Advertisement automatic design method and system based on AI

Abstract

The invention belongs to the technical field of advertisement design, and discloses an automatic advertisement design method and system based on AI. Comprising the steps of collecting historical advertisement data and extension data corresponding to the historical advertisement data; selecting the historical advertisement data according to the extension data corresponding to the historical advertisement data to obtain selected historical advertisement data; the selected historical advertisement data comprises selected advertisement picture data and selected advertisement copywriting data, and feature extraction is performed on the selected advertisement picture data and the selected advertisement copywriting data to obtain selected advertisement feature vectors and selected advertisement copywriting feature words; generating a corresponding design advertisement according to the selected advertisement copywriting feature word and the selected advertisement feature vector, and pushing the design advertisement to the corresponding crowd; historical advertisement data with relatively high benefits are selected as reference advertisement data, so that the benefits of advertisement design are improved, and the sales volume of corresponding commodities is increased.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising design, and more specifically, to an AI-based automatic advertising design method and system. Background Art

[0002] The patent application with the publication number CN114004917A discloses an intelligent design method for store marketing posters, which includes the following steps: Step 1, build a material library; Step 2, learn the designer's design habits and rules for different styles from the data through the planner; Step 3, make the image meet the aesthetic standards through the optimizer; Step 4, construct the rendered image through the generator; and realize automatic expansion and generation based on fixed templates through image retrieval. In this way, precise marketing customization of "thousands of faces for thousands of people" can be achieved. Pixel-level classification is achieved to facilitate AI autonomous learning. No professional technicians are required to participate to reduce energy waste. Through database processing and application, a large amount of repeated investment required for advertising activities can be reduced.

[0003] However, in the process of building the material library, the materials were not selected, resulting in the presence of low-efficiency materials in the material library. Corresponding marketing posters were generated based on the low-efficiency materials, resulting in low efficiency of the marketing posters, which in turn led to lower revenue efficiency of the store. In addition, the corresponding marketing posters were not pushed based on user feedback, which also led to lower revenue efficiency of the store.

[0004] In view of this, the present invention proposes an AI-based automatic advertisement design method and system to solve the above-mentioned problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an AI-based automatic advertising design method, comprising:

[0006] S1: Collect historical advertising data sets, which include O_P group historical advertising data and its corresponding extended data;

[0007] S2: selecting the historical advertising data according to the extended data corresponding to the historical advertising data, the selection process includes preprocessing the extended data, obtaining a comment word set, obtaining a comment factor according to the part of speech of the words in the comment word set, obtaining a comment coefficient according to the comment factor, marking the comment coefficient, obtaining a selected comment coefficient according to the comment coefficient and its marking, and then obtaining a purchaser comment coefficient, obtaining a purchase benefit coefficient according to the extended data, obtaining a selected extension coefficient according to the purchaser comment coefficient and the purchase benefit coefficient, and eliminating the corresponding historical advertising data according to the selected extension coefficient to obtain selected historical advertising data;

[0008] S3: selecting historical advertisement data including selected advertisement image data and selected advertisement text data, respectively extracting features from the selected advertisement image data and the selected advertisement text data to obtain selected advertisement feature vectors and selected advertisement text feature words;

[0009] S4: Generate corresponding designed advertisements according to the selected advertisement copy feature words and the selected advertisement feature vectors, and push the designed advertisements to corresponding groups of people.

[0010] Furthermore, the process of collecting the historical advertisement data set includes:

[0011] The historical advertising data includes advertising image data and advertising copy data, and the extended data corresponding to the historical advertising data includes advertising views, product purchases and purchaser comments;

[0012] Setting a historical advertisement data collection terminal and an extended data collection terminal, wherein the historical advertisement data collection terminal is used to collect historical advertisement data, and the extended data collection terminal is used to collect extended data corresponding to the historical advertisement data;

[0013] The historical advertising data collection terminal and the extended data collection terminal are configured to generate corresponding configuration channels. The historical advertising data collection terminal and the extended data collection terminal are respectively connected to the corresponding server through the generated corresponding configuration channels. The historical advertising data collection terminal and the extended data collection terminal obtain the historical advertising data and the corresponding extended data in the server through the corresponding configuration channels.

[0014] Furthermore, the process of selecting the historical advertisement data according to the extended data corresponding to the historical advertisement data to obtain the selected historical advertisement data includes:

[0015] Preprocess the buyer's comments to obtain a set of buyer's comment words;

[0016] A sentiment word database is preset, wherein the sentiment word database is provided with a positive word storage space, a medium word storage space and a negative word storage space, wherein the positive word storage space stores positive words, the medium word storage space stores medium words, and the negative word storage space stores negative words;

[0017] Compare the comment words in the comment word set with the words stored in the sentiment word database, and obtain the number of positive comment words, the number of medium comment words, and the number of negative comment words in the comment word set according to the comparison result;

[0018] The review factors of the review word set are obtained according to the number of positive review words, the number of medium review words and the number of negative review words, and the review factors include the positive review factor YJ j, Medium Comment Factor YZ j and negative review factor YX j , j is the index of the purchaser, j = 1, 2, 3, ..., m, m is a positive integer;

[0019] Get the comment coefficient JZX of the comment word set according to the comment factor of the comment word set j ;

[0020] Among them, JZX j =max{YJ j , YZ j , YX j};

[0021] The comment coefficients of the comment word set are marked, including positive, medium and negative;

[0022] According to the comment coefficient and annotation of the comment word set, the selected comment coefficient ZZX of the comment word set is obtained j ;

[0023] in, up is the adjustment coefficient, and 0 <up<1;

[0024] Obtain the purchaser review coefficient GMR according to the selected review coefficient of the review word set;

[0025] in,

[0026] Obtain the purchase benefit coefficient GMX based on the number of ad views and product purchases;

[0027] in, LM is the number of product purchases, and LD is the number of ad views;

[0028] Obtain the selection extension coefficient ZXY of each set of historical advertising data according to the purchaser's comment coefficient and the purchase benefit coefficient;

[0029] Among them, ZXY = tp_1×GMR+tp_2×GMX; tp_1 is the adjustment coefficient of the purchaser's comment coefficient, and tp_2 is the adjustment coefficient of the purchase benefit coefficient;

[0030] Set the selection extension factor threshold;

[0031] The selection extension coefficient of each group of historical advertising data is compared with the selection extension coefficient threshold. When the selection extension coefficient is less than the selection extension coefficient threshold, the corresponding group of historical advertising data is eliminated to obtain the selected historical advertising data.

[0032] Furthermore, the process of preprocessing the buyer's comments to obtain the buyer's comment word set includes:

[0033] Punctuation marks, special symbols, numbers and blank characters are removed from the buyer's comments to obtain standard buyer's comments, and a word database is preset, wherein k words are stored in the word database, where k is a positive integer;

[0034] Preset the traversal length, where the traversal length is the length of the longest word in the word database, and start from the first character of the standard purchaser's comment and perform forward traversal;

[0035] The forward traversal process includes: obtaining a character string with the same length as the traversal, comparing the obtained character string with the words in the word database, if there is a word in the word database that is the same as the character string, the forward traversal is successful, and the character string is recorded as a comment word, if there is no word in the word database that is the same as the character string, the forward traversal fails, the traversal length is reduced by one, the character string with the same length as the traversal is obtained, and the obtained character string is compared with the words in the word database to obtain a comparison result;

[0036] Repeat the forward traversal process until there is a word in the word database that is identical to the string or the traversal length is zero. If there is a word in the word database that is identical to the string, continue to start from the next character of the comment word and perform forward traversal until all standard purchaser comments are traversed. If the traversal length is zero, continue to start from the second character and perform forward traversal until all standard purchaser comments are traversed to obtain a set of forward traversed comment words;

[0037] Starting from the last character of the standard purchaser's comment, reverse traversal is performed based on the forward traversal process until a word in the word database is identical to the string or the traversal length is zero. If a word in the word database is identical to the string, reverse traversal is continued from the previous character of the comment word until the entire standard purchaser's comment is traversed. If the traversal length is zero, reverse traversal is continued from the second to last character until the entire standard purchaser's comment is traversed to obtain a set of reverse traversal comment words;

[0038] The union of the set of forward traversal comment words and the set of reverse traversal comment words is obtained. The union of the set of forward traversal comment words and the set of reverse traversal comment words is the buyer's comment word set.

[0039] Furthermore, the process of obtaining the number of positive comment words, the number of medium comment words and the number of negative comment words in the comment word set according to the comparison result, and then obtaining the comment factor of the comment word set includes:

[0040] If the comment word is the same as any positive word stored in the positive word storage space, the comment word is recorded as a positive comment word; if the comment word is the same as any medium word stored in the medium word storage space, the comment word is recorded as a medium comment word; if the comment word is the same as any negative word stored in the negative word storage space, the comment word is recorded as a negative comment word;

[0041] in, kp_1 is the weight of the positive review factor, kp_2 is the weight of the medium review factor, kp_3 is the weight of the negative review factor, RJ j , RZ j and RX j They are the number of positive comment words, the number of medium comment words and the number of negative comment words in the comment word set respectively.

[0042] Furthermore, the process of marking the comment coefficients of the comment word set includes:

[0043] If JZX j =YJ j , then the comment coefficient is marked as positive. If JZX j =YZ j , then the comment coefficient is marked as medium. If JZX j =YX j , then the comment coefficient is marked as negative;

[0044] When YJ j =YZ j =YX j When YJ j =YX j And YJ j When YZ is the largest, the comment coefficient is marked as negative. j =YX j And YZ j When YJ is the largest, the comment coefficient is marked as negative. j =YZ j And YJ j When it is the largest, the comment coefficient is marked as medium.

[0045] Furthermore, the process of extracting features from the selected advertisement copy data to obtain the selected advertisement copy feature words includes:

[0046] Eliminate punctuation marks, special symbols, numbers and blank characters in the selected advertisement copy data to obtain standard selected advertisement copy data;

[0047] Based on the method of forward traversal of standard purchaser comments, forward traversal is performed on the standard selected advertising copy data to obtain the word set corresponding to the forward traversal;

[0048] Based on the method of reverse traversal of standard purchaser comments, reverse traversal is performed on the standard selected advertising copy data to obtain a word set corresponding to the reverse traversal;

[0049] The union of the word set corresponding to the forward traversal and the word set corresponding to the reverse traversal is obtained, and the words in the union of the word set corresponding to the forward traversal and the word set corresponding to the reverse traversal are the selected advertising copy feature words.

[0050] Furthermore, the process of extracting features from the selected advertisement image data to obtain the selected advertisement feature vector includes:

[0051] Preset a feature extraction model, wherein the feature extraction model includes a convolution layer, a pooling layer, a global average pooling layer, and an activation function;

[0052] The selected advertising image data is input into the convolution layer, and a convolution kernel of size K×K is used to perform a convolution operation on the selected advertising image data. The convolution operation is expressed by the formula:

[0053] Among them, Y v，p，u is the pixel value of the uth channel at position (v, p) in the convolutional layer output feature map, u is the index of the channel of the convolutional layer output feature map, C is the number of channels of the selected advertising image data, c is the index of the channel of the selected advertising image data, w, e are the indexes of the convolution kernel size, WC w，e，c，u is the weight of the convolution kernel at the wth row, eth column, cth channel to uth channel, X v+w，p+e，c To select the pixel value of the cth channel of the advertisement image data at position (v+w, p+e), bc u is the bias term;

[0054] The output feature map of the convolutional layer is input to the pooling layer, and the feature map is pooled. The pooling operation is expressed by the formula: v，p，u =max(w,e)(Y v+w，p+e，u );

[0055] Among them, P v，p，u is the pixel value at position (v, p) in the u-th channel output by the pooling layer, (w, e) is the pooling window, and Y v+w，p+e，u is the pixel value of the area covered by the pooling window in the u-th channel of the convolutional layer output, max(w, e)(Y v+w，p+e，u ) is to select the maximum pixel value in the pooling window for output;

[0056] The output of the pooling layer is input into the activation function for activation operation. The activation operation is expressed by the formula f(t)=max(0, t);

[0057] Where f(t) is the output of the activation function, t is the input value to the activation function, and max(0, t) is the larger value of the input 0 and t;

[0058] The output of the activation function is input to the convolution layer, the output of the convolution layer is input to the pooling layer, the output of the pooling layer is input to the activation function, and the convolution operation, pooling operation and activation operation are repeated L times to obtain the selected advertising feature map of size HY×WY and the pixel values ​​Z at different positions and channels of the selected advertising feature map. a，r，u , (a, r) ​​is the position of the pixel in the selected advertisement feature map, HY is the height of the selected advertisement feature map, and WY is the width of the selected advertisement feature map;

[0059] The result of each convolution operation, pooling operation, and activation operation is expressed mathematically as: Z (L) =f(W (L) *Z (L-1) +b (L) );

[0060] Among them, Z (L) is the output of the Lth activation function, W (L) is the L-th convolution kernel weight, Z (L-1) is the output of the L-1th activation function, b (L) is the L-th bias term, * represents the convolution operation;

[0061] The results of L convolution operations, pooling operations, and activation operations are input to the global average pooling layer, and the average pooling operation is performed to obtain the selected advertising feature vector. The average pooling operation is expressed by the formula:

[0062] Among them, vc u is the uth value in the selected advertisement feature vector.

[0063] Furthermore, the process of generating a corresponding designed advertisement according to the selected advertisement copy feature words and the selected advertisement feature vector, and pushing the designed advertisement to the corresponding group of people includes:

[0064] Using the selected advertising copy feature words as seed words for the designed advertising copy, expanding the selected advertising copy feature words to obtain the designed advertising copy, comparing the designed advertising copy with the corresponding selected advertising copy data to obtain the similarity between the designed advertising copy and the corresponding selected advertising copy data, and setting a similarity threshold range; if the similarity between the designed advertising copy and the corresponding selected advertising copy data is not within the similarity threshold range, re-expanding the selected advertising copy feature words until the similarity between the designed advertising copy and the corresponding selected advertising copy data is within the similarity threshold range;

[0065] Taking the selected advertisement feature vector as the basic feature vector of the designed advertisement picture, the selected advertisement feature vector is expanded to obtain the selected advertisement expanded feature vector, the designed advertisement picture is obtained according to the selected advertisement expanded feature vector, the designed advertisement picture is compared with the corresponding selected advertisement picture data, the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is obtained, and a similarity threshold range is set; if the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is not within the similarity threshold range, the selected advertisement feature vector is expanded again until the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is within the similarity threshold range;

[0066] Combine the designed advertising image and the designed advertising copy to obtain the designed advertisement, obtain the crowd who have browsed the corresponding selected historical advertising data, and push the designed advertisement to the obtained crowd.

[0067] An AI-based automatic advertising design system, which is used to implement the AI-based automatic advertising design method, includes:

[0068] A data collection module is used to collect historical advertising data and its corresponding extended data;

[0069] A data selection module is used to select the historical advertising data according to the extended data corresponding to the historical advertising data. The selection process includes preprocessing the extended data, obtaining a comment word set, processing the comment word set, obtaining a purchaser comment coefficient, obtaining a purchase benefit coefficient according to the extended data, obtaining a selection extension coefficient according to the purchaser comment coefficient and the purchase benefit coefficient, and eliminating the corresponding historical advertising data according to the selection extension coefficient to obtain selected historical advertising data;

[0070] A feature extraction module is used to extract features from selected historical advertisement data to obtain selected advertisement feature vectors and selected advertisement copy feature words;

[0071] The advertisement push module is used to generate corresponding designed advertisements according to the selected advertisement copy feature words and the selected advertisement feature vectors, and push the designed advertisements to the corresponding groups of people.

[0072] The technical effects and advantages of the AI-based automatic advertising design method and system of the present invention are as follows:

[0073] Collect historical advertising data and its corresponding extended data, select the historical advertising data according to the extended data corresponding to the historical advertising data to obtain selected historical advertising data, the selected historical advertising data includes selected advertising image data and selected advertising copy data, perform feature extraction on the selected advertising image data and the selected advertising copy data respectively to obtain selected advertising feature vectors and selected advertising copy feature words, generate corresponding designed advertisements according to the selected advertising copy feature words and the selected advertising feature vectors, and push the designed advertisements to corresponding groups of people; select the historical advertising data, eliminate historical advertising data with lower benefits, and retain historical advertising data with higher benefits, so as to provide an accurate reference for the subsequent generation of advertisements, so that the generated advertisements have good benefits, thereby increasing the attractiveness of the products and improving the sales of the products, and push the generated advertisements to corresponding users according to user feedback, thereby increasing the number of advertisement views, and further increasing the sales of the products. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A schematic diagram of an AI-based automatic advertisement design method of the present invention;

[0075] Figure 2 Schematic diagram of an AI-based automatic advertising design system of the present invention. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] Example 1

[0078] See also Figure 1 As shown, the AI-based automatic advertisement design method described in this embodiment includes:

[0079] S1: Collect historical advertising data sets, which include O_P group historical advertising data and its corresponding extended data;

[0080] S2: selecting the historical advertising data according to the extended data corresponding to the historical advertising data, the selection process includes preprocessing the extended data, obtaining a comment word set, obtaining a comment factor according to the part of speech of the words in the comment word set, obtaining a comment coefficient according to the comment factor, marking the comment coefficient, obtaining a selected comment coefficient according to the comment coefficient and its marking, and then obtaining a purchaser comment coefficient, obtaining a purchase benefit coefficient according to the extended data, obtaining a selected extension coefficient according to the purchaser comment coefficient and the purchase benefit coefficient, and eliminating the corresponding historical advertising data according to the selected extension coefficient to obtain selected historical advertising data;

[0081] S3: selecting historical advertisement data including selected advertisement image data and selected advertisement text data, respectively extracting features from the selected advertisement image data and the selected advertisement text data to obtain selected advertisement feature vectors and selected advertisement text feature words;

[0082] S4: Generate corresponding designed advertisements according to the selected advertisement copy feature words and the selected advertisement feature vectors, and push the designed advertisements to corresponding groups of people.

[0083] The process of collecting historical advertising datasets includes:

[0084] The historical advertising data includes advertising image data and advertising copy data. The extended data corresponding to the historical advertising data includes advertising views, product purchases, and purchaser comments. The product purchases are the purchases of the product corresponding to the advertisement, and the purchaser comments are comments from people who purchased the product.

[0085] It should be explained that the people who purchased the product in the comments of the people who purchased the product are the people who viewed the advertisement, and the purchase volume of the product corresponding to the advertisement is the purchase volume of the people who viewed the advertisement;

[0086] Setting a historical advertisement data collection terminal and an extended data collection terminal, wherein the historical advertisement data collection terminal is used to collect historical advertisement data, and the extended data collection terminal is used to collect extended data corresponding to the historical advertisement data;

[0087] The historical advertising data collection terminal and the extended data collection terminal are configured to generate corresponding configuration channels. The historical advertising data collection terminal and the extended data collection terminal are respectively connected to the corresponding server through the generated corresponding configuration channels. The historical advertising data collection terminal and the extended data collection terminal obtain the historical advertising data and the corresponding extended data in the server through the corresponding configuration channels.

[0088] The historical advertising data is selected according to the extended data corresponding to the historical advertising data. The process of obtaining the selected historical advertising data includes:

[0089] Preprocess the buyer's comments to obtain a set of buyer's comment words;

[0090] Furthermore, the process of preprocessing the buyer's comments to obtain the buyer's comment word set includes:

[0091] Punctuation marks, special symbols, numbers and blank characters (such as spaces, line breaks, etc.) in the purchaser's comments are removed to obtain standard purchaser comments, and a word database is preset, wherein the word database stores a number of words, and the words are words in the dictionary;

[0092] Preset the traversal length, where the traversal length is the length of the longest word in the word database, and start from the first character of the standard purchaser's comment and perform forward traversal;

[0093] The forward traversal process includes: obtaining a character string with the same length as the traversal, comparing the obtained character string with the words in the word database, if there is a word in the word database that is the same as the character string, the forward traversal is successful, and the character string is recorded as a comment word, if there is no word in the word database that is the same as the character string, the forward traversal fails, the traversal length is reduced by one, the character string with the same length as the traversal is obtained, and the obtained character string is compared with the words in the word database to obtain a comparison result;

[0094] Repeat the forward traversal process until there is a word in the word database that is identical to the string or the traversal length is zero. If there is a word in the word database that is identical to the string, continue to start from the next character of the comment word and perform forward traversal until all standard purchaser comments are traversed. If the traversal length is zero, continue to start from the second character and perform forward traversal until all standard purchaser comments are traversed to obtain a set of forward traversed comment words;

[0095] Starting from the last character of the standard purchaser's comment, reverse traversal is performed based on the forward traversal process until a word in the word database is identical to the string or the traversal length is zero. If a word in the word database is identical to the string, reverse traversal is continued from the previous character of the comment word until the entire standard purchaser's comment is traversed. If the traversal length is zero, reverse traversal is continued from the second to last character until the entire standard purchaser's comment is traversed to obtain a set of reverse traversal comment words;

[0096] Obtain the union of the set of forward traversal comment words and the set of reverse traversal comment words, the union of the set of forward traversal comment words and the set of reverse traversal comment words is the buyer's comment word set;

[0097] A sentiment word database is preset based on the word database, wherein the sentiment word database is provided with a positive word storage space, a medium word storage space and a negative word storage space, wherein the positive word storage space stores a number of positive words, such as words with positive evaluations such as "good", "great", and "perfect", wherein the medium word storage space stores a number of medium words, such as words with medium evaluations such as "general", "mediocre", and the negative word storage space stores a number of negative words, such as words with negative evaluations such as "bad", "bad review", and "angry", etc.;

[0098] Compare the comment words in the comment word set with the words stored in the sentiment word database. If the comment word is identical to any positive word stored in the positive word storage space, the comment word is recorded as a positive comment word. If the comment word is identical to any medium word stored in the medium word storage space, the comment word is recorded as a medium comment word. If the comment word is identical to any negative word stored in the negative word storage space, the comment word is recorded as a negative comment word. Obtain the number of positive comment words, the number of medium comment words, and the number of negative comment words.

[0099] The review factors of the review word set are obtained according to the number of positive review words, the number of medium review words and the number of negative review words, and the review factors include the positive review factor YJ j , Medium Comment Factor YZ j and negative review factor YX j , j is the index of the purchaser, j = 1, 2, 3, ..., m, m is a positive integer;

[0100] in, kp_1 is the weight of the positive review factor, kp_2 is the weight of the medium review factor, kp_3 is the weight of the negative review factor, RJ j , RZ j and RX j They are the number of positive comment words, the number of medium comment words, and the number of negative comment words in the comment word set;

[0101] Get the comment coefficient JZX of the comment word set according to the comment factor of the comment word set j ;

[0102] Among them, JZX j =max{YJ j , YZ j , YX j}; max{x, y, z} means taking the maximum value among x, y, z;

[0103] Mark the comment coefficients of the comment word set. If JZX j =YJ j , then the comment coefficient is marked as positive. If JZX j =YZ j , then the comment coefficient is marked as medium. If JZX j =YX j , then the comment coefficient is marked as negative;

[0104] What needs to be explained is that when YJ j =YZ j =YX j When YJ j =YX j And YJ j When YZ is the largest, the comment coefficient is marked as negative. j =YX j And YZ j When YJ is the largest, the comment coefficient is marked as negative. j =YZ j And YJ j When it is the largest, the comment coefficient is marked as medium state;

[0105] According to the comment coefficient and annotation of the comment word set, the selected comment coefficient ZZX of the comment word set is obtained j ;

[0106] in, up is the adjustment coefficient, and 0 <up<1;

[0107] Obtain the purchaser review coefficient GMR according to the selected review coefficient of the review word set;

[0108] in,

[0109] Obtain the purchase benefit coefficient GMX based on the number of ad views and product purchases;

[0110] in, LM is the number of product purchases, and LD is the number of ad views;

[0111] Obtain the selection extension coefficient ZXY of each set of historical advertising data according to the purchaser's comment coefficient and the purchase benefit coefficient;

[0112] Among them, ZXY = tp_1×GMR+tp_2×GMX; tp_1 is the adjustment coefficient of the purchaser's comment coefficient, and tp_2 is the adjustment coefficient of the purchase benefit coefficient;

[0113] Set the selection extension factor threshold;

[0114] The selection extension coefficient of each group of historical advertising data is compared with the selection extension coefficient threshold. When the selection extension coefficient is less than the selection extension coefficient threshold, the corresponding group of historical advertising data is eliminated to obtain the selected historical advertising data.

[0115] The process of extracting features from the selected advertising copy data and obtaining the feature words of the selected advertising copy includes:

[0116] Remove punctuation marks, special symbols, numbers and blank characters (such as spaces, line breaks, etc.) from the selected advertising copy data to obtain standard selected advertising copy data;

[0117] Based on the method of forward traversal of standard purchaser comments, forward traversal is performed on the standard selected advertising copy data to obtain the word set corresponding to the forward traversal;

[0118] Based on the method of reverse traversal of standard purchaser comments, reverse traversal is performed on the standard selected advertising copy data to obtain a word set corresponding to the reverse traversal;

[0119] Obtaining the union of the word set corresponding to the forward traversal and the word set corresponding to the reverse traversal, the words in the union of the word set corresponding to the forward traversal and the word set corresponding to the reverse traversal are the selected advertising copy feature words;

[0120] The process of extracting features from the selected advertisement image data and obtaining the selected advertisement feature vector includes:

[0121] Preset a feature extraction model, wherein the feature extraction model includes a convolution layer, a pooling layer, a global average pooling layer, and an activation function;

[0122] The selected advertisement image data is input into the convolution layer, which is used to extract local features from the input image data. A convolution kernel of size K×K is used to perform a convolution operation on the selected advertisement image data. The convolution operation is expressed by the formula:

[0123] Among them, Y v，p，u is the pixel value of the uth channel at position (v, p) in the convolutional layer output feature map, u is the index of the channel of the convolutional layer output feature map, C is the number of channels of the selected advertising image data, c is the index of the channel of the selected advertising image data, w, e are the indexes of the convolution kernel size, WC w，e，c，u is the weight of the convolution kernel at the wth row, eth column, cth channel to uth channel, X v+w，p+e，c To select the pixel value of the cth channel of the advertisement image data at position (v+w, p+e), bc u is the bias term;

[0124] It needs to be explained that the convolution operation can extract different features of the image by learning different convolution kernels;

[0125] The output feature map of the convolutional layer is input to the pooling layer, which is used to reduce the size of the feature map while retaining important information. The feature map is pooled. The pooling operation is expressed by the formula: v，p，u =max(w,e)(Y v+w，p+e，u );

[0126] Among them, P v，p，u is the pixel value at position (v, p) in the u-th channel output by the pooling layer, (w, e) is the pooling window, and Y v+w，p+e，u is the pixel value of the area covered by the pooling window in the u-th channel of the convolutional layer output, max(w, e)(Y v+w，p+e，u ) is to select the maximum pixel value in the pooling window for output;

[0127] The output of the pooling layer is input into the activation function for activation operation. The activation operation is expressed by the formula f(t)=max(0, t);

[0128] Where f(t) is the output of the activation function, t is the input value to the activation function, and max(0, t) is the larger value of the input 0 and t;

[0129] The output of the activation function is input to the convolution layer, the output of the convolution layer is input to the pooling layer, the output of the pooling layer is input to the activation function, and the convolution operation, pooling operation and activation operation are repeated L times to obtain the selected advertising feature map of size HY×WY and the pixel values ​​Z at different positions and channels of the selected advertising feature map. a，r，u , (a, r) ​​is the position of the pixel in the selected advertising feature map, HY is the height of the selected advertising feature map (the number of pixels in the vertical direction), and WY is the width of the selected advertising feature map (the number of pixels in the horizontal direction);

[0130] The result of each convolution operation, pooling operation, and activation operation is expressed mathematically as: Z (L) =f(W (L) *Z (L-1) +b (L) );

[0131] Among them, Z (L) is the output of the Lth activation function, W (L) is the L-th convolution kernel weight, Z (L-1) is the output of the L-1th activation function, b (L) is the L-th bias term, * represents the convolution operation;

[0132] The results of L convolution operations, pooling operations, and activation operations are input to the global average pooling layer, and the average pooling operation is performed to obtain the selected advertising feature vector. The average pooling operation is expressed by the formula:

[0133] Among them, vc u is the uth value in the selected advertisement feature vector.

[0134] The process of generating corresponding designed advertisements according to selected advertisement copy feature words and selected advertisement feature vectors, and pushing the designed advertisements to corresponding groups of people includes:

[0135] Using the selected advertising copy feature words as seed words for the designed advertising copy, expanding the selected advertising copy feature words to obtain the designed advertising copy, comparing the designed advertising copy with the corresponding selected advertising copy data to obtain the similarity between the designed advertising copy and the corresponding selected advertising copy data, and setting a similarity threshold range; if the similarity between the designed advertising copy and the corresponding selected advertising copy data is not within the similarity threshold range, re-expanding the selected advertising copy feature words until the similarity between the designed advertising copy and the corresponding selected advertising copy data is within the similarity threshold range;

[0136] Taking the selected advertisement feature vector as the basic feature vector of the designed advertisement picture, the selected advertisement feature vector is expanded to obtain the selected advertisement expanded feature vector, the designed advertisement picture is obtained according to the selected advertisement expanded feature vector, the designed advertisement picture is compared with the corresponding selected advertisement picture data, the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is obtained, and a similarity threshold range is set; if the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is not within the similarity threshold range, the selected advertisement feature vector is expanded again until the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is within the similarity threshold range;

[0137] Combine the designed advertisement image and the designed advertisement copy to obtain the designed advertisement, obtain the crowd that has browsed the corresponding selected historical advertisement data, and push the designed advertisement to the obtained crowd;

[0138] It needs to be explained that the selected historical advertising data formed by combining the selected advertising image data corresponding to the basic feature vector of the designed advertising image and the selected advertising copy data corresponding to the seed word of the designed advertising copy is recorded as marked selected historical advertising data, and the marked selected historical advertising data is the corresponding selected historical advertising data.

[0139] In this embodiment, historical advertising data is selected, historical advertising data with lower effectiveness is eliminated, and historical advertising data with higher effectiveness is retained, providing an accurate reference for the subsequent generation of advertisements, so that the generated advertisements have good effectiveness, thereby increasing the attractiveness of the products and increasing the sales of the products. The generated advertisements are pushed to the corresponding users based on the user's feedback, thereby increasing the number of advertisement views and further increasing the sales of the products.

[0140] Example 2

[0141] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an AI-based automatic advertising design system, including:

[0142] A data collection module is used to collect historical advertising data and its corresponding extended data;

[0143] A data selection module is used to select the historical advertising data according to the extended data corresponding to the historical advertising data. The selection process includes preprocessing the extended data, obtaining a comment word set, processing the comment word set, obtaining a purchaser comment coefficient, obtaining a purchase benefit coefficient according to the extended data, obtaining a selection extension coefficient according to the purchaser comment coefficient and the purchase benefit coefficient, and eliminating the corresponding historical advertising data according to the selection extension coefficient to obtain selected historical advertising data;

[0144] A feature extraction module is used to extract features from selected historical advertisement data to obtain selected advertisement feature vectors and selected advertisement copy feature words;

[0145] The advertisement push module is used to generate corresponding designed advertisements according to the selected advertisement copy feature words and the selected advertisement feature vectors, and push the designed advertisements to the corresponding groups of people.

[0146] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0147] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0148] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0149] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An AI-based automatic advertising design method, characterized in that: The AI-based automatic advertising design method includes: S1: Collect historical advertising data sets, which include O_P group historical advertising data and its corresponding extended data; S2: selecting the historical advertising data according to the extended data corresponding to the historical advertising data, the selection process includes preprocessing the extended data, obtaining a comment word set, obtaining a comment factor according to the part of speech of the words in the comment word set, obtaining a comment coefficient according to the comment factor, marking the comment coefficient, obtaining a selected comment coefficient according to the comment coefficient and its marking, and then obtaining a purchaser comment coefficient, obtaining a purchase benefit coefficient according to the extended data, obtaining a selected extension coefficient according to the purchaser comment coefficient and the purchase benefit coefficient, and eliminating the corresponding historical advertising data according to the selected extension coefficient to obtain selected historical advertising data; S3: selecting historical advertisement data including selected advertisement image data and selected advertisement text data, respectively extracting features from the selected advertisement image data and the selected advertisement text data to obtain selected advertisement feature vectors and selected advertisement text feature words; S4: Generate corresponding designed advertisements according to the selected advertisement copy feature words and the selected advertisement feature vectors, and push the designed advertisements to corresponding groups of people.

2. The AI-based automatic advertising design method according to claim 1, characterized in that: The process of collecting the historical advertising data set includes: The historical advertising data includes advertising image data and advertising copy data, and the extended data corresponding to the historical advertising data includes advertising views, product purchases and purchaser comments; Setting a historical advertisement data collection terminal and an extended data collection terminal, wherein the historical advertisement data collection terminal is used to collect historical advertisement data, and the extended data collection terminal is used to collect extended data corresponding to the historical advertisement data; The historical advertising data collection terminal and the extended data collection terminal are configured to generate corresponding configuration channels. The historical advertising data collection terminal and the extended data collection terminal are respectively connected to the corresponding server through the generated corresponding configuration channels. The historical advertising data collection terminal and the extended data collection terminal obtain the historical advertising data and the corresponding extended data in the server through the corresponding configuration channels.

3. The AI-based automatic advertising design method according to claim 2, characterized in that: The process of selecting the historical advertisement data according to the extended data corresponding to the historical advertisement data to obtain the selected historical advertisement data includes: Preprocess the buyer's comments to obtain a set of buyer's comment words; A sentiment word database is preset, wherein the sentiment word database is provided with a positive word storage space, a medium word storage space and a negative word storage space, wherein the positive word storage space stores positive words, the medium word storage space stores medium words, and the negative word storage space stores negative words; Compare the comment words in the comment word set with the words stored in the sentiment word database, and obtain the number of positive comment words, the number of medium comment words, and the number of negative comment words in the comment word set according to the comparison result; The review factors of the review word set are obtained according to the number of positive review words, the number of medium review words and the number of negative review words, and the review factors include the positive review factor YJ j , Medium Comment Factor YZ j and negative review factor YX j , j is the index of the purchaser, j = 1, 2, 3, ..., m, m is a positive integer; Get the comment coefficient JZX of the comment word set according to the comment factor of the comment word set j ; Among them, JZX j =max{YJ j , YZ j , YX j }; The comment coefficients of the comment word set are marked, including positive, medium and negative; According to the comment coefficient and annotation of the comment word set, the selected comment coefficient ZZX of the comment word set is obtained j ; in, up is the adjustment coefficient, and 0 <up<1; Obtain the purchaser review coefficient GMR according to the selected review coefficient of the review word set; in, Obtain the purchase benefit coefficient GMX based on the number of ad views and product purchases; in, LM is the number of product purchases, and LD is the number of ad views; Obtain the selection extension coefficient ZXY of each set of historical advertising data according to the purchaser's comment coefficient and the purchase benefit coefficient; Among them, ZXY = tp_1 × GMR + tp_2 × GMX; tp_1 is the adjustment coefficient of the purchaser's comment coefficient, and tp_2 is the adjustment coefficient of the purchase benefit coefficient; Set the selection extension factor threshold; The selection extension coefficient of each group of historical advertising data is compared with the selection extension coefficient threshold. When the selection extension coefficient is less than the selection extension coefficient threshold, the corresponding group of historical advertising data is eliminated to obtain the selected historical advertising data.

4. The AI-based automatic advertising design method according to claim 3, characterized in that: The process of preprocessing the buyer's comments to obtain the buyer's comment word set includes: Punctuation marks, special symbols, numbers and blank characters are removed from the buyer's comments to obtain standard buyer's comments, and a word database is preset, wherein k words are stored in the word database, where k is a positive integer; Preset the traversal length, where the traversal length is the length of the longest word in the word database, and perform forward traversal starting from the first character of the standard purchaser's comment; The forward traversal process includes: obtaining a character string with the same length as the traversal, comparing the obtained character string with the words in the word database, if there is a word in the word database that is the same as the character string, the forward traversal is successful, and the character string is recorded as a comment word, if there is no word in the word database that is the same as the character string, the forward traversal fails, the traversal length is reduced by one, the character string with the same length as the traversal is obtained, and the obtained character string is compared with the words in the word database to obtain a comparison result; Repeat the forward traversal process until there is a word in the word database that is identical to the string or the traversal length is zero. If there is a word in the word database that is identical to the string, continue to start from the next character of the comment word and perform forward traversal until all standard purchaser comments are traversed. If the traversal length is zero, continue to start from the second character and perform forward traversal until all standard purchaser comments are traversed to obtain a set of forward traversed comment words; Starting from the last character of the standard purchaser's comment, reverse traversal is performed based on the forward traversal process until a word in the word database is identical to the string or the traversal length is zero. If a word in the word database is identical to the string, reverse traversal is continued from the previous character of the comment word until the entire standard purchaser's comment is traversed. If the traversal length is zero, reverse traversal is continued from the second to last character until the entire standard purchaser's comment is traversed to obtain a set of reverse traversal comment words; The union of the set of forward traversal comment words and the set of reverse traversal comment words is obtained. The union of the set of forward traversal comment words and the set of reverse traversal comment words is the buyer's comment word set.

5. The AI-based automatic advertising design method according to claim 4, characterized in that: The process of obtaining the number of positive comment words, the number of medium comment words, and the number of negative comment words in the comment word set according to the comparison result, and then obtaining the comment factor of the comment word set includes: If the comment word is the same as any positive word stored in the positive word storage space, the comment word is recorded as a positive comment word; if the comment word is the same as any medium word stored in the medium word storage space, the comment word is recorded as a medium comment word; if the comment word is the same as any negative word stored in the negative word storage space, the comment word is recorded as a negative comment word; in, kp_1 is the weight of the positive review factor, kp_2 is the weight of the medium review factor, kp_3 is the weight of the negative review factor, RJ j , RZ j and RX j They are the number of positive comment words, the number of medium comment words and the number of negative comment words in the comment word set respectively.

6. The AI-based automatic advertising design method according to claim 5, characterized in that: The process of marking the comment coefficients of the comment word set includes: If JZX j =YJ j , then the comment coefficient is marked as positive. If JZX j =YZ j , then the comment coefficient is marked as medium. If JZX j =YX j , then the comment coefficient is marked as negative; When YJ j =YZ j =YX j When YJ j =YX j And YJ j When YZ is the largest, the comment coefficient is marked as negative. j =YX j And YZ j When YJ is the largest, the comment coefficient is marked as negative. j =YZ j And YJ j When it is the largest, the comment coefficient is marked as medium.

7. The AI-based automatic advertising design method according to claim 6, characterized in that: The process of extracting features from the selected advertisement copy data to obtain the selected advertisement copy feature words includes: Eliminate punctuation marks, special symbols, numbers and blank characters in the selected advertisement copy data to obtain standard selected advertisement copy data; Based on the method of forward traversal of standard purchaser comments, forward traversal is performed on the standard selected advertising copy data to obtain a word set corresponding to the forward traversal; Based on the method of reverse traversal of standard purchaser comments, reverse traversal is performed on the standard selected advertising copy data to obtain a word set corresponding to the reverse traversal; The union of the word set corresponding to the forward traversal and the word set corresponding to the reverse traversal is obtained, and the words in the union of the word set corresponding to the forward traversal and the word set corresponding to the reverse traversal are the selected advertising copy feature words.

8. The AI-based automatic advertising design method according to claim 7, characterized in that: The process of extracting features from the selected advertisement image data to obtain the selected advertisement feature vector includes: Preset a feature extraction model, wherein the feature extraction model includes a convolution layer, a pooling layer, a global average pooling layer, and an activation function; The selected advertising image data is input into the convolution layer, and a convolution kernel of size K×K is used to perform a convolution operation on the selected advertising image data. The convolution operation is expressed by the formula: Among them, Y v，p，u is the pixel value of the uth channel at position (v, p) in the convolutional layer output feature map, u is the index of the channel of the convolutional layer output feature map, C is the number of channels of the selected advertising image data, c is the index of the channel of the selected advertising image data, w, e are the indexes of the convolution kernel size, WC w，e，c，u is the weight of the convolution kernel at the wth row, eth column, cth channel to uth channel, X v+w，p+e，c To select the pixel value of the cth channel of the advertisement image data at position (v+w, p+e), bc u is the bias term; The output feature map of the convolutional layer is input to the pooling layer, and the feature map is pooled. The pooling operation is expressed by the formula: v，p，u =max(w,e)(Y v+w，p+e，u ); Among them, P v，p，u is the pixel value at position (v, p) in the u-th channel output by the pooling layer, (w, e) is the pooling window, and Y v+w，p+e，u is the pixel value of the area covered by the pooling window in the u-th channel of the convolutional layer output, max(w, e)(Y v+w，p+e，u ) is to select the maximum pixel value in the pooling window for output; The output of the pooling layer is input into the activation function for activation operation. The activation operation is expressed by the formula f(t)=max(0, t); Where f(t) is the output of the activation function, t is the input value to the activation function, and max(0, t) is the larger value of the input 0 and t; The output of the activation function is input to the convolution layer, the output of the convolution layer is input to the pooling layer, the output of the pooling layer is input to the activation function, and the convolution operation, pooling operation and activation operation are repeated L times to obtain the selected advertising feature map of size HY×WY and the pixel values ​​Z at different positions and channels of the selected advertising feature map. a，r，u , (a, r) ​​is the position of the pixel in the selected advertisement feature map, HY is the height of the selected advertisement feature map, and WY is the width of the selected advertisement feature map; The result of each convolution operation, pooling operation, and activation operation is expressed mathematically as: Z (L) =f(W (L) *Z (L-1) +b (L) ); Among them, Z (L) is the output of the Lth activation function, W (L) is the L-th convolution kernel weight, Z (L-1) is the output of the L-1th activation function, b (L) is the L-th bias term, * represents the convolution operation; The results of L convolution operations, pooling operations, and activation operations are input to the global average pooling layer, and the average pooling operation is performed to obtain the selected advertising feature vector. The average pooling operation is expressed by the formula: Among them, vc u is the uth value in the selected advertisement feature vector.

9. The AI-based automatic advertising design method according to claim 8, characterized in that: The process of generating a corresponding designed advertisement according to the selected advertisement copy feature words and the selected advertisement feature vector, and pushing the designed advertisement to the corresponding group of people includes: Using the selected advertising copy feature words as seed words for the designed advertising copy, expanding the selected advertising copy feature words to obtain the designed advertising copy, comparing the designed advertising copy with the corresponding selected advertising copy data to obtain the similarity between the designed advertising copy and the corresponding selected advertising copy data, and setting a similarity threshold range; if the similarity between the designed advertising copy and the corresponding selected advertising copy data is not within the similarity threshold range, re-expanding the selected advertising copy feature words until the similarity between the designed advertising copy and the corresponding selected advertising copy data is within the similarity threshold range; Taking the selected advertisement feature vector as the basic feature vector of the designed advertisement picture, the selected advertisement feature vector is expanded to obtain the selected advertisement expanded feature vector, the designed advertisement picture is obtained according to the selected advertisement expanded feature vector, the designed advertisement picture is compared with the corresponding selected advertisement picture data, the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is obtained, and a similarity threshold range is set; if the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is not within the similarity threshold range, the selected advertisement feature vector is expanded again until the similarity between the designed advertisement picture and the corresponding selected advertisement picture data is within the similarity threshold range; Combine the designed advertising pictures and the designed advertising copy to obtain the designed advertising, obtain the crowd who have browsed the corresponding selected historical advertising data, and push the designed advertising to the obtained crowd.

10. An AI-based automatic advertising design system, used to implement an AI-based automatic advertising design method according to any one of claims 1 to 9, characterized in that: include: A data collection module is used to collect historical advertising data and its corresponding extended data; A data selection module is used to select the historical advertising data according to the extended data corresponding to the historical advertising data. The selection process includes preprocessing the extended data, obtaining a comment word set, processing the comment word set, obtaining a purchaser comment coefficient, obtaining a purchase benefit coefficient according to the extended data, obtaining a selection extension coefficient according to the purchaser comment coefficient and the purchase benefit coefficient, and eliminating the corresponding historical advertising data according to the selection extension coefficient to obtain selected historical advertising data; A feature extraction module is used to extract features from selected historical advertisement data to obtain selected advertisement feature vectors and selected advertisement copy feature words; The advertisement push module is used to generate corresponding designed advertisements according to the selected advertisement copy feature words and the selected advertisement feature vectors, and push the designed advertisements to the corresponding groups of people.

Images