Advertisement design effect evaluation method and system based on artificial intelligence

Abstract

The invention discloses an advertisement design effect evaluation method and system based on artificial intelligence, and belongs to the technical field of effect evaluation, and the method comprises the steps of data collection, data preprocessing, advertisement data balance processing, advertisement design effect evaluation model construction and advertisement design effect evaluation. According to the scheme, the number of data needing to be synthesized by each sub-cluster is obtained based on the average density, the importance of the advertisement data in the sub-cluster to which the advertisement data belong and the similarity of the advertisement data with other sub-clusters are comprehensively considered, the selection probability is obtained, new advertisement data is synthesized based on the cluster center and the random adjustment factor, and the balanced data set is constructed according to the spatial distribution entropy; the method comprises the following steps: extracting potential information by using one-dimensional convolution, adjusting an adjacent matrix, adding a self-connecting edge, calculating spatial information, carrying out feature updating based on a convolution layer, obtaining an output label, designing a self-adaptive division proportion to divide individual positions, and respectively updating the individual positions by using a sine and cosine algorithm. And the accuracy and practicability of advertisement design effect evaluation are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of effect evaluation, and specifically refers to an advertising design effect evaluation method and system based on artificial intelligence. Background Art

[0002] The advertising design effect evaluation method uses advanced artificial intelligence technology to collect and analyze a large amount of advertising-related data to accurately measure the effect of advertising design, helping designers and marketing teams understand the advantages and disadvantages of advertising. However, the existing advertising design effect evaluation method has the problem of unbalanced advertising data categories and uneven distribution, resulting in inaccurate effect evaluation of minority ads; the existing advertising design effect evaluation method has the problem of complex and diverse advertising data, complex nonlinear relationships, and difficulty in effectively extracting features, resulting in inaccurate advertising design effect evaluation and insufficient adaptability. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an advertising design effect evaluation method and system based on artificial intelligence. In view of the problem that the advertising data categories are unbalanced and unevenly distributed in the existing advertising design effect evaluation methods, resulting in inaccurate effect evaluation of minority advertisements, this scheme clusters the minority class data sets to better characterize the local distribution characteristics of the advertising data; based on the average density, the number of data required to be synthesized for each sub-cluster is obtained, and new data is synthesized in a targeted manner according to the data sparsity of different sub-clusters; the importance of the advertising data in the sub-cluster to which it belongs and the similarity with other sub-clusters are comprehensively considered to obtain the selection probability of each advertising data in each sub-cluster, thereby improving the quality of the synthesized data; new advertising data is synthesized based on the cluster center and the random adjustment factor, and a balanced data set is constructed according to the spatial distribution entropy, which alleviates the problem of data imbalance and contributes to more comprehensive and objective advertising. Advertising design effect evaluation; In view of the problems that the existing advertising design effect evaluation methods have complex and diverse advertising data, complex nonlinear relationships, and it is difficult to effectively extract features, resulting in inaccurate advertising design effect evaluation and insufficient adaptability, this solution uses one-dimensional convolution to extract potential information, adjusts the adjacency matrix to add self-connected edges, so that nodes can aggregate neighbor features and retain their own feature representations; calculates spatial information, updates features based on the convolution layer, obtains output labels through the fully connected layer and softmax, and obtains richer and more effective advertising data features; uses individual positions as representatives of model parameters, designs adaptive division ratios to divide individual positions, uses sine-cosine algorithms to update individual positions separately, and constructs an advertising design effect evaluation model based on the search for optimal parameters, thereby improving the accuracy and practicality of advertising design effect evaluation and enhancing adaptability and generalization capabilities to different advertising data.

[0004] The technical solution adopted by the present invention is as follows: The artificial intelligence-based advertising design effect evaluation method provided by the present invention comprises the following steps:

[0005] Step S1: data collection;

[0006] Step S2: data preprocessing;

[0007] Step S3: advertising data balancing processing;

[0008] Step S4: constructing an advertising design effect evaluation model;

[0009] Step S5: Evaluate the advertising design effect.

[0010] Furthermore, in step S1, the data collection is to collect historical advertising data; the historical advertising data includes delivery data, user behavior data, user feedback data, product sales growth data and effect level; the delivery data includes delivery platform data and delivery time data; the user behavior data includes click-through rate, browsing time, conversion rate and number of shares; the user feedback data includes comment data and rating data; the effect level includes excellent, good, average and poor, and the effect level is used as a data label.

[0011] Furthermore, in step S2, the data preprocessing is to perform data cleaning, data conversion, data normalization and data set construction on the collected data; data cleaning includes processing noise, outliers, duplicate values ​​and missing values; data conversion is to convert the data into vector form; data normalization is to unify the data range based on the maximum and minimum normalization method; data set construction is to construct an advertising data set based on the data after data cleaning, data conversion and data normalization, and divide the advertising data set into a minority class data set and a majority class data set.

[0012] Furthermore, in step S3, the advertising data balancing process specifically includes the following steps:

[0013] Step S31: clustering, using the density peak clustering algorithm to cluster the minority class data set to obtain F sub-clusters;

[0014] Step S32: Calculate the required synthesis amount, and obtain the amount of data required to be synthesized for each sub-cluster based on the average density. The formula used is as follows:

[0015] ;

[0016] ;

[0017] In the formula, h f is the required synthesis amount of the f-th sub-cluster, and are the average densities of the f-th and a-th sub-clusters, respectively, and S f and S a are the number of advertising data in the fth and ath sub-clusters, respectively. f and a are sub-cluster indexes. S1 and S2 are the number of advertising data in the majority class dataset and the minority class dataset, respectively. and are the i-th advertisement data in the f-th sub-cluster and the j-th advertisement data in the advertisement dataset G, respectively. i and j are advertisement data indexes. yes and The Euclidean distance between c is the cutoff distance; is an indicator function, if ,but ,otherwise, ;

[0018] Step S33: Calculate the selection probability, comprehensively consider the importance of the advertisement data in the sub-cluster to which it belongs and its similarity with other sub-clusters, and obtain the selection probability of each advertisement data in each sub-cluster. The formula used is as follows:

[0019] ;

[0020] In the formula, is the selection probability of the i-th advertisement data in the f-th sub-cluster, ρ i is the local density of the i-th advertisement data in the f-th subcluster, , c a is the cluster center of the a-th subcluster, yes and c a The Euclidean distance between

[0021] Step S34: synthesizing data, including the following steps:

[0022] Step S341: Calculate the average Euclidean distance between all advertisement data and the cluster center in each sub-cluster as the synthesis radius of the sub-cluster, and then randomly select L advertisement data from the sub-cluster based on the selection probability, obtain L vectors from the cluster center to the L advertisement data, design a random adjustment factor, and synthesize new advertisement data based on the cluster center and the random adjustment factor. The formula used is as follows:

[0023] ;

[0024] In the formula, is the newly synthesized advertisement data of the fth sub-cluster, c f is the cluster center of the f-th subcluster, r is the random adjustment factor, θ is the smoothing factor, and b f is the composite radius of the f-th subcluster, is r subject to mean b f , the variance is The normal distribution of is the g-th vector of the f-th subcluster, where g is the vector index;

[0025] Step S342: Repeat step S341 until the amount of newly synthesized advertisement data of each sub-cluster is equal to the required amount of synthesis, and construct a synthetic data set based on all the newly synthesized advertisement data;

[0026] Step S35: construct a balanced data set, calculate the spatial distribution entropy E1 of the advertising data set, and the spatial distribution entropy E2 of the union of the synthetic data set and the advertising data set. If the difference E2-E1 between the spatial distribution entropies of the two is less than zero, the union of the synthetic data set and the advertising data set is used as the balanced data set, and 70% of the advertising data are randomly selected from the balanced data set to construct the training data set, and the remaining 30% of the advertising data are used to construct the test data set; otherwise, the synthetic data set is discarded and the process returns to step S34; the formula used to calculate the spatial distribution entropy is as follows:

[0027] ;

[0028] Where E is the spatial distribution entropy, M is the number of advertising data, and They are centered on the i-th advertisement data and cutoff distance d c is the number of majority class data and minority class data within the neighborhood formed by the radius.

[0029] Furthermore, in step S4, the construction of the advertisement design effect evaluation model specifically includes the following steps:

[0030] Step S41: Initialization, pre-set the distance threshold, take each advertisement data in the training data set as a node, the feature vector of the node is the corresponding advertisement data, and obtain a B×O feature matrix Z; if the Euclidean distance between two advertisement data is less than the distance threshold, there is an edge between the corresponding two nodes, and the weight of the edge is 1; otherwise, there is no edge between the corresponding two nodes, and the weight of the edge is 0; construct a B×B adjacency matrix A based on the edge weight; define a B×C adaptive two-dimensional node embedding matrix Q; where B is the number of nodes, O is the feature dimension, and C is the embedding dimension;

[0031] Step S42: One-dimensional convolution processing, using a one-dimensional convolution layer with a kernel size of 5 to process the feature matrix Z and extract the potential information in the feature matrix Z. The formula used is as follows:

[0032] ;

[0033] In the formula, is the potential information in the feature matrix Z, is a one-dimensional convolution function, is the Hadamard product, α1 is the first control factor;

[0034] Step S43: Adjust the adjacency matrix and add self-connection edges to each node. The formula used is as follows:

[0035] ;

[0036] Where I is the identity matrix, λ is the weight, is the adjusted adjacency matrix;

[0037] Step S44: Calculate the spatial information using the following formula:

[0038] ;

[0039] In the formula, is the spatial information, T is the transposition operation, and α2 is the second control factor;

[0040] Step S45: Feature update, initial feature vector , the feature vector is updated based on the U-layer convolutional layer. The formula used is as follows:

[0041] ;

[0042] In the formula, and are the feature vectors output by the u+1th and uth convolutional layers, respectively, u is the convolutional layer index, is the weight matrix of the u-th convolutional layer, σ(·) is the activation function, yes The degree matrix of

[0043] Step S46: Evaluation, the feature vector output by the U-th convolutional layer Input to the fully connected layer, and then use the softmax activation function to get the probability distribution corresponding to each data label, and take the data label with the highest probability as the final output label. The formula used is as follows:

[0044] ;

[0045] Where Y is the probability distribution, and They are the weight matrix and bias vector of the fully connected layer respectively;

[0046] Step S47: Model parameter search, including the following steps:

[0047] Step S471: Initial position, establish a parameter search space for the first control factor α1, the weight λ and the second control factor α2, randomly initialize K individual positions in the parameter search space, use the individual positions as representatives of model parameters, and use the cross entropy loss function of the advertising design effect evaluation model trained based on the individual positions for the test data set as the individual fitness value;

[0048] Step S472: Individual position division: set a marking factor for each individual position and initialize the value to 0, arrange the individual positions in order from small to large based on the size of the fitness value, design an adaptive division ratio v, and divide the previous The individual positions are divided into excellent individual positions, and the value of the corresponding marking factor is changed to 1. The individual positions are divided into general individual positions, and the corresponding marking factor value is changed to 0; the formula used to calculate the adaptive division ratio is as follows:

[0049] ;

[0050] In the formula, t and t max are the search index and the maximum search number, m is the scale factor, β is the perturbation factor, rand(0,1) generates a random number in the range of (0,1), tan(•) is the tangent function, is rounded down;

[0051] Step S473: Position update, using the sine-cosine algorithm to update the excellent individual position and the general individual position respectively; the formula used is as follows:

[0052] ;

[0053] In the formula, and are the positions of the wth individual in the t+1th and tth searches, respectively. w is the individual index. sin(•) and cos(•) are the sine and cosine functions, respectively. q1 and q2 are random numbers uniformly distributed in the range of (0, 2π) and (0, 2), respectively. is the global optimal position at the tth search, δ w is the labeling factor of the w-th individual position, is the step size factor;

[0054] Step S474: Determine the optimal parameters, pre-set the fitness threshold, and update the individual fitness value. When the fitness value of the global optimal position is less than the fitness threshold, the parameters corresponding to the global optimal position are the optimal parameters. Based on the optimal parameters, build an advertising design effect evaluation model to complete the parameter search; otherwise, if the maximum number of searches is reached, go to step S471 to reinitialize; otherwise, add 1 to the number of searches and return to step S472 to continue the iterative search.

[0055] Furthermore, in step S5, the advertising design effect evaluation is to collect real-time advertising data; the real-time advertising data includes delivery data, user behavior data, user feedback data and product sales growth data. After pre-processing, the real-time advertising data is input into the advertising design effect evaluation model constructed based on optimal parameters for processing, and based on the output label, the effect level corresponding to the real-time advertising data is obtained.

[0056] The artificial intelligence-based advertising design effect evaluation system provided by the present invention comprises a data acquisition module, a data preprocessing module, an advertising data balance processing module, a module for constructing an advertising design effect evaluation model, and an advertising design effect evaluation module;

[0057] The data collection module collects historical advertising data and sends the data to the data preprocessing module;

[0058] The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data, and sends the data to the advertising data balancing processing module;

[0059] The advertising data balancing processing module clusters the minority class data set, obtains the amount of data that needs to be synthesized for each sub-cluster based on the average density, comprehensively considers the importance of the advertising data in the sub-cluster to which it belongs and the similarity with other sub-clusters, obtains the selection probability of each advertising data in each sub-cluster, synthesizes new advertising data based on the cluster center and the random adjustment factor, constructs a balanced data set according to the spatial distribution entropy, and sends the data to the module for constructing the advertising design effect evaluation model;

[0060] The module for constructing the advertising design effect evaluation model uses one-dimensional convolution to extract potential information, adjusts the adjacency matrix to add self-connected edges, calculates spatial information, updates features based on the convolution layer, obtains output labels, designs an adaptive division ratio to divide individual positions, uses the sine-cosine algorithm to update individual positions respectively, constructs an advertising design effect evaluation model based on optimal parameters, and sends data to the advertising design effect evaluation module;

[0061] The advertisement design effect evaluation module pre-processes the real-time advertisement data and then inputs the pre-processed data into an advertisement design effect evaluation model constructed based on optimal parameters to obtain the effect level corresponding to the real-time advertisement data.

[0062] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0063] (1) In view of the problem that the existing advertising design effect evaluation methods have imbalanced advertising data categories and uneven distribution, resulting in inaccurate effect evaluation of minority advertisements, this scheme uses the density peak clustering algorithm to cluster the minority class data set to better characterize the local distribution characteristics of the advertising data; based on the average density, the number of data required to be synthesized for each sub-cluster is obtained, and new data is synthesized in a targeted manner according to the data sparsity of different sub-clusters; the importance of the advertising data in the sub-cluster to which it belongs and the similarity with other sub-clusters are comprehensively considered to obtain the selection probability of each advertising data in each sub-cluster, and the rich information of the original advertising data is fully utilized when synthesizing new data, thereby improving the quality of the synthesized data; based on the cluster center and the random adjustment factor, new advertising data is synthesized, and a balanced data set is constructed according to the spatial distribution entropy to generate new data with a distribution similar to that of the original advertising data, further alleviating the data imbalance problem, which is conducive to a more comprehensive and objective advertising design effect evaluation and improving the evaluation accuracy.

[0064] (2) In view of the problems of existing advertising design effect evaluation methods, such as complex and diverse advertising data with complex nonlinear relationships, it is difficult to effectively extract features, resulting in inaccurate advertising design effect evaluation and insufficient adaptability. This solution uses one-dimensional convolution to extract potential information and improve the ability to extract advertising features; adjusts the adjacency matrix to add self-connected edges so that nodes can aggregate neighbor features and retain their own feature representations, thereby enhancing the expressive power of the model; calculates spatial information, updates features based on the convolutional layer, obtains output labels through the fully connected layer and softmax, obtains richer and more effective advertising data features, and accurately outputs evaluation results; uses individual positions as representatives of model parameters, designs adaptive partitioning ratios to divide individual positions, uses the sine-cosine algorithm to update individual positions separately, and constructs an advertising design effect evaluation model based on the search for optimal parameters, thereby improving the accuracy and practicality of advertising design effect evaluation and enhancing adaptability and generalization to different advertising data. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A schematic diagram of the flow chart of the advertising design effect evaluation method based on artificial intelligence provided by the present invention;

[0066] Figure 2 A schematic diagram of an artificial intelligence-based advertising design effect evaluation system provided by the present invention;

[0067] Figure 3 is a schematic diagram of the process of step S3;

[0068] Figure 4It is a schematic diagram of the process of step S4.

[0069] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0071] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0072] Example 1, see Figure 1 The invention provides an artificial intelligence-based advertising design effect evaluation method, which comprises the following steps:

[0073] Step S1: data collection, collecting historical advertising data;

[0074] Step S2: Data preprocessing, data cleaning, data conversion, data normalization and data set construction for the collected data;

[0075] Step S3: Balancing advertising data: clustering the minority class data set, obtaining the amount of data that needs to be synthesized for each sub-cluster based on the average density, comprehensively considering the importance of the advertising data in the sub-cluster to which it belongs and its similarity with other sub-clusters, obtaining the selection probability of each advertising data in each sub-cluster, synthesizing new advertising data based on the cluster center and random adjustment factor, and constructing a balanced data set based on the spatial distribution entropy;

[0076] Step S4: construct an advertising design effect evaluation model, use one-dimensional convolution to extract potential information, adjust the adjacency matrix to add self-connected edges, calculate spatial information, update features based on the convolution layer, obtain output labels, design an adaptive division ratio to divide individual positions, use the sine-cosine algorithm to update individual positions respectively, and construct an advertising design effect evaluation model based on the optimal parameters;

[0077] Step S5: Advertising design effect evaluation: After pre-processing the real-time advertising data, input it into the advertising design effect evaluation model constructed based on the optimal parameters for processing, and obtain the effect level corresponding to the real-time advertising data.

[0078] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the historical advertising data includes delivery data, user behavior data, user feedback data, product sales growth data and effect level; the delivery data includes delivery platform data and delivery time data; the user behavior data includes click-through rate, browsing time, conversion rate and number of shares; the user feedback data includes comment data and rating data; the effect level includes excellent, good, average and poor, and the effect level is used as a data label.

[0079] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, data cleaning includes processing noise, outliers, duplicate values ​​and missing values; data conversion is to convert data into vector form; data normalization is to unify the data range based on the maximum and minimum normalization method; constructing a data set is to construct an advertising data set based on the data after data cleaning, data conversion and data normalization, and divide the advertising data set into a minority class data set and a majority class data set.

[0080] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the advertisement data balancing process specifically includes the following steps:

[0081] Step S31: clustering, using the density peak clustering algorithm to cluster the minority class data set to obtain F sub-clusters;

[0082] Step S32: Calculate the required synthesis amount, and obtain the amount of data required to be synthesized for each sub-cluster based on the average density. The formula used is as follows:

[0083] ;

[0084] ;

[0085] In the formula, h f is the required synthesis amount of the f-th sub-cluster, and are the average densities of the f-th and a-th sub-clusters, respectively, and S f and S a are the number of advertising data in the fth and ath sub-clusters, respectively. f and a are sub-cluster indexes. S1 and S2 are the number of advertising data in the majority class dataset and the minority class dataset, respectively. and are the i-th advertisement data in the f-th sub-cluster and the j-th advertisement data in the advertisement dataset G, respectively. i and j are advertisement data indexes. yes and The Euclidean distance between c is the cutoff distance; is an indicator function, if ,but ,otherwise, ;

[0086] Step S33: Calculate the selection probability, comprehensively consider the importance of the advertisement data in the sub-cluster to which it belongs and its similarity with other sub-clusters, and obtain the selection probability of each advertisement data in each sub-cluster. The formula used is as follows:

[0087] ;

[0088] In the formula, is the selection probability of the i-th advertisement data in the f-th sub-cluster, ρ i is the local density of the i-th advertisement data in the f-th subcluster, , c a is the cluster center of the a-th subcluster, yes and c a The Euclidean distance between

[0089] Step S34: synthesizing data, including the following steps:

[0090] Step S341: Calculate the average Euclidean distance between all advertisement data and the cluster center in each sub-cluster as the synthesis radius of the sub-cluster, and then randomly select L advertisement data from the sub-cluster based on the selection probability, obtain L vectors from the cluster center to the L advertisement data, design a random adjustment factor, and synthesize new advertisement data based on the cluster center and the random adjustment factor. The formula used is as follows:

[0091] ;

[0092] In the formula, is the newly synthesized advertisement data of the fth sub-cluster, c f is the cluster center of the f-th subcluster, r is the random adjustment factor, θ is the smoothing factor in the range of (0, 1), and b f is the composite radius of the f-th subcluster, is r subject to mean b f , the variance is The normal distribution of is the g-th vector of the f-th subcluster, where g is the vector index;

[0093] Step S342: Repeat step S341 until the amount of newly synthesized advertisement data of each sub-cluster is equal to the required amount of synthesis, and construct a synthetic data set based on all the newly synthesized advertisement data;

[0094] Step S35: construct a balanced data set, calculate the spatial distribution entropy E1 of the advertising data set, and the spatial distribution entropy E2 of the union of the synthetic data set and the advertising data set. If the difference E2-E1 between the spatial distribution entropies of the two is less than zero, the union of the synthetic data set and the advertising data set is used as the balanced data set, and 70% of the advertising data are randomly selected from the balanced data set to construct the training data set, and the remaining 30% of the advertising data are used to construct the test data set; otherwise, the synthetic data set is discarded and the process returns to step S34; the formula used to calculate the spatial distribution entropy is as follows:

[0095] ;

[0096] Where E is the spatial distribution entropy, M is the number of advertising data, and They are centered on the i-th advertisement data and cutoff distance d c is the number of majority class data and minority class data within the neighborhood formed by the radius.

[0097] By performing the above operations, in view of the problem that the existing advertising design effect evaluation methods have imbalanced advertising data categories and uneven distribution, resulting in inaccurate effect evaluation of minority advertisements, this scheme uses the density peak clustering algorithm to cluster the minority class data set to better characterize the local distribution characteristics of the advertising data; based on the average density, the number of data required to be synthesized for each sub-cluster is obtained, and new data is synthesized in a targeted manner according to the data sparsity of different sub-clusters; the importance of the advertising data in the sub-cluster to which it belongs and the similarity with other sub-clusters are comprehensively considered to obtain the selection probability of each advertising data in each sub-cluster, and the rich information of the original advertising data is fully utilized when synthesizing new data, thereby improving the quality of the synthesized data; based on the cluster center and the random adjustment factor, new advertising data is synthesized, and a balanced data set is constructed according to the spatial distribution entropy, and new data with a distribution similar to the original advertising data is generated, which further alleviates the data imbalance problem, helps to conduct a more comprehensive and objective advertising design effect evaluation, and improves the evaluation accuracy.

[0098] Example 5, see Figure 1 and Figure 4 Based on the above embodiment, in step S4, building an advertisement design effect evaluation model specifically includes the following steps:

[0099] Step S41: Initialization, pre-set the distance threshold, take each advertisement data in the training data set as a node, the feature vector of the node is the corresponding advertisement data, and obtain a B×O feature matrix Z; if the Euclidean distance between two advertisement data is less than the distance threshold, there is an edge between the corresponding two nodes, and the weight of the edge is 1; otherwise, there is no edge between the corresponding two nodes, and the weight of the edge is 0; construct a B×B adjacency matrix A based on the edge weight; define a B×C adaptive two-dimensional node embedding matrix Q; where B is the number of nodes, O is the feature dimension, and C is the embedding dimension;

[0100] Step S42: One-dimensional convolution processing, using a one-dimensional convolution layer with a kernel size of 5 to process the feature matrix Z and extract the potential information in the feature matrix Z. The formula used is as follows:

[0101] ;

[0102] In the formula, is the potential information in the feature matrix Z, is a one-dimensional convolution function, is the Hadamard product, α1 is the first control factor in the range of (0, 1);

[0103] Step S43: Adjust the adjacency matrix and add self-connection edges to each node so that the node can not only aggregate the features of neighboring nodes during graph convolution, but also retain its own feature representation. The formula used is as follows:

[0104] ;

[0105] Where I is the identity matrix, λ is the weight in the range of (0, 1), is the adjusted adjacency matrix;

[0106] Step S44: Calculate the spatial information using the following formula:

[0107] ;

[0108] In the formula, is the spatial information, T is the transposition operation, and α2 is the second control factor in the range of (0, 1);

[0109] Step S45: Feature update, initial feature vector , the feature vector is updated based on the U-layer convolutional layer. The formula used is as follows:

[0110] ;

[0111] In the formula, and are the feature vectors output by the u+1th and uth convolutional layers, respectively, u is the convolutional layer index, is the weight matrix of the u-th convolutional layer, σ(·) is the activation function, yes The degree matrix of

[0112] Step S46: Evaluation, the feature vector output by the U-th convolutional layer Input to the fully connected layer, and then use the softmax activation function to get the probability distribution corresponding to each data label, and take the data label with the highest probability as the final output label. The formula used is as follows:

[0113] ;

[0114] Where Y is the probability distribution, and They are the weight matrix and bias vector of the fully connected layer respectively;

[0115] Step S47: Model parameter search, including the following steps:

[0116] Step S471: Initial position, establish a parameter search space for the first control factor α1, the weight λ and the second control factor α2, randomly initialize K individual positions in the parameter search space, use the individual positions as representatives of model parameters, and use the cross entropy loss function of the advertising design effect evaluation model trained based on the individual positions for the test data set as the individual fitness value;

[0117] Step S472: Individual position division: set a marking factor for each individual position and initialize the value to 0, arrange the individual positions in order from small to large based on the size of the fitness value, design an adaptive division ratio v, and divide the previous The individual positions are divided into excellent individual positions, and the value of the corresponding marking factor is changed to 1. The individual positions are divided into general individual positions, and the corresponding marking factor value is changed to 0; the formula used to calculate the adaptive division ratio is as follows:

[0118] ;

[0119] In the formula, t and t max are the search count index and the maximum search count, respectively. m is the scale factor in the range of (0, 1). β is the perturbation factor in the range of (0.1, 0.5). rand(0, 1) generates a random number in the range of (0, 1). tan(•) is the tangent function. is rounded down;

[0120] Step S473: Position update, using the sine-cosine algorithm to update the excellent individual position and the general individual position respectively; the formula used is as follows:

[0121] ;

[0122] In the formula, and are the positions of the wth individual in the t+1th and tth searches, respectively. w is the individual index. sin(•) and cos(•) are the sine and cosine functions, respectively. q1 and q2 are random numbers uniformly distributed in the range of (0, 2π) and (0, 2), respectively. is the global optimal position at the tth search. The global optimal position is the individual position with the smallest fitness value. w is the labeling factor of the w-th individual position, is a step factor in the range (0.01, 0.1);

[0123] Step S474: Determine the optimal parameters, pre-set the fitness threshold, and update the individual fitness value. When the fitness value of the global optimal position is less than the fitness threshold, the parameters corresponding to the global optimal position are the optimal parameters. Based on the optimal parameters, build an advertising design effect evaluation model to complete the parameter search; otherwise, if the maximum number of searches is reached, go to step S471 to reinitialize; otherwise, add 1 to the number of searches and return to step S472 to continue the iterative search.

[0124] By performing the above operations, the existing advertising design effect evaluation methods have the problem that the advertising data is complex and diverse, has complex nonlinear relationships, and is difficult to effectively extract features, resulting in inaccurate advertising design effect evaluation and insufficient adaptability. This solution uses one-dimensional convolution to extract potential information and improve the ability to extract advertising features; adjusts the adjacency matrix to add self-connected edges so that nodes can aggregate neighbor features and retain their own feature representations, thereby enhancing the expressive power of the model; calculates spatial information, updates features based on the convolution layer, obtains output labels through the fully connected layer and softmax, obtains richer and more effective advertising data features, and accurately outputs evaluation results; uses individual positions as representatives of model parameters, designs adaptive partitioning ratios to divide individual positions, uses sine-cosine algorithms to update individual positions separately, and constructs an advertising design effect evaluation model based on the search for optimal parameters, thereby improving the accuracy and practicality of advertising design effect evaluation and enhancing adaptability and generalization to different advertising data.

[0125] Example 6, see Figure 1This embodiment is based on the above embodiment. In step S5, the advertisement design effect evaluation is to collect real-time advertisement data; the real-time advertisement data includes delivery data, user behavior data, user feedback data and product sales growth data. After preprocessing the real-time advertisement data, it is input into the advertisement design effect evaluation model constructed based on the optimal parameters for processing, and based on the output label, the effect level corresponding to the real-time advertisement data is obtained.

[0126] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, the artificial intelligence-based advertising design effect evaluation system provided by the present invention includes a data acquisition module, a data preprocessing module, an advertising data balance processing module, a module for building an advertising design effect evaluation model and an advertising design effect evaluation module;

[0127] The data collection module collects historical advertising data and sends the data to the data preprocessing module;

[0128] The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data, and sends the data to the advertising data balancing processing module;

[0129] The advertising data balancing processing module clusters the minority class data set, obtains the amount of data that needs to be synthesized for each sub-cluster based on the average density, comprehensively considers the importance of the advertising data in the sub-cluster to which it belongs and the similarity with other sub-clusters, obtains the selection probability of each advertising data in each sub-cluster, synthesizes new advertising data based on the cluster center and the random adjustment factor, constructs a balanced data set according to the spatial distribution entropy, and sends the data to the module for constructing the advertising design effect evaluation model;

[0130] The module for constructing the advertising design effect evaluation model uses one-dimensional convolution to extract potential information, adjusts the adjacency matrix to add self-connected edges, calculates spatial information, updates features based on the convolution layer, obtains output labels, designs an adaptive division ratio to divide individual positions, uses the sine-cosine algorithm to update individual positions respectively, constructs an advertising design effect evaluation model based on optimal parameters, and sends data to the advertising design effect evaluation module;

[0131] The advertisement design effect evaluation module pre-processes the real-time advertisement data and then inputs the pre-processed data into an advertisement design effect evaluation model constructed based on optimal parameters to obtain the effect level corresponding to the real-time advertisement data.

[0132] 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.

[0133] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0134] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An advertising design effect evaluation method based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: data collection, collecting historical advertising data; Step S2: data preprocessing; Step S3: advertising data balancing processing; Step S4: constructing an advertising design effect evaluation model; Step S5: advertising design effect evaluation, after pre-processing the real-time advertising data, input it into the advertising design effect evaluation model constructed based on the optimal parameters for processing, and obtain the effect level corresponding to the real-time advertising data; Step S4 includes step S472: individual position division, setting a marking factor for each individual position and initializing the value to 0, arranging the individual positions in order from small to large based on the size of the fitness value, designing an adaptive division ratio v, and dividing the previous The individual positions are divided into excellent individual positions, and the value of the corresponding marking factor is changed to 1. The individual positions are divided into general individual positions, and the corresponding marking factor value is changed to 0; the formula used to calculate the adaptive division ratio is as follows: ； In the formula, t and t max are the search count index and the maximum search count, m is the scale factor, β is the perturbation factor, rand(0,1) generates a random number in the range (0,1), tan(•) is the tangent function, K is the number of individual positions, It is rounded down.

2. The advertising design effect evaluation method based on artificial intelligence according to claim 1 is characterized by: In step S3, the advertisement data balancing process specifically includes the following steps: Step S31: clustering, using the density peak clustering algorithm to cluster the minority class data set to obtain F sub-clusters; Step S32: Calculate the required synthesis amount, and obtain the amount of data required to be synthesized for each sub-cluster based on the average density. The formula used is as follows: ； ； In the formula, h f is the required synthesis amount of the f-th sub-cluster, and are the average densities of the f-th and a-th sub-clusters, respectively, and S f and S a are the number of advertising data in the fth and ath sub-clusters, respectively. f and a are sub-cluster indexes. S1 and S2 are the number of advertising data in the majority class dataset and the minority class dataset, respectively. and are the i-th advertisement data in the f-th sub-cluster and the j-th advertisement data in the advertisement dataset G, respectively. i and j are advertisement data indexes. yes and The Euclidean distance between c is the cutoff distance; is the indicator function; Step S33: Calculate the selection probability, comprehensively consider the importance of the advertisement data in the sub-cluster to which it belongs and its similarity with other sub-clusters, and obtain the selection probability of each advertisement data in each sub-cluster. The formula used is as follows: ； In the formula, is the selection probability of the i-th advertisement data in the f-th sub-cluster, ρ i is the local density of the i-th advertisement data in the f-th subcluster, , c a is the cluster center of the a-th subcluster, yes and c a The Euclidean distance between Step S34: synthesizing data; Step S35: construct a balanced data set, calculate the spatial distribution entropy E1 of the advertising data set, and the spatial distribution entropy E2 of the union of the synthetic data set and the advertising data set. If the difference E2-E1 between the spatial distribution entropies of the two is less than zero, the union of the synthetic data set and the advertising data set is used as the balanced data set, and 70% of the advertising data are randomly selected from the balanced data set to construct the training data set, and the remaining 30% of the advertising data are used to construct the test data set; otherwise, the synthetic data set is discarded and the process returns to step S34; the formula used to calculate the spatial distribution entropy is as follows: ； Where E is the spatial distribution entropy, M is the number of advertising data, and They are centered on the i-th advertisement data and cutoff distance d c is the number of majority class data and minority class data within the neighborhood formed by the radius.

3. The advertising design effect evaluation method based on artificial intelligence according to claim 2 is characterized in that: In step S34, the synthesized data specifically includes the following steps: Step S341: Calculate the average Euclidean distance between all advertisement data and the cluster center in each sub-cluster as the synthesis radius of the sub-cluster, and then randomly select L advertisement data from the sub-cluster based on the selection probability, obtain L vectors from the cluster center to the L advertisement data, design a random adjustment factor, and synthesize new advertisement data based on the cluster center and the random adjustment factor. The formula used is as follows: ； In the formula, is the newly synthesized advertisement data of the fth sub-cluster, c f is the cluster center of the f-th subcluster, r is the random adjustment factor, θ is the smoothing factor, and b f is the composite radius of the f-th subcluster, is r subject to mean b f , the variance is The normal distribution of is the g-th vector of the f-th subcluster, where g is the vector index; Step S342: Repeat step S341 until the amount of newly synthesized advertisement data of each sub-cluster is equal to the required synthesis amount, and construct a synthetic data set based on all the newly synthesized advertisement data.

4. The advertising design effect evaluation method based on artificial intelligence according to claim 1 is characterized in that: In step S4, the construction of the advertisement design effect evaluation model specifically includes the following steps: Step S41: Initialization, pre-set the distance threshold, take each advertisement data in the training data set as a node, the feature vector of the node is the corresponding advertisement data, and obtain a B×O feature matrix Z; if the Euclidean distance between two advertisement data is less than the distance threshold, there is an edge between the corresponding two nodes, and the weight of the edge is 1; otherwise, there is no edge between the corresponding two nodes, and the weight of the edge is 0; construct a B×B adjacency matrix A based on the edge weight; define a B×C adaptive two-dimensional node embedding matrix Q; where B is the number of nodes, O is the feature dimension, and C is the embedding dimension; Step S42: One-dimensional convolution processing, using a one-dimensional convolution layer with a kernel size of 5 to process the feature matrix Z and extract the potential information in the feature matrix Z. The formula used is as follows: ； In the formula, is the potential information in the feature matrix Z, is a one-dimensional convolution function, is the Hadamard product, α1 is the first control factor; Step S43: Adjust the adjacency matrix and add self-connection edges to each node. The formula used is as follows: ； Where I is the identity matrix, λ is the weight, is the adjusted adjacency matrix; Step S44: Calculate spatial information using the following formula: ； In the formula, is the spatial information, T is the transposition operation, and α2 is the second control factor; Step S45: Feature update, initial feature vector , the feature vector is updated based on the U-layer convolutional layer. The formula used is as follows: ； In the formula, and are the feature vectors output by the u+1th and uth convolutional layers, respectively, u is the convolutional layer index, is the weight matrix of the u-th convolutional layer, σ(·) is the activation function, yes The degree matrix of Step S46: Evaluation, the feature vector output by the U-th convolutional layer Input to the fully connected layer, and then use the softmax activation function to get the probability distribution corresponding to each data label, and take the data label with the highest probability as the final output label. The formula used is as follows: ； Where Y is the probability distribution, and They are the weight matrix and bias vector of the fully connected layer respectively; Step S47: Model parameter search.

5. The method for evaluating the advertising design effect based on artificial intelligence according to claim 4, characterized in that: In step S47, the model parameter search specifically includes the following steps: Step S471: Initial position, establish a parameter search space for the first control factor α1, the weight λ and the second control factor α2, randomly initialize K individual positions in the parameter search space, use the individual positions as representatives of model parameters, and use the cross entropy loss function of the advertising design effect evaluation model trained based on the individual positions for the test data set as the individual fitness value; Step S472: individual position division; Step S473: Position update, using the sine-cosine algorithm to update the excellent individual position and the general individual position respectively; the formula used is as follows: ； In the formula, and are the positions of the wth individual in the t+1th and tth searches, respectively. w is the individual index. sin(•) and cos(•) are the sine and cosine functions, respectively. q1 and q2 are random numbers uniformly distributed in the range of (0, 2π) and (0, 2), respectively. is the global optimal position at the tth search, δ w is the labeling factor of the w-th individual position, is the step size factor; Step S474: Determine the optimal parameters, pre-set the fitness threshold, and update the individual fitness value. When the fitness value of the global optimal position is less than the fitness threshold, the parameters corresponding to the global optimal position are the optimal parameters. Based on the optimal parameters, build an advertising design effect evaluation model to complete the parameter search; otherwise, if the maximum number of searches is reached, go to step S471 to reinitialize; otherwise, add 1 to the number of searches and return to step S472 to continue the iterative search.

6. The advertising design effect evaluation method based on artificial intelligence according to claim 1 is characterized by: In step S1, the data collection is to collect historical advertising data; the historical advertising data includes delivery data, user behavior data, user feedback data, product sales growth data and effect level; the delivery data includes delivery platform data and delivery time data; the user behavior data includes click-through rate, browsing time, conversion rate and number of shares; the user feedback data includes comment data and rating data; the effect level includes excellent, good, average and poor, and the effect level is used as a data label.

7. The advertising design effect evaluation method based on artificial intelligence according to claim 1 is characterized by: In step S5, the advertising design effect evaluation is to collect real-time advertising data; the real-time advertising data includes delivery data, user behavior data, user feedback data and product sales growth data. After pre-processing, the real-time advertising data is input into the advertising design effect evaluation model constructed based on the optimal parameters for processing, and based on the output label, the effect level corresponding to the real-time advertising data is obtained.

8. The method for evaluating advertising design effects based on artificial intelligence according to claim 1, characterized in that: In step S2, the data preprocessing is to perform data cleaning, data conversion, data normalization and data set construction on the collected data; data cleaning includes processing noise, outliers, duplicate values ​​and missing values; Data conversion is to convert data into vector form; data normalization is to unify the data range based on the maximum and minimum normalization method; data set construction is to construct an advertising data set based on the data after data cleaning, data conversion and data normalization, and divide the advertising data set into minority class data set and majority class data set.

9. An artificial intelligence-based advertising design effect evaluation system, used to implement the artificial intelligence-based advertising design effect evaluation method as described in any one of claims 1 to 8, characterized in that: It includes a data collection module, a data preprocessing module, an advertising data balance processing module, a module for building an advertising design effect evaluation model, and an advertising design effect evaluation module; The data collection module collects historical advertising data and sends the data to the data preprocessing module; The data preprocessing module performs data cleaning, data conversion, data normalization and data set construction on the collected data, and sends the data to the advertising data balancing processing module; The advertising data balancing processing module clusters the minority class data set, obtains the amount of data that needs to be synthesized for each sub-cluster based on the average density, comprehensively considers the importance of the advertising data in the sub-cluster to which it belongs and the similarity with other sub-clusters, obtains the selection probability of each advertising data in each sub-cluster, synthesizes new advertising data based on the cluster center and the random adjustment factor, constructs a balanced data set according to the spatial distribution entropy, and sends the data to the module for constructing the advertising design effect evaluation model; The module for constructing the advertising design effect evaluation model uses one-dimensional convolution to extract potential information, adjusts the adjacency matrix to add self-connected edges, calculates spatial information, updates features based on the convolution layer, obtains output labels, designs an adaptive division ratio to divide individual positions, uses the sine-cosine algorithm to update individual positions respectively, constructs an advertising design effect evaluation model based on optimal parameters, and sends data to the advertising design effect evaluation module; The advertisement design effect evaluation module pre-processes the real-time advertisement data and then inputs the pre-processed data into an advertisement design effect evaluation model constructed based on optimal parameters to obtain the effect level corresponding to the real-time advertisement data.

Images