidirectional encoder representation model based on a transformer (Bidirectional Encoder Representations from Transformers, BERT); the label score can be obtained by scoring the label according to preset rules.

[0042] By considering the characteristics of the label itself, the characteristics of the associated labels and the association relationship between the labels, the label to be evaluated can be comprehensively evaluated from multiple dimensions, thereby improving the accuracy of the evaluation.

[0043] S102, constructing label graph data to be evaluated according to the label features to be evaluated, the associated label features and the label association relationships to be evaluated, wherein the label graph data to be evaluated is composed of a plurality of nodes and edges of nodes, wherein the nodes are label features to be evaluated or associated label features, and the edges of the nodes are label association relationships to be evaluated.

[0044] When constructing the label graph data to be evaluated, the label features to be evaluated and the associated label features are used as multiple nodes of the label graph data to be evaluated, and the nodes with associated relationships are connected as edges of the nodes according to the associated relationships of the labels to be evaluated.

[0045] By converting label features and label associations into graph data structures and visualizing complex label relationships, we can more intuitively represent the associations between labels and help discover potential label association patterns.

[0046] S103, inputting the label graph data to be evaluated into the label evaluation model, encoding the label features to be evaluated and the target label features to obtain encoding results, and determining the predicted score of the label to be evaluated according to the encoding results, wherein the target label features are associated label features associated with the label features to be evaluated.

[0047] In some embodiments, after the label graph data to be evaluated is input into the label evaluation model, the label evaluation model selects the associated label features associated with the label to be evaluated as the target label features based on the label connection relationship in the label graph data to be evaluated; a pre-trained neural network is used to perform a forward propagation operation on the label features to be evaluated and the target label features to obtain embedded representations of the label to be evaluated and the target label, and the obtained embedded representations are weighted summed according to the pre-trained weights to obtain a predicted score for the label to be evaluated.

[0048] The evaluation label features and target label features are encoded and weightedly summed through the label evaluation model. The evaluation model can capture the correlation and mutual influence between labels, integrate the information of multiple features, and improve the accuracy of the evaluation.

[0049] The connection between labels is established by obtaining label features and label associations. The label graph data to be evaluated is constructed based on the features of the label to be evaluated, the features of the associated labels, and the associations of the labels to be evaluated. This can more clearly show the associations and feature information between labels, and is conducive to modeling and analyzing the complex relationships between labels. The label graph data to be evaluated is input into the label evaluation model, and the features of the label to be evaluated and the target label features are encoded and weighted summed to obtain the predicted score of the label to be evaluated; by using the collected data to construct the label graph data, the complex label relationship is converted into an easy-to-process graph structure and the model is trained. By using the label vectors and label scores in the label features, a more comprehensive and comprehensive node feature representation can be provided, the association information of the training labels can be more effectively utilized, the use value of the labels can be automatically scored, and the accuracy of user label evaluation can be improved.

[0050] In some embodiments, obtaining the associated tag features of the tag to be evaluated may include:

[0051] Get the associated label vector and associated label score of the label to be evaluated;

[0052] Convert the associated label score into an associated label score vector;

[0053] Concatenate the associated label vector and the associated label score vector to obtain the associated label features of the label to be evaluated.

[0054] The associated label vector can represent the label information in the form of a vector, and the associated label score provides a quantitative assessment of the importance of the label. By concatenating the score vector with the original label vector, qualitative and quantitative information can be integrated to provide a more comprehensive label feature representation.

[0055] In some embodiments, Figure 2 As shown, obtaining the label vector to be evaluated and the associated label vector may include: S201 to S204.

[0056] S201, obtaining a tag to be evaluated and an associated tag of the tag to be evaluated;

[0057] S202, using a preset word segmenter to perform word segmentation processing on the tag to be evaluated and the associated tag to obtain a string of the tag to be evaluated and a string of the associated tag;

[0058] S203, converting the tag string to be evaluated and the associated tag string into corresponding identifiers to obtain a tag identifier to be evaluated and an associated tag identifier;

[0059] S204, inputting the tag identifier to be evaluated and the associated tag identifier into a tag encoding model, encoding the tag identifier to be evaluated and the associated tag identifier respectively through the tag encoding model to obtain a tag vector to be evaluated and an associated tag vector.

[0060] Converting strings into identifiers and into the specified input format of the evaluation model, and through automated word segmentation and vectorization processing, it is possible to deeply understand the semantic content of labels, capture contextual information, and improve the accuracy and efficiency of label evaluation.

[0061] In some embodiments, Figure 3 As shown, before inputting the tag identifier to be evaluated and the associated tag identifier into the tag encoding model, the method may further include: S301 to S305.

[0062] S301, obtaining an identifier of a training label and a category of the training label;

[0063] S302, inputting the identifier of the training label into the encoding model, forward propagating the identifier of the training label through the neural network of the encoding model, and obtaining a label vector of the training label, wherein the neural network of the encoding model has a preset first model parameter;

[0064] S303, calculating the vector distance between the label vectors of the training labels according to the label vectors of the training labels;

[0065] S304, determining a target loss function value of the encoding model according to the category of the training label and the vector distance between the label vectors of the training label;

[0066] S305, when the target loss function value is greater than the set threshold, redetermine the first model parameter according to the target loss function value and update the encoding model until the target loss function value is less than the set threshold, thereby obtaining a label encoding model.

[0067] Through the forward propagation of the neural network, deep features can be extracted from the label identifiers. Calculating the vector distance can help evaluate the similarity between labels. By calculating the loss function through the vector distance and updating the model, the accuracy of label encoding and the generalization ability of the model can be improved.

[0068] In some embodiments, determining the target loss function value of the encoding model according to the category of the training label and the vector distance between the training labels may include:

[0069] When the categories of the training labels are the same, determining a target first loss function value as a target loss function value according to a relationship between the first vector distance and the first loss function value;

[0070] When the categories of the training labels are different, the target second loss function is determined as the target loss function value according to the relationship between the second vector distance and the second loss function value.

[0071] By selecting different target loss function values ​​according to the analogy of the training labels, we can better learn the similarities between labels of the same category and the differences between labels of different categories, and improve the generalization ability of the model.

[0072] The relationship between the first vector distance and the first loss function value can be:

[0073]

[0074] Among them, L 1 is the first loss function value, D pos is the first vector distance.

[0075] The relationship between the second vector distance and the second loss function value can be:

[0076] L 2 =0.5*max(0,margin-D neg ) 2

[0077] Among them, L2 is the value of the second loss function, margin is the preset hyperparameter, D neg is the second vector distance, and this preset hyperparameter can be set to 1.

[0078] In some embodiments, re-determining the first model parameter according to the target loss function value may include:

[0079] Determine the gradient of a first model parameter of the encoding model according to the target loss function value;

[0080] The first model parameter is re-determined according to the gradient of the first model parameter.

[0081] The size of the gradient determines the step size of the parameter update. By calculating the gradient, the model can dynamically adjust the parameters according to the current training status, thereby improving the accuracy of the model.

[0082] In some embodiments, the label evaluation model includes a graph encoder and a scorer, which encodes the features of the label to be evaluated and the features of the target label to obtain an encoding result, and determines the predicted score of the label to be evaluated according to the encoding result, which may include:

[0083] The features of the label to be evaluated and the features of the target label are encoded through the neural network of the graph encoder to obtain the intermediate vector of the label to be evaluated and the intermediate vector of the target label;

[0084] The neural network of the scorer performs a weighted summation of the intermediate vector of the label to be evaluated and the intermediate vector of the target label to obtain the predicted score of the label to be evaluated.

[0085] By using graph encoders and scorers, the model can combine the capabilities of multiple neural network models. By combining multiple models, we can fully utilize the advantages of different models and improve the performance and generalization ability of the model.

[0086] In some embodiments, when the label graph data to be evaluated is input into the label evaluation model, a graph attention mechanism is used to assign different weights to neighbor nodes, and their attention coefficients can be calculated by the following formula at the kth iteration:

[0087]

[0088] Among them, Sv is the set of neighbors of node v; and are the intermediate vectors of the target node v and the adjacent nodes in the previous iteration (k-1); α k and W k are vectors and matrices that can be learned for linear transformation. When k = 0, the intermediate vector is the feature vector of the node (the 13-dimensional value and the embedding concatenation of the label name), that is, And σ1 is the leaklyReLU function, σ 1 =max(0.2x,x).

[0089] At the kth iteration, the middle vector of the entire neighborhood It can be calculated by the following formula:

[0090]

[0091] Among them, AGG is the aggregation function, is the attention coefficient of node u to node v at the kth iteration.

[0092] The intermediate vector of each node in the next iteration can be updated by the following two steps:

[0093]

[0094] in, and is a learnable weight matrix for linear transformation, and the nonlinear activation function σ 2 =max(0,x) is also a ReLU function.

[0095] After k iterations, we get the intermediate vector of each node, which is also the negative sample in subsequent training.

[0096] In some embodiments, Figure 4 As shown, before inputting the label graph data to be evaluated into the label evaluation model, the method may further include: S401 to S404.

[0097] S401, obtaining training label features and training label association relationships of training labels, where the training label features include training label vectors and training label scores;

[0098] S402, generating training label graph data according to training label features and training label association relationships;

[0099] S403, inputting the training label graph data into the evaluation model, determining the training label intermediate vector through the graph encoder and determining the predicted score of the training label through the scorer, wherein the graph encoder of the evaluation model has a preset second model parameter, and the scorer has a preset third model parameter;

[0100] S404, when the difference between the training label score and the predicted score is greater than a set threshold, the second model and the third model parameters are re-determined according to the difference until the difference is less than the set threshold, thereby obtaining a label evaluation model.

[0101] Integrating the vectors and scores of training labels can make full use of information from different data sources and improve the model's discriminative ability; dynamically adjusting model parameters based on the difference between predicted scores and actual scores enables the model to learn adaptively, improving the model's performance and prediction accuracy.

[0102] In some embodiments, the label evaluation model further includes a discriminator. After re-determining the second model and the third model parameters according to the difference, the method may further include:

[0103] The training label intermediate vector is input into the discriminator, and the predicted classification value of the intermediate vector is determined based on the pre-trained classification parameters;

[0104] The second model parameters are adjusted based on the predicted classification values.

[0105] By inputting the intermediate vector of the training label into the discriminator, the predicted classification value of the intermediate vector can be determined according to the pre-trained classification parameters. The second model parameters are adjusted according to the predicted classification value, which helps the intermediate vector generated by the model gradually approach the true vector.

[0106] In some embodiments, when adjusting the second model parameters according to the predicted classification value, the second model parameters are adjusted according to the predicted classification to determine the loss function, and the loss function of the graph encoder can be:

[0107]

[0108] Among them, L DIS (w; x, z) is the loss function value of the discriminator, w is the parameter of the discriminator, x is the input data from the real data distribution, z is the random noise from the latent space distribution, is the expected value symbol, which means that for all g (z) The random noise z of the distribution is averaged, P g (z) is the latent space distribution used by the generator, usually a standard normal distribution, d w It is the output of the discriminator network for random noise z, usually a probability value, indicating the probability that z z comes from the latent space.

[0109] In some embodiments, Figure 5 As shown, before inputting the training label intermediate vector into the discriminator, the method may further include: S501 to S504.

[0110] S501, obtaining training samples, where the training samples include positive samples and negative samples, wherein the positive samples are training label features selected according to a priori distribution, and the negative samples are training label intermediate vectors;

[0111] S502, inputting the training sample into the discriminator, and determining the predicted classification value of the training sample according to the preset classification parameters;

[0112] S503, re-determining the classification parameters according to the predicted classification values ​​to obtain target classification parameters;

[0113] S504, updating the discriminator according to the target classification parameters.

[0114] In some embodiments, when updating the discriminator according to the target classification parameters, the loss function is determined according to the target classification score to update the discriminator. The loss function of the discriminator can be:

[0115]

[0116] in, is the loss function value, is the parameter of the discriminator, x is the input data, is the expected value symbol, which means averaging all input data from the data distribution, P data (x) is the distribution of real data, is the discriminator function, d w is the output of the discriminator network, which represents the probability that the input data x is the real data.

[0117] By using both positive and negative samples, the model can learn to distinguish features of different categories and enhance its discrimination ability; it helps to reduce overfitting of the model on training data and improve the performance of the model on unknown data.

[0118] In some embodiments, obtaining the associated tag score may include:

[0119] Get the score values ​​of associated tags in multiple dimensions;

[0120] The score values ​​of the associated tags in multiple dimensions are weighted and summed according to the preset weights of the multiple dimensions to obtain the associated tag score.

[0121] By considering the scoring values ​​of multiple dimensions, we can pay comprehensive attention to the different attributes of the label, which helps to deeply analyze the characteristics of the label; at the same time, by setting different weights for the scoring values ​​of different dimensions, we can adjust the importance of each dimension according to business needs, thereby improving the flexibility of scoring.

[0122] In some embodiments, the tag coverage platform score C p The calculation formula can be:

[0123]

[0124] The tag coverage platform score indicates how many platforms the tag has appeared on that provide user tags for business selection. The more platforms a tag appears on, the better the universality of the tag, and the more platforms recognize its use value.

[0125] Label semantic relevance score C r The calculation formula can be:

[0126]

[0127] The tag semantic relevance score represents whether the tag has a certain semantic relevance to the user's final search and browsing behavior, that is, whether the tag can truly reflect the user's target interests and target needs.

[0128] Tag attractiveness score A t The calculation formula can be:

[0129] A t =100×(ee min ) / (e max -e min )

[0130] Among them, e is the click rate of the tag, e min is the minimum value of click rate of all tags, e max The maximum click rate of all tags.

[0131] The tag attractiveness score represents whether the tag can stimulate user interest, increase user engagement, and promote user loyalty. If a tag can attract users, they will click on the tag and use it. Therefore, this score is related to the click-through rate of the tag.

[0132] In some embodiments, the calculation formula of the tag coverage score C can be:

[0133]

[0134] Where N is the number of users covered by the tag, and P% is the proportion of covered users.

[0135] The tag coverage score represents the scale of users associated with the tag. Generally speaking, the fewer users a tag covers, the narrower the user range that can be selected when using the tag, and the more restricted the use of the tag. Therefore, the higher the proportion of users covered by the tag, the higher the score. At the same time, when the absolute number of covered users reaches a certain threshold, it is sufficient for marketing, and the coverage ratio factor is no longer considered. By the same token, when the absolute number of covered users is less than a certain threshold, the use of the tag is of little significance, and the coverage ratio factor is no longer considered.

[0136] The label information score represents the uncertainty of each possible event of the information source. If the label has n values: U 1 …U i …U n , the calculation formula of label information volume H(U) can be:

[0137]

[0138] Among them, U is the label value, n is the number of labels, and p i is the probability of the ith label appearing, and E is the uncertainty of a single value -logp i The statistical mean of .

[0139] Label information score E t The calculation formula can be:

[0140] E t =100×(aa min ) / (a max -a min )

[0141] Among them, a is the amount of information of the label, a min is the minimum value of the information of all labels, a max is the maximum value of the information content of all tags.

[0142] The label timeliness score represents the sensitivity of the label. The higher the label update frequency, the more it can capture the most subtle changes of the user. The better the timeliness, the higher the sensitivity. The main update frequencies of labels are: hourly update, daily update, weekly update, and monthly update. The calculation formula of the label timeliness score F can be:

[0143]

[0144] The data quality score is the quality level of the user tag in the mathematical properties of the data itself; it can include the score of the number of users covered by the tag, the score of the proportion of users covered by the tag, the score of the amount of information in the tag, and the score of the timeliness of the tag.

[0145] The feedback from system users, or the users of the tags, can directly represent the closeness of the tag's connection with the business, and also directly represent the value and significance of the tag to the system users. Moreover, the user's recent feedback is more meaningful than the feedback from earlier times. This is because the value of tags changes over time. The more recent the user feedback is, the more it can represent the value of the tag at this stage. In addition, in order to prevent drastic fluctuations and interference from other factors, previous user feedback data must also be taken into consideration when calculating the score.

[0146] In some embodiments, Figure 6 As shown, suppose the number of times the label is selected and used from date T-14 to date T-1 is F 1 The number of times the label is selected and used from date T-14 to date T-1 is F. 2 , the number of times the tag was selected and used before date T-28 is F 3 , label usage score T 1 The calculation formula can be:

[0147]

[0148] Among them, P 1 F 1 The scoring rule is: when F 1 ≥5, the value is 100; otherwise, the value is F 1 / 5*100. 2 F 2 The scoring rule is: when F 2 ≥5, the value is 100; otherwise, the value is F 2 / 5*100. 3 F 3 The scoring rule is: when F 3 ≥5, the value is 100; otherwise, the value is F 3 / 5*100.

[0149] Assume that the number of searches for a tag from date T-14 to date T-1 is S. 1 , the number of times the tag was searched from date T-14 to date T-1 is S 2 , the number of times the tag was searched before date T-28 is S 3 . Then the calculation formula for the tag search frequency score Se can be:

[0150]

[0151] Among them, Q 1 For S 1 The scoring rule is: when S 1 ≥5, the value is 100; otherwise, the value is S 1 / 5*100.Q 2 For S 2 The scoring rule is: when S 2 ≥5, the value is 100; otherwise, the value is S 2 / 5*100.Q 3 For S 3 The scoring rule is: when S 3 ≥5, the value is 100; otherwise, the value is S 3 / 5*100.

[0152] Assume that the number of times a tag is viewed from date T-14 to date T-1 is W. 1 、The number of times the tag was viewed from date T-14 to date T-1 is W 2 、The number of times the tag was viewed before date T-28 is W 3 . Then the calculation formula for the tag viewing score V can be:

[0153]

[0154] Among them, L 1 W 1 The scoring rule is: when W 1 ≥5, the value is 100; otherwise the value is W 1 / 5*100.L 2 W 2 The scoring rule is: when W 2 ≥5, the value is 100; otherwise the value is W 2 / 5*100.L 3 W 3 The scoring rule is: when W 3 ≥5, the value is 100; otherwise the value is W 3 / 5*100.

[0155] Assume that the number of times a tag is collected from date T-14 to date T-1 is H. 1 、The number of times a tag is collected from date T-14 to date T-1 is H 2 、The number of times the tag has been collected before date T-28 is H 3 . Then the tag collection count score M can be:

[0156]

[0157] Among them, Y 1 H 1 The scoring rule is: when H 1 ≥5, the value is 100; otherwise, the value is H 1 / 5*100. 2 H 2 The scoring rule is: when H 2 ≥5, the value is 100; otherwise, the value is H 2 / 5*100. 3 H 3 The scoring rule is: when H 3 ≥5, the value is 100; otherwise, the value is H 3 / 5*100.

[0158] In some embodiments, Figure 7 As shown, assuming that the response rate ratio of the label after it is used from date T-14 to date T-1 is R 1 , the ratio of the response rate after the label is used from date T-14 to date T-1 is R 2 , the response rate ratio of the label used before date T-28 is R 3 Response rate improvement score K 1 The calculation formula can be:

[0159]

[0160] Among them, O 1 For R 1 The scoring rule is: when R 1 ≥2, the value is 100; otherwise, the value is R 1 / 2*100. 2 For R 2 The scoring rule is: when R 2 ≥2, the value is 100; otherwise, the value is R 2 / 2*100. 3 For R 3 The scoring rule is: when R 3 ≥2, the value is 100; otherwise, the value is R 3 / 2*100.

[0161] Assume that the return on investment ratio of the tag after it is used for 14 days from date T-14 to date T-1 is I 1 The ratio of the return on investment of the tag after it is used from date T-14 to date T-1 is I 2 , the return on investment ratio of the tag after it is used before date T-28 is I 3 . ROI improvement score K 2 The calculation formula can be:

[0162]

[0163] Among them, X 1 For I 1 The scoring rule is: when I 1 ≥2, the value is 100; otherwise, the value is 1 1 / 2*100.X 2 For I 2 The scoring rule is: when I 2 ≥2, the value is 100; otherwise, the value is 1 2 / 2*100.X 3 For I3 The scoring rule is: when R 3 ≥2, the value is 100; otherwise, the value is R 3 / 2*100.

[0164] The larger the ratio of the marketing campaign response rate after using the tag to the marketing campaign response rate and return on investment before using the tag, the more obvious the effect of the tag is, and the better the effect of the tag is. Moreover, the significance of the tag's recent use effect is greater than the significance of the tag's earlier use effect. Because over time, the use effect of the tag will change, and the more recent the use effect, the more representative it is of the current effect. In addition, in order to prevent drastic fluctuations and interference from other factors, the previous use effect should also be taken into consideration in the use effect score.

[0165] Figure 8 A schematic diagram of the structure of a device 800 for evaluating user tags provided in an embodiment of the present application is shown, and the device may include:

[0166] The acquisition module 801 is used to acquire the label features to be evaluated, the associated label features and the label association relationship, the label features to be evaluated include the label vector to be evaluated, and the associated label features include the associated label vector and the associated label score;

[0167] A construction module 802 is used to construct label graph data to be evaluated according to the label features to be evaluated, the associated label features and the label association relationship, wherein the label graph data to be evaluated is composed of a plurality of nodes and edges of nodes, wherein the nodes are label features to be evaluated or associated label features, and the edges of the nodes are label association relationships;

[0168] The evaluation module 803 is used to input the label graph data to be evaluated into the label evaluation model, encode the label features to be evaluated and the target label features and perform weighted summation to obtain the predicted score of the label to be evaluated, and the target label features are the associated label features associated with the label features to be evaluated.

[0169] In some embodiments, the user tag evaluation device 800 may further include:

[0170] The acquisition module 801 is also used to acquire the associated tag vector and the associated tag score;

[0171] A conversion module, used to convert the associated label score into an associated label score vector;

[0172] The concatenation module is used to concatenate the associated label vector and the associated label score vector to obtain the associated label feature of the label to be evaluated.

[0173] In some embodiments, the user tag evaluation device 800 may further include:

[0174] The acquisition module 801 is also used to acquire the tag to be evaluated and the associated tag;

[0175] A word segmentation module is used to use a preset word segmenter to perform word segmentation processing on the to-be-evaluated label and the associated label to obtain a to-be-evaluated label string and an associated label string;

[0176] The conversion module is further used to convert the tag string to be evaluated and the associated tag string into corresponding identifiers;

[0177] The encoding module is used to input the tag identifier to be evaluated and the associated tag identifier into the tag encoding model, and encode the tag identifier to be evaluated and the associated tag identifier respectively through the tag encoding model to obtain the tag vector to be evaluated and the associated tag vector.

[0178] In some embodiments, the user tag evaluation device 800 may further include:

[0179] The acquisition module 801 is further used to acquire the identifier of the training label and the category of the training label;

[0180] The encoding module is further used to input the identifier of the training label into the encoding model, and forward propagate the identifier of the training label through the neural network of the encoding model and the preset first model parameters to obtain the label vector of the training label;

[0181] A calculation module, used for calculating the vector distance between training labels according to the label vectors of the training labels;

[0182] A determination module, used to determine a target loss function value of the encoding model according to the category of the training label and the vector distance between the training labels;

[0183] The determination module is also used to redetermine the first model parameters and update the encoding model according to the target loss function value when the target loss function value is greater than the set threshold, until the target loss function value is less than the set threshold, to obtain the pre-trained first model parameters and the label encoding model updated based on the first model parameters.

[0184] In some embodiments, the determination module is further used to determine the target first loss function value as the target loss function value according to the relationship between the first vector and the first loss function value when the categories of the training labels are the same;

[0185] The determination module is also used to determine the target second loss function as the target loss function value according to the relationship between the second vector distance and the second loss function value when the categories of the training labels are different.

[0186] In some embodiments, the determination module is further used to determine the gradient of the first model parameter of the encoding model according to the target loss function value;

[0187] The determination module is also used to re-determine the first model parameter according to the gradient of the first model parameter.

[0188] In some embodiments, the encoding module is further used to encode the label features to be evaluated and the target label features through the neural network of the graph encoder and the pre-trained second model parameters to obtain the intermediate vector of the label to be evaluated and the intermediate vector of the target label;

[0189] The calculation module is also used to perform weighted summation on the intermediate vector of the label to be evaluated and the intermediate vector of the target label through the neural network of the scorer and the pre-trained third model parameters to obtain the predicted score of the label to be evaluated.

[0190] In some embodiments, the acquisition module 801 is further used to acquire training label features and training label association relationships, where the training label features include training label vectors and training label scores;

[0191] The construction module 802 is further used to generate training label graph data according to the training label features and the training label association relationship;

[0192] The evaluation module 803 is further used to input the label graph data into the evaluation model, determine the training label intermediate vector through the graph encoder and the preset second model parameters, and determine the predicted score of the training label through the scorer and the preset third model parameters;

[0193] The determination module is also used to re-determine the second model and the third model parameters according to the difference between the training label score and the predicted score when it is greater than a set threshold, until the difference is less than the set threshold, thereby obtaining the pre-trained second model parameters, the third model parameters and the label evaluation model updated based on the pre-trained second model parameters and the third model parameters.

[0194] In some embodiments, the user tag evaluation device 800 may further include:

[0195] The determination module is further used to input the training label intermediate vector into the discriminator and determine the predicted classification value of the intermediate vector based on the pre-trained classification parameters;

[0196] An adjustment module is used to adjust the second model parameters according to the predicted classification value.

[0197] In some embodiments, the user tag evaluation device 800 may further include:

[0198] The acquisition module 801 is also used to acquire training samples, where the training samples include positive samples and negative samples, wherein the positive samples are training label features selected according to a priori distribution, and the negative samples are training label intermediate vectors;

[0199] The determination module is further used to input the training sample into the discriminator and determine the predicted classification value of the training sample according to the preset classification parameters;

[0200] The determination module is also used to re-determine the classification parameters according to the predicted classification values ​​to obtain the target classification parameters;

[0201] An update module is used to update the discriminator according to the target classification parameters.

[0202] In some embodiments, the acquisition module 801 is further used to obtain the score values ​​of multiple dimensions of the associated tags;

[0203] The calculation module is also used to perform weighted summation on the score values ​​according to preset weights of multiple dimensions to obtain the associated tag score.

[0204] Figure 8 The various modules in the device shown can be implemented Figure 1 Each step in the above steps can achieve the corresponding technical effect, which will not be repeated here for the sake of brevity.

[0205] Fig. 9 A schematic diagram of the hardware structure of a terminal device provided in an embodiment of the present application is shown.

[0206] The terminal device may include a processor 901 and a memory 902 storing computer program instructions.

[0207] Specifically, the processor 901 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0208] The memory 902 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 902 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In one example, the memory 902 may include a removable or non-removable (or fixed) medium, or the memory 902 is a non-volatile solid-state memory. The memory 902 may be inside or outside the integrated gateway disaster recovery device.

[0209] In one example, the memory 902 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method for evaluating user tags according to the present disclosure.

[0210] The processor 901 reads and executes the computer program instructions stored in the memory 902 to implement Figure 1 The method of user tag evaluation in the illustrated embodiment.

[0211] In one example, the terminal device may further include a communication interface 903 and a bus 904. Fig. 9 As shown, the processor 901, the memory 902, and the communication interface 903 are connected via a bus 904 and communicate with each other.

[0212] The communication interface 903 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0213] Bus 904 includes hardware, software or both, and the parts of terminal equipment are coupled to each other.For example, but not limitation, bus may include accelerated graphics port (Accelerated Graphics Port, AGP) or other graphics bus, enhanced industry standard architecture (Extended Industry Standard Architecture, EISA) bus, front side bus (Front Side Bus, FSB), hypertransmission (Hyper Transport, HT) interconnection, industry standard architecture (IndustryStandard Architecture, ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 904 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0214] In addition, in combination with the user tag evaluation method in the above embodiment, the embodiment of the present application can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any user tag evaluation method in the above embodiment is implemented.

[0215] An embodiment of the present application also provides a computer program product, including a computer program, which implements any one of the user tag evaluation methods in the above embodiments when the computer program is processed and executed.

[0216] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0217] The functional blocks shown in the structural block diagram described above can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), appropriate firmware, plug-in, function card, etc. When implemented in software, the elements of the present application are programs or text segments used to perform the required tasks. The program or text segment can be stored in a machine-readable medium, or transmitted on a transmission medium or communication link via a data signal carried in a carrier. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (Read-Only Memory, ROM), flash memory, erasable read-only memory (Erasable ReadOnly Memory, EROM), floppy disks, compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical discs, hard disks, optical fiber media, radio frequency (Radio Frequency, RF) links, etc. The text segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0218] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0219] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0220] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A method for evaluating user tags, characterized in that: include: Obtaining a tag feature to be evaluated, an associated tag feature, and an associated relationship of the tag to be evaluated, wherein the tag feature to be evaluated includes a tag vector to be evaluated, and the associated tag feature includes an associated tag vector and an associated tag score; Constructing label graph data to be evaluated according to the label features to be evaluated, the associated label features and the label association relationships to be evaluated, wherein the label graph data to be evaluated is composed of a plurality of nodes and edges of the nodes, wherein the nodes are label features to be evaluated or associated label features, and the edges of the nodes are label association relationships to be evaluated; The label graph data to be evaluated is input into the label evaluation model, the label features to be evaluated and the target label features are encoded to obtain an encoding result, and the predicted score of the label to be evaluated is determined according to the encoding result, wherein the target label feature is an associated label feature associated with the label feature to be evaluated.

2. The method for evaluating user tags according to claim 1, characterized in that: Get the associated label features of the label to be evaluated, including: Obtaining an associated tag vector and an associated tag score of the tag to be evaluated; Convert the associated label score into an associated label score vector; The associated label vector and the associated label score vector are concatenated to obtain the associated label feature of the label to be evaluated.

3. The method for evaluating user tags according to claim 1, characterized in that: Get the label vector to be evaluated and the associated label vector, including: Get the tag to be evaluated and the associated tag of the tag to be evaluated; Using a preset word segmenter to perform word segmentation processing on the tag to be evaluated and the associated tag to obtain a tag string to be evaluated and an associated tag string; Convert the tag string to be evaluated and the associated tag string into corresponding identifiers to obtain a tag identifier to be evaluated and an associated tag identifier; The tag identifier to be evaluated and the associated tag identifier are input into the tag encoding model, and the tag identifier to be evaluated and the associated tag identifier are respectively encoded by the tag encoding model to obtain a tag vector to be evaluated and an associated tag vector.

4. The method for evaluating user labels according to claim 3, characterized in that: Before inputting the tag identifier to be evaluated and the associated tag identifier into the tag encoding model, the method further includes: Get the identifier of the training label and the category of the training label; Inputting the identifier of the training label into the encoding model, forward propagating the identifier of the training label through the neural network of the encoding model to obtain a label vector of the training label, wherein the neural network of the encoding model has a preset first model parameter; Calculate the vector distance between the label vectors of the training labels according to the label vectors of the training labels; Determining a target loss function value of an encoding model according to the category of the training label and the vector distance between the label vectors of the training label; When the target loss function value is greater than a set threshold, the first model parameters are re-determined according to the target loss function value and the encoding model is updated until the target loss function value is less than the set threshold, thereby obtaining the label encoding model.

5. The method for evaluating user tags according to claim 4, characterized in that: The step of determining the target loss function value of the encoding model according to the category of the training label and the vector distance between the training labels comprises: When the categories of the training labels are the same, determining a target first loss function value as a target loss function value according to a relationship between the first vector distance and the first loss function value; When the categories of the training labels are different, the target second loss function value is determined as the target loss function value according to the relationship between the second vector distance and the second loss function value.

6. The method for evaluating user tags according to claim 4, characterized in that: The re-determining the first model parameter according to the target loss function value comprises: Determine the gradient of a first model parameter of the encoding model according to the target loss function value; The first model parameter is re-determined according to the gradient of the first model parameter.

7. The method for evaluating user tags according to claim 1, characterized in that: The label evaluation model includes a graph encoder and a scorer, wherein encoding the features of the label to be evaluated and the features of the target label obtains an encoding result, and determining the predicted score of the label to be evaluated according to the encoding result includes: The features of the label to be evaluated and the features of the target label are encoded through the neural network of the graph encoder to obtain the intermediate vector of the label to be evaluated and the intermediate vector of the target label; The neural network of the scorer performs a weighted summation on the intermediate vector of the label to be evaluated and the intermediate vector of the target label to obtain a predicted score of the label to be evaluated.

8. The method for evaluating user tags according to claim 7, characterized in that: Before inputting the to-be-evaluated label graph data into the label evaluation model, the method further includes: Acquire training label features and training label association relationships of training labels, wherein the training label features include training label vectors and training label scores; Generate training label graph data according to the training label features and the training label association relationship; Inputting the training label graph data into an evaluation model, determining the training label intermediate vector through a graph encoder and determining the predicted score of the training label through a scorer, wherein the graph encoder of the evaluation model has a preset second model parameter, and the scorer has a preset third model parameter; When the difference between the training label score and the predicted score is greater than a set threshold, the second model and the third model parameters are re-determined according to the difference until the difference is less than the set threshold, thereby obtaining a label evaluation model.

9. The method for evaluating user tags according to claim 8, characterized in that: The label evaluation model further includes a discriminator. After re-determining the parameters of the second model and the third model according to the difference, the method further includes: Inputting the training label intermediate vector into the discriminator to obtain a predicted classification value of the intermediate vector; The second model parameters are adjusted based on the predicted classification value.

10. The method for evaluating user tags according to claim 9, characterized in that: Before inputting the training label intermediate vector into the discriminator, the method further comprises: Acquire training samples, wherein the training samples include positive samples and negative samples, wherein the positive samples are training label features selected according to a priori distribution, and the negative samples are training label intermediate vectors; Inputting the training sample into the discriminator to obtain a predicted classification value of the training sample; Re-determine the classification parameters according to the predicted classification values ​​to obtain target classification parameters; The discriminator is updated according to the target classification parameters.

11. The method for evaluating user tags according to claim 1, characterized in that: Get the associated tag scores, including: Obtaining score values ​​of the associated tags in multiple dimensions; The score of the associated tag is obtained by weighted summing the score values ​​of the associated tag in multiple dimensions according to the preset weights of the multiple dimensions.

12. A device for evaluating user labels, characterized in that: The device comprises: An acquisition module is used to acquire the tag feature to be evaluated, the associated tag feature and the tag association relationship to be evaluated of the tag to be evaluated, wherein the tag feature to be evaluated includes the tag vector to be evaluated, and the associated tag feature includes the associated tag vector and the associated tag score; A construction module, used to construct label graph data to be evaluated according to the label features to be evaluated, the associated label features and the label association relationship to be evaluated, wherein the label graph data to be evaluated is composed of a plurality of nodes and edges of nodes, wherein the nodes are label features to be evaluated or associated label features, and the edges of the nodes are label association relationships to be evaluated; An evaluation module is used to input the label graph data to be evaluated into a label evaluation model, encode the label features to be evaluated and the target label features to obtain an encoding result, and determine the predicted score of the label to be evaluated based on the encoding result, wherein the target label features are associated label features associated with the label features to be evaluated.

13. A terminal device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for evaluating user tags as described in any one of claims 1 to 11 is implemented.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method for evaluating user tags according to any one of claims 1 to 11 is implemented.

15. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method for user tag evaluation as described in any one of claims 1-11.

Images