An intelligent advertisement content generation and optimization system based on big data

Abstract

The application relates to the technical field of advertisement marketing, and discloses an intelligent advertisement content generation and optimization system based on big data, which comprises a customer data acquisition module, a competitive product advertisement acquisition module, a competitive product advertisement analysis module, an intelligent generation and prediction module, a content optimization module and an interaction module; the competitive product advertisement acquisition module and the competitive product advertisement analysis module are arranged, the competitive product of a product is selected, corresponding competitive products are selected according to different competitive product standards, the foundation for analyzing the advertisement of the competitive product is laid, and the precision is improved; the advertisement data of the competitive product is analyzed, the relationship between the advertisement and the user is obtained from the side, if the advertisement effect is good, the like degree and the receiving degree of the user to the advertisement are higher, therefore, the information of the user does not need to be collected, only the big data acquisition and analysis of the advertisement data are needed, and the privacy protection of the user is ensured.

Description

Technical Field

[0001] This invention relates to the field of advertising and marketing technology, and more specifically to an intelligent advertising content generation and optimization system based on big data. Background Technology

[0002] With the rapid development of the Internet, the advertising industry is also constantly changing. Traditional advertising production methods require a lot of time and manpower, and the advertising effect is difficult to guarantee. Therefore, how to improve the efficiency of advertising production and advertising effect has become an urgent problem to be solved in the advertising industry. The emergence of big data technology has brought new opportunities to the advertising industry. By analyzing user data, more targeted advertising content can be generated and advertising effect can be improved.

[0003] The publicly available document CN117541322B discloses a method and system for intelligent advertising content generation based on big data analysis, relating to the field of advertising and marketing technology. The method includes: acquiring advertising target data; obtaining advertising target clustering information based on data processing; obtaining target preference feature information based on the advertising target clustering information; intelligently generating advertising content based on advertising content generation information; analyzing the advertising target audience through hierarchical clustering to accurately determine user characteristic information related to advertising effectiveness, precisely grasping the needs and preferences of the target audience; accurately generating targeted advertising content by calculating the advertising matching index to better meet user needs, provide more attractive and effective information, and enhance user recognition and acceptance of the advertising; and evaluating advertising effectiveness through advertising effectiveness data, obtaining an advertising effectiveness evaluation index to adjust and optimize advertising content in a timely manner, thereby improving the effectiveness and efficiency of advertising.

[0004] However, when using big data to obtain user characteristic information, it includes users' sensitive personal information, such as user behavior data and social media data obtained from big data in public documents, which can easily infringe on users' personal privacy. At the same time, when intelligently generating advertising content, if it relies entirely on artificial intelligence, the advertising content may contain some harmful content and negative opinions, which may bring security, data and ethical risks. Therefore, while using artificial intelligence to intelligently generate advertising, it is also necessary to identify and optimize the advertising content, so as to optimize the advertising content, rather than just optimizing the advertising effect.

[0005] In view of this, the present invention proposes an intelligent advertising content generation and optimization system based on big data, which avoids user privacy issues in the intelligent generation of advertising content, and at the same time identifies and optimizes the generated advertising content to make the advertising content higher quality. Summary of the Invention

[0006] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides an intelligent advertising content generation and optimization system based on big data to solve the problems existing in the background art.

[0007] The present invention provides the following technical solution: an intelligent advertising content generation and optimization system based on big data, including a customer data collection module, a competitor advertising acquisition module, a competitor advertising analysis module, an intelligent generation and prediction module, a content optimization module, and an interaction module;

[0008] The customer data collection module is used to collect customer demand data and product data, and transmit them to the competitor advertising acquisition module and the competitor advertising analysis module.

[0009] The competitor advertising acquisition module is used to receive data from the customer data collection module, acquire competitor advertising data, and then transmit it to the competitor advertising analysis module after dividing it into comparative advertising data and numerical advertising data.

[0010] The competitor advertising analysis module is used to receive data from the competitor advertising acquisition module and the customer data collection module, analyze the numerical advertising data and comparative advertising data, and obtain the advantages and mandatory items.

[0011] The intelligent generation and prediction module generates advertisements intelligently based on advantages and mandatory options, then inputs them into the advertisement prediction model to predict the advertising effect, selects the advertisement with the best performance as the candidate advertisement, and then transmits the candidate advertisement and the predicted performance to the content optimization module.

[0012] The content optimization module outputs candidate ads and predicted performance to the interaction module for display. The client can modify the ad content through the content optimization module. After modification, the modified ad is fed back to the intelligent generation and prediction module for performance prediction, and the prediction result is then sent back to the content optimization module. If the client does not need to modify the ad content, the candidate ad is directly used as the final generated ad and transmitted to the interaction module for display.

[0013] The interaction module is used to display the final generated advertisement on the human-computer interaction interface.

[0014] Preferably, the customer demand data includes the advertising content, copywriting elements, and visual elements that the customer expects to see; the advertising content includes product promotion content, service promotion content, brand image content, and promotional activity content; the product promotion content includes product features, usage methods, product advantages and competitor comparisons, and product pricing; the service promotion content includes service content, service process and timeliness, service advantages and features, and customer reviews; the brand image content includes brand story and history, brand values ​​and philosophy, brand partners and honors, and brand public welfare activities and responsibilities; and the promotional activities include limited-time discounts and offers, holiday promotions, points redemption and gift giveaways, and joint promotions and partnerships. The accompanying text element data includes text content, text font, font format, and font layout; the visual element data includes image color, image content, animation content, and video content; the product data is data related to the advertised product, including product name, product image, product function, product price, product sales, product category, and product historical data; the product historical data includes historical sales quantity, sales amount, sales duration, and historical advertising data; the advertising data includes advertising content data, advertising conversion rate, advertising click-through rate, and advertising exposure rate; the product is the advertised product; the advertising content data includes text elements and visual elements in the advertising content.

[0015] Preferably, the specific method by which the competitor advertising acquisition module acquires competitor advertising data is as follows:

[0016] Product data features are extracted and fused. The feature fusion adopts linear fusion, which uses a linear combination to fuse different features. New fused features are obtained through weight allocation, and the new fused features are used as the fused product data features.

[0017] Using big data technology, collect data on N products of the same category as those advertised. Extract the product data features of the N products and then merge them to obtain the merged product data features. Calculate the similarity between the merged product data features and the merged product data features. Select the products corresponding to the top n merged product data features with the highest similarity as competitors.

[0018] The most recent ad data from each of the n competitors is taken as the competitor ad data, meaning there are n competitor ad data.

[0019] Preferably, the competitor advertising acquisition module divides competitor advertising data into comparative advertising data and numerical advertising data, including the following steps:

[0020] Step S01: Encode the advertising data of n competitors, obtain chromosomes, and construct an initial population;

[0021] Step S02: Determine the fitness function;

[0022] Step S03: Perform natural selection on chromosomes in the population;

[0023] Step S04: Perform crossover recombination on chromosomes in the population;

[0024] Step S05: Mutate the chromosomes in the population;

[0025] Step S06: Obtain a new population, with a preset population generation number of L and a fitness threshold of Q, where L is an integer greater than 0 and Q is a real number greater than 0; repeat steps S03-S05 until the new population reaches generation number L or a chromosome in the new population has a fitness value greater than or equal to the fitness threshold Q, then the loop ends, and the competitor's advertising data corresponding to the chromosome with the highest fitness in the new population is taken as the optimal advertising data;

[0026] The competitor's ad data corresponding to the best ad data is used as the comparison ad data, and the remaining n-1 competitor's ad data are used as the numerical ad data.

[0027] Preferably, in step S01, each competitor's advertising data is encoded as A, where A is a chromosome, and B chromosomes are randomly generated to form an initial population C. , where C b Let b be the b-th chromosome, where b = 1, 2, 3, ..., B;

[0028] The fitness function is expressed as follows: ; where f b GX represents the fitness corresponding to the b-th chromosome. b The advertising performance of the competitor ads corresponding to the competitor ads data for the b-th chromosome;

[0029] , among which, SR b _gg represents the advertising revenue generated by competitor ads corresponding to the competitor ad data of chromosome b, and CB represents the advertising revenue generated by competitor ads. b _gg represents the advertising cost of the competitor's ad corresponding to the competitor's ad data for the b-th chromosome.

[0030] Preferably, the similarity between the fused commodity data features and the fused product data features is calculated using cosine similarity, expressed by the formula:

[0031] ;

[0032] Wherein, cosθ iLet RH be the similarity between the i-th fused commodity data feature and the fused product data feature. i _sp is the ith fused product data feature vector, RH_cp is the fused product data feature vector, and ||RH i _sp‖ is the modulus of the i-th fused product data feature vector,‖RH_cp‖ is the modulus of the fused product data feature vector, and "∙" is the dot product, i=1,2,3,...,N.

[0033] Preferably, the competitor advertising analysis module obtains the advantages and mandatory options in the following way:

[0034] Similarity calculations are performed on the text and visual elements in numerical advertising data and comparative advertising data. Both text and visual elements contain text factors and visual factors. All data items in the text element data are text factors, and all data items in the visual element data are visual factors. The text and visual elements are combined into an element set, where both text factors and visual factors are factors within this element set. Therefore:

[0035] The comparison of advertising data is expressed as follows: , among which, U D To compare the set of advertising data elements, D j To compare the data characteristics of the j-th factor in the advertising data element set, m is the total number of factors in the element set, i.e., the sum of the number of copy factors and visual factors;

[0036] Numerical advertising data is represented as follows: , among which, U s_a Let s_a be the set of numerical advertisement data elements for the a-th element. j Let a be the data feature of the j-th factor in the a-th numerical advertising data element set, where a = 1, 2, 3, ..., n-1;

[0037] The factors in the comparative advertising data element set and the numerical advertising data element set are matched one-to-one.

[0038] Calculate the similarity between all factor data features in each numerical advertising data element set and all corresponding factor data features in the comparison advertising data element set. Set a screening threshold YU1. If the similarity does not reach the screening threshold, mark the factor in the comparison advertising data element set as an advantage.

[0039] The required options are factors that must be present in the generated advertising content.

[0040] Preferably, the intelligent generation and prediction module generates P initial advertisements, and inputs the P initial advertisements sequentially into the constructed advertising prediction model to predict the corresponding advertising effect; the input layer of the advertising prediction model has P nodes for inputting the P initial advertisements, and the output layer also has P nodes for outputting the advertising effect of the P initial advertisements; the initial advertisement with the best predicted advertising effect is selected as the candidate advertisement.

[0041] Preferably, the training process of the advertising prediction model is as follows:

[0042] Beforehand, collect advertising content data of competitors in group R1 and historical advertising content data of products in group R2, and use them as analysis data. Then there are a total of (R1+R2) sets of analysis data, where R1 and R2 are both integers greater than 1. Transform a set of analysis data and its corresponding campaign performance into a set of feature vectors.

[0043] Each set of feature vectors is used as input to the advertising prediction model. The advertising prediction model outputs a set of delivery effects corresponding to each set of analysis data, and aims at the actual delivery effect corresponding to each set of analysis data. The actual delivery effect is the pre-collected delivery effect corresponding to the analysis data. The training objective is to minimize the sum of prediction errors of all analysis data. The formula for the prediction error is expressed as: , where ε q The prediction error is represented by q, where q is the group number of the feature vector corresponding to the analyzed data, and θ is the prediction error. q For the q-th set of analysis data, μ represents the delivery effect. q To determine the actual delivery effect corresponding to the qth set of analysis data, the advertising prediction model is trained until the sum of prediction errors converges and training stops.

[0044] The advertising prediction model is a deep neural network model, which includes an input layer, a hidden layer, and an output layer. Each hidden layer contains multiple neurons, and each neuron is connected to the neurons in the next layer, with weights included in the connections.

[0045] Preferably, the content optimization module includes an ad modification unit and an element bar unit; the ad modification unit is used by the customer to modify the content of candidate ads. After the candidate ads are displayed on the operating terminal, the customer can modify the candidate ads. In the ad content of the candidate ads, each text element and visual element is a separate modification item. The customer clicks on the modification item to replace or modify it; the element bar unit is used to display the text elements and visual elements for replacement and modification.

[0046] After the client completes the modifications, they save the revised ad and send it back to the intelligent generation and prediction module. The module then inputs the modified ad into the ad prediction model to predict its performance. The modified ad's performance is compared to that of alternative ads. If the modified ad performs better, it replaces the alternative ad and is sent back to the content optimization module. If the alternative ad performs better, it is not replaced, and the modified ad's performance is displayed in the interaction module. The interaction module then queries the client again to determine whether to replace the alternative ad.

[0047] The technical effects and advantages of this invention are as follows:

[0048] (1) This invention has a competitor advertising acquisition module and a competitor advertising analysis module, which is beneficial to select competitors of the product and select corresponding competitors according to different competitor standards, laying the foundation for subsequent analysis of competitor advertising and improving accuracy; by analyzing the advertising data of competitors, the relationship between advertising and users can be obtained indirectly. If the advertising effect is good, it means that the user's liking and acceptance of the advertising is higher. Therefore, there is no need to collect user information. Only big data collection and analysis of advertising data is required, which ensures the protection of user privacy. At the same time, corresponding advantages can be obtained from competitor advertising, thereby improving the content quality and placement effect of product advertising.

[0049] (2) By providing an intelligent generation and prediction module, this invention is conducive to predicting the advertising effect by constructing an advertising prediction model. By integrating a trained neural network model, it fully utilizes the nonlinear and multimodal data modeling capabilities of the neural network to effectively capture complex data relationships. Based on the parallel computing processing mechanism, it improves the computing efficiency, thereby ensuring the comprehensiveness and accuracy of the analysis. In turn, it can ensure that the intelligently generated advertising has the best advertising effect. Attached Figure Description

[0050] Figure 1 This is a structural diagram of the intelligent advertising content generation and optimization system based on big data according to the present invention. Detailed Implementation

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The intelligent advertising content generation and optimization system based on big data involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, the present invention provides an intelligent advertising content generation and optimization system based on big data, including a customer data collection module, a competitor advertising acquisition module, a competitor advertising analysis module, an intelligent generation and prediction module, a content optimization module, and an interaction module;

[0053] The customer data collection module is used to collect customer demand data and product data, and transmit them to the competitor advertising acquisition module and the competitor advertising analysis module.

[0054] The competitor advertising acquisition module is used to receive data from the customer data collection module, acquire competitor advertising data, and then transmit it to the competitor advertising analysis module after dividing it into comparative advertising data and numerical advertising data.

[0055] The competitor advertising analysis module is used to receive data from the competitor advertising acquisition module and the customer data collection module, analyze the numerical advertising data and comparative advertising data, and obtain the advantages and mandatory items.

[0056] The intelligent generation and prediction module generates advertisements intelligently based on advantages and mandatory options, then inputs them into the advertisement prediction model to predict the advertising effect, selects the advertisement with the best performance as the candidate advertisement, and then transmits the candidate advertisement and the predicted performance to the content optimization module.

[0057] The content optimization module outputs the candidate ads and the predicted delivery effect to the interaction module for display. The customer can modify the ad content through the content optimization module. After the modification is completed, the modified ad is sent back to the intelligent generation and prediction module for delivery effect prediction. The prediction result is then transmitted to the content optimization module. If the customer does not need to modify the ad content, the candidate ad is directly used as the final generated ad and transmitted to the interaction module for display.

[0058] The interaction module is used to display the final generated advertisement on the human-computer interaction interface.

[0059] In this embodiment, it should be specifically noted that the customer demand data includes the advertising content, copywriting element data, and visual element data that the customer expects to be presented; the advertising content includes product promotion content, service promotion content, brand image content, and promotional activity content, etc.; the product promotion content includes, but is not limited to, product feature introduction, product usage methods, product advantages and competitor comparisons, and product prices, etc.; the service promotion content includes, but is not limited to, service content, service process and timeliness, service advantages and features, and customer reviews, etc.; the brand image content includes, but is not limited to, brand story and history, brand values ​​and philosophy, brand partners and honors, and brand public welfare activities and responsibilities, etc.; the promotional activities include, but are not limited to, limited-time discounts and offers, holiday promotions, points redemption and gift giving, and joint promotions with partners, etc. The textual element data includes, but is not limited to, text content, text font, font format, and font layout; the visual element data includes, but is not limited to, image color, image content, animation content, and video content; the product data is data related to the advertised product, including, but not limited to, product name, product image, product function, product price, product sales revenue, product category, and product historical data; the product historical data includes, but is not limited to, historical sales volume, sales amount, sales duration, and historical advertising data; the advertising data includes, but is not limited to, advertising content data, advertising conversion rate, advertising click-through rate, and advertising exposure rate; the product is the advertised product; the advertising content data includes, but is not limited to, textual elements and visual elements in the advertising content;

[0060] By interviewing customers and having them complete electronic questionnaires related to their needs, we can obtain the advertising content, copywriting elements, and visual elements that customers want to present; we can also obtain product data through product catalogs and product descriptions provided by customers, and utilize big data technology to obtain historical product data.

[0061] In this embodiment, it should be specifically explained that the specific method by which the competitor advertising acquisition module acquires competitor advertising data is as follows:

[0062] Product data features are extracted and fused. The feature fusion adopts linear fusion, which uses a linear combination to fuse different features. New fused features are obtained through weight allocation, and the new fused features are used as the fused product data features.

[0063] Using big data technology, N products of the same category as those advertised are collected. Product data features of these N products are extracted and fused to obtain fused product data features. The similarity between the fused product data features and the fused product data features is calculated. The products corresponding to the top n similarity-ranked fused product data features are identified as competitors. The product categories include, but are not limited to, apparel, food, cosmetics, and furniture. The method for fusing product data features is the same as the method for fusing product data features, employing linear fusion with consistent weight allocation.

[0064] The method for extracting product data features is the same as that for extracting commodity data features. After data preprocessing, text data features are extracted using methods such as bag-of-words model, word embedding, and TF-IDF, while image data features are extracted using color histogram, edge detection, and convolutional neural network model.

[0065] The most recent ad data from n competitors is taken as the competitor ad data, meaning there are n competitor ad data.

[0066] The weights can be set by those skilled in the art or by the customer. If competitors focus on selecting products with similar functions, while neglecting factors such as product price, then the weight value corresponding to the data characteristics of product function will be greater when allocating weights. In this case, more emphasis will be placed on whether the function of the product is similar to that of the product, and the price will have a smaller impact on the selection of competitors. If competitors focus on selecting products with similar prices, while neglecting factors such as product function, then the weight value corresponding to the data characteristics of product price will be greater when allocating weights. In this case, more emphasis will be placed on whether the price of the product is similar to that of the product. If the price difference is too large, then the possibility of considering that product as a competitor is relatively small.

[0067] In this embodiment, it should be specifically explained that the competitor advertising acquisition module divides competitor advertising data into comparative advertising data and numerical advertising data, including the following steps:

[0068] Step S01: Encode the advertising data of n competitors, obtain chromosomes, and construct an initial population;

[0069] Step S02: Determine the fitness function;

[0070] Step S03: Perform natural selection on chromosomes in the population;

[0071] Step S04: Perform crossover recombination on chromosomes in the population;

[0072] Step S05: Mutate the chromosomes in the population;

[0073] Step S06: Obtain a new population. The preset population generation number is L, and the fitness threshold is Q, where L is an integer greater than 0 and Q is a real number greater than 0. Repeat steps S03-S05 until the new population generation number is L or a chromosome in the new population has a fitness value greater than or equal to the fitness threshold Q. The loop ends when the competitor's advertising data corresponding to the chromosome with the highest fitness in the new population is taken as the optimal advertising data. For example, if the preset population generation number is 1, then natural selection, crossover and recombination are performed on the chromosomes in the initial population to obtain a new population. At this time, the generation number of the new population is 1, so the loop ends.

[0074] Use the competitor's ad data corresponding to the best ad data as the comparison ad data, and use the remaining n-1 competitor's ad data as the numerical ad data;

[0075] Each competitor's ad data is coded as A, where A is a chromosome. B chromosomes are randomly generated to form the initial population C. , where C b Let b be the b-th chromosome, where b = 1, 2, 3, ..., B;

[0076] The fitness function is expressed as follows: ; where f b GX represents the fitness corresponding to the b-th chromosome. b The advertising performance of the competitor ads corresponding to the competitor ads data for the b-th chromosome;

[0077] , among which, SR b _gg represents the advertising revenue generated by competitor ads corresponding to the competitor ad data of chromosome b, and CB represents the advertising revenue generated by competitor ads. b _gg represents the advertising cost of the competitor's ad corresponding to the competitor's ad data for chromosome b; the advertising effect can also be obtained through ad click-through rate, ad conversion rate, and customer acquisition cost, etc. The customer acquisition cost is expressed as the total advertising expenditure divided by the number of new customers acquired, that is, the average cost required to acquire a new customer; the ad click-through rate is expressed as the number of ad clicks divided by the number of ad impressions, used to measure the ad's ability to attract the target audience to click; the ad conversion rate is expressed as the number of users who converted divided by the number of ad clicks, used to measure the proportion of users who completed a specific action after clicking the ad, the specific action including but not limited to purchase, registration, and download, etc.

[0078] In this embodiment, it should be specifically noted that the similarity between the fused commodity data features and the fused product data features can be calculated using suitable methods for obtaining data feature similarity, such as cosine similarity or Jaccard similarity coefficient. This embodiment uses cosine similarity for calculation, and the formula is expressed as:

[0079] ;

[0080] Wherein, cosθ i Let RH be the similarity between the i-th fused commodity data feature and the fused product data feature. i _sp is the ith fused product data feature vector, RH_cp is the fused product data feature vector, and ||RH i _sp‖ is the magnitude of the i-th fused product data feature vector,‖RH_cp‖ is the magnitude of the fused product data feature vector, and "∙" is the dot product, i=1,2,3,...,N;

[0081] The similarity between each fused product data feature and the fused product data feature is sorted. Since the cosine similarity ranges from -1 to 1, the closer it is to 1, the higher the similarity and the higher the ranking. The top n fused product data features are selected and their corresponding products are taken as competitors.

[0082] In this embodiment, it should be specifically noted that the competitor advertising analysis module obtains the advantages and mandatory options in the following way:

[0083] Similarity calculations are performed on the text and visual elements in numerical advertising data and comparative advertising data. Both text and visual elements contain several text and visual factors. The data items in the text element data are all text factors, such as text content, text font, and font format. Similarly, the data items in the visual element data are all visual factors, such as image color, image content, animation content, and video content. When text and visual elements are combined into an element set, where both text and visual factors are factors within the element set, then:

[0084] The comparison of advertising data is expressed as follows: , among which, U D To compare the set of advertising data elements, D j To compare the data characteristics of the j-th factor in the advertising data element set, m is the total number of factors in the element set, i.e., the sum of the number of copy factors and visual factors;

[0085] Numerical advertising data is represented as follows: , among which, U s_a Let s_a be the set of numerical advertisement data elements for the a-th element. j Let a be the data feature of the j-th factor in the a-th numerical advertising data element set, where a = 1, 2, 3, ..., n-1;

[0086] The factors in the comparative advertising data element set and the numerical advertising data element set are matched one-to-one.

[0087] Calculate the similarity between all factor data features in each numerical advertising data element set and all corresponding factor data features in the comparison advertising data element set. The similarity calculation can be performed by someone skilled in the art using a suitable data feature similarity calculation method, or cosine similarity can still be used. For visual factors, keyframes and audio features can be extracted from animations or videos to obtain corresponding data features as visual factor data features. Set a screening threshold YU1. If the similarity does not reach the screening threshold, the factor in the comparison advertising data element set is marked as an advantage. The screening threshold YU1 can be calculated by someone skilled in the art. The value range is set by the technicians themselves, and the value range satisfies YU1∈(-1,1). For example, if the factor in the first set of numerical advertising data elements and the set of comparative advertising data elements that does not reach the screening threshold is factor y1, then factor y1 is marked as an advantage item. If the advantage item is a factor in the comparative advertising data that is significantly different from the numerical advertising data, then the reason why the performance of the comparative advertising data is better than that of the numerical advertising data is mostly due to the existence of the advantage item, rather than common factors with excessive similarity. Therefore, selecting the advantage item is beneficial to bring the factor of the advantage item into the product advertising content, and to improve the performance of the product advertising content by utilizing the advantages of the existing advantage item.

[0088] The required options are factors that must exist in the generated advertising content. They can be elements that the customer requires in the advertising content from the customer demand data, or they can be obtained by analyzing the historical advertising data of the product and using the factors that appear most frequently in the historical advertising data as required options, or factors that appear more than a frequency threshold as required options. The frequency threshold is a frequency value that can be set by professionals in the field or by the customer.

[0089] In this embodiment, it should be specifically noted that during the intelligent generation and prediction module's intelligent generation of advertisements, the mandatory options are the content that must be reflected in the advertisement, and the advantageous options are the content that is automatically selected in the advertisement. Therefore, the intelligent generation of advertisements can be regarded as a permutation and combination process, that is, a process of combining the mandatory options and the advantageous options to generate P initial advertisements. When intelligently generating advertisements, artificial intelligence generation tools such as AdCreative.ai, Sivi.ai, and QuickAds can be used for intelligent advertisement generation.

[0090] P initial ads are sequentially input into the constructed ad prediction model to predict the corresponding delivery effect. The input layer of the ad prediction model has P nodes for inputting the P initial ads, and the output layer also has P nodes for outputting the delivery effect of the P initial ads. The initial ad with the best predicted delivery effect is selected as the candidate ad and transmitted to the content optimization module for subsequent operations.

[0091] The training process of the advertising prediction model is as follows:

[0092] Beforehand, collect advertising content data of competitors in group R1 and historical advertising content data of products in group R2, and use them as analysis data. Then there are a total of (R1+R2) groups of analysis data, where R1 and R2 are both integers greater than 1. Transform a group of analysis data and its corresponding campaign performance into a corresponding set of feature vectors. The campaign performance can be calculated in the same way as the advertising performance in the fitness function, or it can be calculated through advertising click-through rate, advertising conversion rate, and customer acquisition cost, etc.

[0093] Each set of feature vectors is used as input to the advertising prediction model. The advertising prediction model outputs a set of delivery effects corresponding to each set of analysis data, and aims at the actual delivery effect corresponding to each set of analysis data. The actual delivery effect is the pre-collected delivery effect corresponding to the analysis data. The training objective is to minimize the sum of prediction errors of all analysis data. The formula for the prediction error is expressed as: , where ε q The prediction error is represented by q, where q is the group number of the feature vector corresponding to the analyzed data, and θ is the prediction error. q For the q-th set of analysis data, μ represents the delivery effect. q To determine the actual delivery effect corresponding to the qth set of analysis data, the advertising prediction model is trained until the sum of prediction errors converges and training stops.

[0094] The advertising prediction model is a deep neural network model, which includes an input layer, a hidden layer, and an output layer. Each hidden layer contains multiple neurons, and each neuron is connected to the neurons in the next layer. The connections contain weights that determine the importance and influence of data transmitted in the neural network. An activation function is applied to each neuron between the hidden layer and the output layer. The activation function introduces non-linearity, allowing the network to learn more complex patterns and features.

[0095] In this embodiment, it should be specifically noted that the content optimization module includes an ad modification unit and an element bar unit. The ad modification unit is used by the customer to modify the content of candidate ads. After the candidate ads are displayed on the operating terminal, the customer can modify them. In the ad content of the candidate ads, each text element and visual element is a separate modification item, and the customer can click on the modification item to replace or modify it. The element bar unit is used to display the text elements and visual elements available for replacement and modification.

[0096] After the client completes the modifications, the revised ad is saved and sent back to the intelligent generation and prediction module. It is then input into the ad prediction model to predict the ad's performance. The modified ad's performance is compared with that of alternative ads. If the modified ad performs better, it is directly replaced with the alternative ad and sent back to the content optimization module. If the alternative ad performs better, it is not replaced, and the modified ad's performance is displayed in the interaction module. The interaction module then queries the client again, asking whether to replace the alternative ad. If the client knows the performance will be lower but still wants to use the modified ad, their request is met. If the modified ad performs better than the alternative ad, it is directly replaced, satisfying the client's needs while achieving better performance.

[0097] The content optimization module can also perform professional checks on the intelligently generated advertising content. Professional personnel can modify or delete harmful content and negative opinions in the advertising content to ensure the quality of the advertising content.

[0098] In this embodiment, it should be specifically explained that the elite method generates F1 offspring chromosomes. For a population with a size of B, the fitness of the B chromosomes is arranged from largest to smallest, and each of the first F1 chromosomes generates one offspring chromosome. The rotation method generates F2 offspring chromosomes, that is, R chromosomes generate F2 offspring chromosomes according to the corresponding rotation probability. F1 + F2 = B, so as to keep the offspring population size B unchanged and the number of generations of the population increases.

[0099] The expression for the rotation probability is: , where ζ b Let be the rotation probability of the competitor's advertising data corresponding to the b-th chromosome;

[0100] The crossover recombination is performed using the PMX method. E chromosomes are randomly selected from the population for crossover recombination to obtain E new chromosomes. After chromosome crossover recombination, the fitness of the E new chromosomes is calculated. The fitness of the E new chromosomes is then sorted from largest to smallest with the fitness of the original E chromosomes to generate a sorting table. The E new chromosomes from the sorting table replace the E chromosomes used in the crossover recombination in the population in ascending order. In this embodiment, E is preferably 0.7B. If the calculated E is not an integer, it is rounded up to ensure that the calculated E is an integer.

[0101] The preset mutation probability is V. Based on the mutation probability, B chromosomes in the population are mutated. The mutation method is to randomly select the positions of two genes in the chromosome and exchange the values ​​of the two genes. In this embodiment, V=0.02 is preferred. The mutation probability is preset by those skilled in the art based on the algorithm efficiency and algorithm accuracy.

[0102] In conclusion, 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 within the protection scope of the present invention.

[0103] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A big data-based intelligent advertising content generation and optimization system, characterized in that: It includes a customer data collection module, a competitor advertising acquisition module, a competitor advertising analysis module, an intelligent generation and prediction module, a content optimization module, and an interaction module; The customer data collection module is used to collect customer demand data and product data, and transmit them to the competitor advertising acquisition module and the competitor advertising analysis module. The competitor advertising acquisition module is used to receive data from the customer data collection module, acquire competitor advertising data, and then transmit it to the competitor advertising analysis module after dividing it into comparative advertising data and numerical advertising data. The competitor advertising analysis module is used to receive data from the competitor advertising acquisition module and the customer data collection module, analyze the numerical advertising data and comparative advertising data, and obtain the advantages and mandatory items. The intelligent generation and prediction module generates advertisements intelligently based on advantages and mandatory options, then inputs them into the advertisement prediction model to predict the advertising effect, selects the advertisement with the best performance as the candidate advertisement, and then transmits the candidate advertisement and the predicted performance to the content optimization module. The content optimization module outputs the candidate ads and the predicted delivery effect to the interaction module for display. The customer can modify the ad content through the content optimization module. After the modification is completed, the modified ad is fed back to the intelligent generation and prediction module for delivery effect prediction, and then the prediction result is sent back to the content optimization module. If the client does not need to modify the ad content, the alternative ads will be directly used as the final generated ads and transmitted to the interaction module for display. The interaction module is used to display the final generated advertisement on the human-computer interaction interface; The competitor advertising acquisition module divides competitor advertising data into comparative advertising data and numerical advertising data, including the following steps: Step S01: Encode the advertising data of n competitors, obtain chromosomes, and construct an initial population; Step S02: Determine the fitness function; Step S03: Perform natural selection on chromosomes in the population; Step S04: Perform crossover recombination on chromosomes in the population; Step S05: Mutate the chromosomes in the population; Step S06: Obtain a new population, with a preset population generation number of L and a fitness threshold of Q, where L is an integer greater than 0 and Q is a real number greater than 0; repeat steps S03-S05 until the new population reaches generation number L or a chromosome in the new population has a fitness value greater than or equal to the fitness threshold Q, then the loop ends, and the competitor's advertising data corresponding to the chromosome with the highest fitness in the new population is taken as the optimal advertising data; The competitor's ad data corresponding to the best ad data is used as the comparison ad data, and the remaining n-1 competitor's ad data are used as the numerical ad data.

2. The intelligent advertising content generation and optimization system based on big data according to claim 1, characterized in that: The customer demand data includes the advertising content, copywriting elements, and visual elements that customers expect to be presented; the advertising content includes product promotion content, service promotion content, brand image content, and promotional activity content; the product promotion content includes product features, usage methods, product advantages and competitor comparisons, and product pricing; the service promotion content includes service content, service process and timeliness, service advantages and features, and customer reviews; the brand image content includes brand story and history, brand values ​​and philosophy, brand partners and honors, and brand public welfare activities and responsibilities; the promotional activities include limited-time discounts and offers, holiday promotions, points redemption and gift giveaways, and joint promotions with partners; The text element data includes text content, text font, font format, and font layout; the visual element data includes image color, image content, animation content, and video content; the product data is data related to the advertised product, including product name, product image, product function, product price, product sales revenue, product category, and product historical data; the product historical data includes historical sales quantity, sales amount, sales duration, and historical advertising data; the advertising data includes advertising content data, advertising conversion rate, advertising click-through rate, and advertising exposure rate; the product is the advertised product; the advertising content data includes text elements and visual elements within the advertising content.

3. The intelligent advertising content generation and optimization system based on big data according to claim 2, characterized in that: The specific method by which the competitor advertising acquisition module acquires competitor advertising data is as follows: Product data features are extracted and fused. The feature fusion adopts linear fusion, which uses a linear combination to fuse different features. New fused features are obtained through weight allocation, and the new fused features are used as the fused product data features. Using big data technology, collect data on N products of the same category as those advertised. Extract the product data features of the N products and then merge them to obtain the merged product data features. Calculate the similarity between the merged product data features and the merged product data features. Select the products corresponding to the top n merged product data features with the highest similarity as competitors. The most recent ad data from each of the n competitors is taken as the competitor ad data, meaning there are n competitor ad data.

4. The intelligent advertising content generation and optimization system based on big data according to claim 3, characterized in that: In step S01, each competitor's advertising data is encoded as A, where A is a chromosome. B chromosomes are randomly generated to form the initial population C. , where C b Let b be the b-th chromosome, where b = 1, 2, 3, ..., B; The fitness function is expressed as follows: ; where f b GX represents the fitness corresponding to the b-th chromosome. b The advertising performance of the competitor ads corresponding to the competitor ads data for the b-th chromosome; , among which, SR b _gg represents the advertising revenue generated by competitor ads corresponding to the competitor ad data of chromosome b, and CB represents the advertising revenue generated by competitor ads. b _gg represents the advertising cost of the competitor's ad corresponding to the competitor's ad data for the b-th chromosome.

5. The intelligent advertising content generation and optimization system based on big data according to claim 4, characterized in that: The similarity between the merged product data features and the merged commodity data features is calculated using cosine similarity, expressed by the following formula: ； Wherein, cosθ i Let RH be the similarity between the i-th fused commodity data feature and the fused product data feature. i _sp is the ith fused product data feature vector, RH_cp is the fused product data feature vector, and ||RH i _sp‖ is the modulus of the i-th fused product data feature vector,‖RH_cp‖ is the modulus of the fused product data feature vector, "∙" is the dot product, i=1,2,3,...,N.

6. The intelligent advertising content generation and optimization system based on big data according to claim 5, characterized in that: The competitive advertising analysis module obtains the advantages and mandatory options in the following way: Similarity is calculated between text and visual elements in numerical advertising data and comparative advertising data. The data items in the text element data are all text factors, and the data items in the visual element data are all visual factors. Text and visual elements are combined into an element set, where both text and visual factors are factors within that element set. Then: The comparison of advertising data is expressed as follows: , among which, U D To compare the set of advertising data elements, D j To compare the data characteristics of the j-th factor in the advertising data element set, m is the total number of factors in the element set, i.e., the sum of the number of copy factors and visual factors; Numerical advertising data is represented as follows: , among which, U s_a Let s_a be the set of numerical advertisement data elements for the a-th element. j Let a be the data feature of the j-th factor in the a-th numerical advertising data element set, where a = 1, 2, 3, ..., n-1; The factors in the comparative advertising data element set and the numerical advertising data element set are matched one-to-one. Calculate the similarity between all factor data features in each numerical advertising data element set and all corresponding factor data features in the comparison advertising data element set. Set a screening threshold YU1. If the similarity does not reach the screening threshold, mark the factor in the comparison advertising data element set as an advantage. The required options are factors that must be present in the generated advertising content.

7. The intelligent advertising content generation and optimization system based on big data according to claim 6, characterized in that: The intelligent generation and prediction module generates P initial advertisements, which are then sequentially input into the constructed advertising prediction model to predict the corresponding advertising performance. The input layer of the advertising prediction model has P nodes for inputting the P initial advertisements, and the output layer also has P nodes for outputting the advertising performance of the P initial advertisements. The initial advertisement with the best predicted advertising performance is selected as the candidate advertisement.

8. The intelligent advertising content generation and optimization system based on big data according to claim 7, characterized in that: The training process of the advertising prediction model is as follows: Beforehand, collect advertising content data of competitors in group R1 and historical advertising content data of products in group R2, and use them as analysis data. Then there are a total of (R1+R2) sets of analysis data, where R1 and R2 are both integers greater than 1. Transform a set of analysis data and its corresponding campaign performance into a set of feature vectors. Each set of feature vectors is used as input to the advertising prediction model. The advertising prediction model outputs a set of delivery effects corresponding to each set of analysis data, and aims at the actual delivery effect corresponding to each set of analysis data. The actual delivery effect is the pre-collected delivery effect corresponding to the analysis data. The training objective is to minimize the sum of prediction errors of all analysis data. The formula for the prediction error is expressed as: , where ε q The prediction error is represented by q, where q is the group number of the feature vector corresponding to the analyzed data, and θ is the prediction error. q For the q-th set of analysis data, μ represents the delivery effect. q To determine the actual delivery effect corresponding to the qth set of analysis data, the advertising prediction model is trained until the sum of prediction errors converges and training stops. The advertising prediction model is a deep neural network model, which includes an input layer, a hidden layer, and an output layer. Each hidden layer contains multiple neurons, and each neuron is connected to the neurons in the next layer, with weights included in the connections.

9. The intelligent advertising content generation and optimization system based on big data according to claim 8, characterized in that: The content optimization module includes an ad modification unit and an element bar unit. The ad modification unit is used by the customer to modify the content of candidate ads. After the candidate ads are displayed on the operation terminal, the customer can modify them. In the ad content of the candidate ads, each text element and visual element is a separate modification item. The customer can click on the modification item to replace or modify it. The element bar unit is used to display the text elements and visual elements available for replacement and modification. After the client completes the modifications, they save the revised ad and send it back to the intelligent generation and prediction module. The module then inputs the modified ad into the ad prediction model to predict its performance. The modified ad's performance is compared to that of alternative ads. If the modified ad performs better, it replaces the alternative ad and is sent back to the content optimization module. If the alternative ad performs better, it is not replaced, and the modified ad's performance is displayed in the interaction module. The interaction module then queries the client again to determine whether to replace the alternative ad.

Images