Advertisement putting effect optimization method based on big data

Abstract

The invention relates to the technical field of big data and digital advertisements, and particularly discloses an advertisement putting effect optimization method based on big data. According to the method, an advertisement creative gene pool is constructed, copywriting, color matching, layout and background music are coded into gene segments, and iterative evolution and self-optimization of advertisement materials are achieved through creative population initialization, dynamic feedback evaluation, natural selection, gene cross recombination and an environment pressure driven mutation mechanism. The system updates fitness scores in real time based on multi-source user behavior data, dynamically adjusts mutation probability in combination with external environment factors, and introduces a high-quality gene pool to accelerate convergence and ensure compliance. According to the technical scheme, creative fatigue is delayed, the novelty and conversion efficiency of advertisements are improved, and double optimization of the marketing effect and the user experience is achieved.

Description

Technical Field

[0001] This invention belongs to the field of big data and digital advertising technology, specifically relating to a method for optimizing advertising performance based on big data. Background Technology

[0002] With the deep integration of big data and artificial intelligence technologies, the digital marketing field has entered a new stage of automation and intelligence, with precise ad placement becoming a core driver for improving business conversion efficiency. In large-scale online advertising systems, through in-depth mining and analysis of massive amounts of user behavior data, the system can create precise profiles of the target audience and match more relevant marketing content to different user groups. As a key factor determining the effectiveness of ad reach, the continuous optimization of placement strategies directly relates to the utilization rate of traffic resources and the improvement of the overall marketing return on investment.

[0003] The automated generation and dynamic optimization of advertising creative materials are key research areas for improving advertising performance. This process typically involves parametric modeling of multiple creative dimensions, including copywriting, visual images, layout configuration, and interaction logic, and utilizes machine learning algorithms to predict the performance of massive combinations of creative materials and perform real-time filtering. By establishing a feedback loop, the system aims to identify high-quality combinations with high click-through rates and high conversion potential from a vast pool of candidate creative materials, adapting to the ever-changing market environment and user preferences.

[0004] Traditional advertising optimization algorithms generally face the serious problem of creative fatigue and decay. Creatives often stagnate after reaching a local optimum, and static filtering mechanisms struggle to cope with audience fatigue caused by prolonged exposure. When dealing with a high-dimensional feature space composed of multiple elements, existing methods suffer from a significant combinatorial explosion problem. The extremely high computational complexity not only consumes substantial computing resources but also limits the system's ability to discover deep cross-boundary creative combinations. Existing optimization models lack real-time evolutionary responses to dynamic environmental pressures and cannot simulate the dynamic process of natural selection, resulting in creatives lacking continuous self-iteration capabilities and struggling to maintain long-term competitiveness in a fiercely competitive market. Therefore, a big data-based method for optimizing advertising performance is desired. Summary of the Invention

[0005] The purpose of this invention is to provide a method for optimizing advertising performance based on big data, which can solve the problems mentioned in the background.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a method for optimizing advertising performance based on big data, comprising the following specific steps:

[0007] Step 1: Construct an advertising creative gene library, encode advertising copy, visual color scheme, page layout and background music into independent gene fragments, and assign a unique identifier and initial fitness score to each gene fragment; Step 2: Initialize the creative population by randomly selecting multiple gene fragments from the advertising creative gene library and combining them to form several initial advertising creative individuals, each individual corresponding to a complete advertising material scheme; Step 3: Deploy a dynamic feedback evaluation mechanism to monitor the click-through rate, conversion rate, and dwell time of each individual ad creative in the creative population in real time based on real user behavior data, and update the fitness score of each individual accordingly; Step 4: Perform natural selection. Sort the creative population according to the fitness score, retain individuals with scores higher than a preset threshold as the parent population, and eliminate the rest of the individuals. Step 5: Perform gene cross-recombination. Randomly select two individuals from the parent population and perform fragment-level exchange on their gene segments to generate new offspring advertising creative individuals. Step 6: Introduce an environmental stress-driven mutation mechanism. During the generation of offspring individuals, dynamically adjust the mutation probability according to changes in external environmental factors, and randomly replace or perturb some gene segments. Step 7: Iterative evolution and population update. The offspring individuals and the surviving parent individuals together form a new generation of creative population, and steps 3 to 6 are repeated until the preset termination conditions are met.

[0008] Preferably, in step 1, the advertising copy gene fragment adopts semantic vector encoding to map keywords, sentiment tendencies, and sentence structures into numerical sequences of fixed dimensions; the visual color matching gene fragment is parametrically described through three dimensions: primary color, secondary color ratio, and color contrast; the page layout gene fragment defines the position, size, and hierarchical relationship of elements using a grid division method; and the background music gene fragment is structurally represented based on rhythm intensity, melody type, and emotional tags.

[0009] Preferably, in step 2, the size of the initial creative population is dynamically set according to the available computing power resources, and the gene combination of each initial advertising creative individual must meet the preset business compliance constraints, including but not limited to brand tone consistency, industry standard restrictions and platform content security rules.

[0010] Preferably, in step 3, the user behavior data comes from a multi-channel integrated log stream, covering display exposure, click interaction, page jump and final conversion events, and recent behaviors are given higher weight through a sliding time window mechanism to reflect the timeliness changes of user preferences.

[0011] Preferably, in step 4, the fitness score is calculated by comprehensively considering short-term conversion efficiency and long-term user value. The short-term indicators include instant click-through rate and conversion rate, and the long-term indicators include user retention rate and repurchase tendency. The final score is generated by a weighted fusion formula, and the weighting coefficient can be dynamically configured according to marketing objectives.

[0012] Preferably, in step 5, the gene cross-recombination operation is performed independently in four dimensions: text, color scheme, layout, and background music. Each recombination only selects one or more dimensions to perform fragment exchange, and the exchange position is determined by random sampling to ensure the diversity and unpredictability of the recombination results.

[0013] Preferably, in step 6, the external environmental factors include the intensity of competitor advertising campaigns, the popularity of trending topics on social media, seasonal consumption trends, and macroeconomic indicators. The system quantifies the impact of each factor on the current market environment by crawling public data sources in real time and performing semantic analysis, and adjusts the mutation probability accordingly, so that the creative population can proactively adapt to external changes.

[0014] Preferably, in step 7, the termination conditions include reaching the maximum number of generations, the average fitness improvement of the population over multiple consecutive generations being less than a preset threshold, or the system detecting a structural mutation in the market environment requiring re-initialization of the population.

[0015] Preferably, the present invention also includes archiving and storing high-fitness gene fragments generated during historical evolution to form a high-quality gene pool, and preferentially extracting fragments from it for the construction of new individuals in subsequent evolutionary cycles, thereby accelerating the convergence process and retaining the creative elements that have been verified.

[0016] Preferably, the present invention also includes a secondary compliance verification of candidate creative individuals before each ad placement to ensure that their content does not violate the latest laws and regulations or platform policies. The verification rule base supports dynamic updates and is seamlessly integrated with the evolution system.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a digital creative ecosystem that simulates biological evolution, enabling advertising materials to have the ability to continuously self-evolve. This overcomes the stagnation problem of traditional optimization algorithms at local optima and slows down the rate of creative fatigue decay.

[0018] 2. This invention utilizes natural selection to efficiently prune ineffective combinations in high-dimensional feature space, significantly reducing computational resource consumption. Through gene crossover and environment-driven mutation, it discovers cross-disciplinary creative combinations that are difficult for human designers to intuitively conceive, enhancing the novelty and competitiveness of advertising content. The introduction of dynamic feedback evaluation and environmental pressure factors enables the system to respond in real time to market changes and user preference shifts, ensuring that the advertising strategy is always in the optimal state, maximizing the conversion efficiency and commercial value of marketing investment while protecting user experience. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall technical solution architecture according to the present invention; Figure 2 This is a schematic diagram of the core principle framework of creative optimization based on simulated biological evolution mechanism according to the present invention; Figure 3 This is a flowchart illustrating the logical process of constructing an advertising creative gene bank and encoding multidimensional genes according to the present invention. Figure 4 This is a flowchart illustrating the dynamic update process of fitness scores based on user behavior feedback according to the present invention. Figure 5 This is a schematic diagram of the multi-level interaction logic of the mutation mechanism driven by external environmental pressure according to the present invention. Detailed Implementation

[0020] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0021] In the execution of the big data-based advertising performance optimization method of this invention, the underlying data management architecture is first activated to support the fragmented storage and structured evolution of large-scale advertising materials. The method specifically includes the following steps: In step 1, an advertising creative gene library is constructed. This process involves the in-depth deconstruction of massive amounts of original advertising materials. Specifically, the system pre-establishes a distributed multimodal storage cluster to decouple the collected advertising copy, visual color schemes, page layout templates, and background music materials.

[0022] For the advertising copy gene fragment, semantic vector encoding is performed. The system calls a pre-trained natural language processing model to perform word segmentation and part-of-speech tagging on the input short text or long case copy. Subsequently, the extracted keywords, implied sentiment values, and sentence structure features are input into a high-dimensional feature mapping module. This high-dimensional feature mapping module transforms this unstructured information into a set of fixed-dimensional numerical sequences, where each value represents the weight of the copy in a specific semantic dimension. The copy gene fragment not only includes the text content itself but also includes language identification, word count, and initial screening status bits for sensitive word compliance.

[0023] For visual color matching gene fragments, the system samples image materials using a color space extraction algorithm. The dominant hue is defined as the color region with the highest pixel proportion in the image, and its features are quantified using red-green-blue color space coordinates or hue, saturation, and brightness coordinates. The secondary color proportion describes the distribution density of secondary colors other than the dominant hue in the image. Color contrast is obtained by calculating the brightness difference between adjacent color blocks and the hue ring span. These parameters are combined into a multi-dimensional floating-point vector, serving as the physical representation of the visual gene.

[0024] For page layout gene fragments, the system adopts a topological structure description method based on a grid system. The advertising display area is divided into several equally divided logical grids, defining the starting row index, starting column index, number of rows occupied, number of columns occupied, and hierarchy depth of each visual element in the grid coordinate system. Each layout gene fragment is actually a configuration list describing the arrangement and combination logic of page elements.

[0025] For background music gene fragments, the system performs an audio feature extraction process. The audio signal is converted from the time domain to the frequency domain using a Fast Fourier Transform (FFT), the average rhythm intensity is calculated, and the basic frequency response of the melody is identified. An audio classification model is then used to label the music with emotional characteristics. These features, after structural processing, form music gene fragments containing rhythmic pulse sequences and frequency feature values.

[0026] After extracting the four types of gene fragments mentioned above, the system generates a globally unique identifier for each gene fragment. This identifier consists of a timestamp, a physical node number, and an auto-incrementing sequence. The initial fitness score is set to a preset baseline value, such as 0.5, to provide an initial selection reference in the early stages of evolution.

[0027] In step 2, the creative population is initialized. The system retrieves gene fragments of different categories from the advertising creative gene library using a random number generator. The system dynamically sets the initial population size, i.e., the total number of individuals, based on the current computing power resource configuration. The generation process of each advertising creative individual is as follows: the system randomly selects one or more gene fragments with compatibility identifiers from the copywriting library, color scheme library, layout library, and music library, and encapsulates these fragments into a complete advertising material scheme description file according to a predefined splicing protocol.

[0028] During this process, the system enforces business compliance checks. The verification module performs logical consistency audits on the generated individuals. For example, it checks whether the product description in the copy matches the image area size in the page layout; whether the brand color scheme conforms to the preset brand visual manual specifications; and whether the background music duration covers the estimated advertising display period. Only individuals that pass all compliance checks are admitted to the initial creative population. Unqualified individuals are marked and discarded, triggering a re-extraction process until the preset population size is reached.

[0029] In step 3, a dynamic feedback evaluation mechanism is deployed. Once an individual from the creative population is pushed to the advertising platform, the system activates a real-time data collection engine. This engine captures real-time behavioral feedback from the user side using multi-channel integrated log streaming technology. The data covers display exposure events, click interactions, page jump depth, and ultimately, purchase or registration conversion behaviors.

[0030] To ensure the real-time nature and accuracy of the evaluation, the system introduces a sliding time window mechanism. Specifically, the system divides the past 24 hours into several equal time slices. Behavioral data occurring within time slices closer to the current moment is assigned a higher decay weighting coefficient. When calculating the overall performance of each individual ad creative, the system counts the click frequency and conversion frequency of that ad creative within each time slice, multiplies these frequencies by their corresponding weighting coefficients, and then sums them. This approach can sensitively reflect instantaneous shifts in user preferences. For example, in the last hour of a promotional campaign, users' sensitivity to discount information increases; the sliding time window mechanism can quickly capture this change and reflect it in the fitness score update.

[0031] In step 4, natural selection is performed. The system periodically triggers the ranking engine to summarize the fitness scores of all individuals in the current population. The fitness score calculation process integrates short-term conversion efficiency with long-term user value. The weighting coefficients involved in the calculation can be adjusted according to the current marketing strategy. For example, during the brand promotion phase, the weights of click-through rate and dwell time are increased; during the harvest phase, the weights of conversion rate and repurchase intention are significantly increased.

[0032] After sorting, the system filters the population based on a preset selection threshold. This threshold can be a fixed percentage or a dynamic boundary based on the mean fitness. The retained individuals are formally labeled as the parent population, representing the best-performing gene combinations in the current environment. The eliminated individuals are removed from the in-memory active queue, and their data is transferred to an offline analysis module to mine feature patterns from these failed cases.

[0033] In step 5, gene crossover and recombination are performed. This is a crucial step in simulating the biological genetic process. The system selects paired parental individuals from the parent population using a random sampling algorithm. For each selected pair of parental individuals, the system initiates fragment-level exchange logic. The recombination operation is independent of each other in four dimensions: text, color scheme, layout, and background music.

[0034] The specific operation process is as follows: The system generates a mask vector containing four binary bits, each bit corresponding to a dimension. If the mask bit is 1, gene swapping occurs along that dimension; if it is 0, it remains unchanged. In this way, the system may generate new offspring individuals that possess the text and layout of parent A, and the color scheme and background music of parent B. The specific location of the swap (i.e., the cut point in the gene sequence) is also determined by random numbers. For example, for a text gene containing multiple phrases, recombination may occur at a certain phrase in the middle. This recombination mechanism can break the original fixed combination, explore new high-conversion feature coupling points, and increase the diversity of creativity.

[0035] In step 6, an environmental pressure-driven mutation mechanism is introduced. To enable the creative population to proactively adapt to market fluctuations, an environmental perception subsystem is integrated into the system. This environmental perception subsystem acquires external environmental factor data through a real-time web crawler interface, including but not limited to competitor advertising density data on major media channels, the popularity values ​​of the top 100 trending topics on mainstream social platforms, the consumption trend index corresponding to the current season, and macroeconomic fluctuation indicators.

[0036] The environmental perception subsystem performs semantic clustering on the acquired text data and normalizes the numerical data. The system pre-defines a mutation probability mapping function: when the volatility of external environmental factors increases, the system automatically raises the mutation probability baseline value. At the binary stream level of offspring generation, mutation manifests as the random replacement of an identifier in a gene segment or a small random perturbation of parameters within a gene segment. This externally pressure-driven mutation enables the population to generate groundbreaking new characteristics when facing drastic market changes, avoiding aesthetic fatigue caused by excessive convergence.

[0037] In step 7, iterative evolution and population renewal occur. The system merges the offspring individuals generated in step 5 with the high-quality parent individuals retained in step 4 to form a completely new population. This population then re-enters the dynamic feedback evaluation process of step 3, undergoing a new round of real-world market testing.

[0038] The system has three types of termination conditions: the first is reaching the preset maximum number of generations (e.g., 500 consecutive generations); the second is when the performance tends to stabilize, i.e., the average fitness score of the population increases less than the preset minimum value for 5 consecutive generations; the third is triggered by a sudden event, i.e., when the environmental perception subsystem detects a major structural change in the market environment (such as a sudden public health event or a major adjustment in industry policies), the system actively terminates the current iteration and forces the population to be reinitialized.

[0039] In a preferred implementation of this embodiment, the system also includes an archiving mechanism for a high-quality gene pool. Whenever an individual's fitness score exceeds 90% of its historical high, the gene fragments it contains and their corresponding combination parameters are permanently stored in a high-performance read-only database. In subsequent new population initialization or mutation operations, the system will preferentially extract elements from this high-quality gene pool with a high probability. This is equivalent to providing "experience memory" for the evolutionary process, which can shorten the optimization cycle after new products are launched.

[0040] This embodiment also includes a pre-launch compliance secondary verification module. This module is linked in real-time with the latest legal and regulatory database. Before the final ad binary file is distributed to CDN nodes, the verification engine checks again whether the copy contains the latest banned words and whether the images comply with the copyright policies of the specific platform. This double-insurance mechanism ensures that while the evolved "creative monster" is commercially attractive, it remains under control from a legal and compliance perspective.

[0041] Example 2: In Example 2, the present invention provides an advertising optimization scheme for a specific application scenario of a short video social platform. This advertising optimization scheme, based on Example 1, represents a significant technical extension tailored to the characteristics of video stream advertising.

[0042] In the construction phase of step 1, the video script gene fragment is refined into three sub-fragments: the golden hook phrase in the first 3 seconds, the description of the benefits in the middle, and the call to action at the end. Each sub-fragment is encoded as a vector containing emotional intensity values ​​and keyword weights. The visual gene not only includes static color schemes but also adds dynamic shot cut frequency, special effects intensity, and contrast dynamic change curves. The background music gene adds audio-visual alignment and synchronization feature points to describe the matching degree between music beats and visual transitions.

[0043] In the feedback evaluation of step 3, in addition to click-through rate and conversion rate, the system forcibly introduces "completion rate" as a core evaluation indicator. Specifically, the system collects user bounce rates at each 1 / 100th of the playback progress through the playback progress monitoring interface. If a large number of users bounce within the first 5 seconds of the video, the system will automatically reduce the fitness weights of the "first 3 seconds text gene" and "initial screen gene" for that individual.

[0044] In the mutation mechanism described in step 6, the system integrates with the trend API of short video platforms. When a specific visual filter or background music suddenly becomes wildly popular on the platform, environmental stressors are quickly fed back to the evolutionary engine. The system increases the mutation activity of specific gene segments, guiding offspring individuals to align with the trend. For example, the system automatically inserts the most popular filter parameters into the visual genes of offspring individuals.

[0045] In the cross-recombination step 5, Example 2 employed an asymmetric cross-recombination strategy. For parent elements with exceptional performance, the system retains their core visual skeleton, recombining only the background music and subtitle text. This protective mechanism prevents excessive recombination from disrupting the already established excellent visual rhythm.

[0046] In the termination condition judgment of step 7, the system adds "creative half-life" monitoring. When it detects that the conversion efficiency of an individual drops sharply and non-linearly with the increase of the number of impressions, it is determined that the individual has entered the fatigue period, and regardless of how high its fitness score is, it is forced to enter the elimination or mutation procedure.

[0047] Example 3: In Example 3, the present invention demonstrates implementation details under a distributed edge computing architecture. This example is primarily used to address network latency and data synchronization issues during multi-regional, global deployment.

[0048] In step 2, the initial population is distributed to edge computing nodes located in different geographical locations. Each edge node independently runs the display and feedback collection work for a small subset of the population individuals. This allows for the formation of subpopulations with locally optimal characteristics based on the cultural preferences of different regions.

[0049] In step 3, each edge node periodically compresses the user behavior features it collects and uploads them to the central server. The central server then performs a global fitness aggregation. To reduce communication overhead, the uploaded data is not the raw log, but rather moment features that have undergone preliminary local aggregation, including statistical features such as the mean and variance of click counts.

[0050] In the gene crossover process of step 5, the system introduces a "migration mechanism." Every certain number of generations, the central server selects gene fragments from individuals that perform exceptionally well in region A and randomly inserts them into the population of region B. This cross-regional gene flow helps break local optima and introduce new creative inspiration, making it particularly suitable for marketing campaigns of global brands.

[0051] In step 6, regarding environmental stress handling, Example 3 configured independent regional environmental parameters for each edge node. For example, the mutation probability of Southeast Asian nodes is affected by local religious festivals and climate change, while European nodes are more influenced by the popularity of local sporting events. This differentiated mutation strategy allows the same advertising system to generate creative content that best suits local aesthetics in different cultural contexts.

[0052] All numerical computation logic involved in the above embodiments of the present invention is executed in the background by a logic processor. For example, the fitness score update logic is manifested at the computer level as follows: obtaining the original score stored in the first register, obtaining the real-time feedback weight stored in the second register, performing a multiplication and accumulation operation, and then writing the result back to the individual feature vector region in memory. All comparison operations, such as threshold filtering, are completed by a comparator driven by clock pulses. These low-level machine actions together realize the above-mentioned complex biological evolution simulation process.

[0053] The encoding methods, grid partitioning methods, semantic vector dimensions, and other technical details described in the specific implementation can all be flexibly adjusted according to specific server performance indicators during actual engineering deployment. For example, when using a high-performance computing cluster, the dimension of the semantic vector can be expanded from 256 dimensions to 2048 dimensions to obtain more refined emotion capture capabilities; in real-time generation scenarios on mobile devices, quantized 8-bit integer vectors can be used to improve computational efficiency.

[0054] In each of the above steps, data transfer between different modules follows a standard asynchronous communication protocol. Each time an S-step completes its logical processing, it generates a message packet containing the processing result status bits, which is pushed to the input buffer of the next step via a high-speed message bus. This decoupled design ensures the system's concurrent stability when processing millions of gene fragments.

[0055] For storing gene fragments, the system employs a combination of key-value database and object storage. Feature vectors and identifiers of gene fragments are stored in an in-memory database to achieve sub-millisecond retrieval responses; while actual text, high-definition images, and audio / video materials are stored in a backend distributed object storage system. This separation of static and dynamic elements ensures extremely high performance during evolutionary computation.

[0056] In terms of system security design, this embodiment also includes an anomaly monitoring thread. This thread continuously scans the characteristic distribution of individuals in the population. If it finds that a gene segment has garbled text, logical breaks, or illegal characters due to excessive mutation, the monitoring thread will immediately trigger a circuit breaker mechanism, withdraw the release of that individual, and record an anomaly log for algorithm optimization reference.

[0057] This big data-based evolutionary method can handle not only single-media advertisements but also dynamic H5 advertisements that include interactive logic. In the H5 scenario, layout gene fragments are expanded into interactive genes that include event-triggered logic, such as defining animated feedback when a user clicks on different areas of the screen. These interactive genes also participate in crossover, mutation, and selection processes, evolving into the interactive flow that best guides users to complete conversion actions.

[0058] Through the above multi-dimensional embodiments, it can be seen that this invention constructs a continuously self-iterable productivity system in the field of digital advertising by simulating the core logic of biological evolution. This systematic and engineered approach changes the traditional material production model that relies on human experience, transforming the creative generation process into a controllable evolutionary process driven by data.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing advertising performance based on big data, characterized in that, Includes the following steps: Step 1: Construct an advertising creative gene library, encode advertising copy, visual color scheme, page layout and background music into independent gene fragments, and assign a unique identifier and initial fitness score to each gene fragment; Step 2: Initialize the creative population by randomly selecting multiple gene fragments from the advertising creative gene library and combining them to form several initial advertising creative individuals, each individual corresponding to a complete advertising material scheme; Step 3: Deploy a dynamic feedback evaluation mechanism to monitor the click-through rate, conversion rate, and dwell time of each individual ad creative in the creative population in real time based on real user behavior data, and update the fitness score of each individual accordingly; Step 4: Perform natural selection. Sort the creative population according to the fitness score, retain individuals with scores higher than a preset threshold as the parent population, and eliminate the rest of the individuals. Step 5: Perform gene cross-recombination. Randomly select two individuals from the parent population and perform fragment-level exchange on their gene segments to generate new offspring advertising creative individuals. Step 6: Introduce an environmental stress-driven mutation mechanism. During the generation of offspring individuals, dynamically adjust the mutation probability according to changes in external environmental factors, and randomly replace or perturb some gene segments. Step 7: Iterative evolution and population update. The offspring individuals and the surviving parent individuals together form a new generation of creative population, and steps 3 to 6 are repeated until the preset termination conditions are met.

2. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, In step 1, the process of constructing the advertising creative gene pool includes: For advertising copy gene fragments, the semantic analysis module is called to perform word segmentation and part-of-speech tagging on the copy, extract keywords, sentiment tendencies and sentence structure features, and map them into a fixed-dimensional numerical sequence. Language identifiers, word count statistics and preliminary screening status bits for sensitive word compliance are added to the fragments. For visual color matching gene fragments, the image material is sampled by a color space extraction algorithm. The color with the highest pixel proportion in the image is defined as the main color tone, and the color space coordinates of the main color tone are obtained. The distribution density of auxiliary colors in the image, the brightness difference between adjacent color blocks, and the hue circle span are calculated to form a multi-dimensional floating-point vector. For page layout gene fragments, a topological structure description method based on a grid system is adopted to divide the advertising display area into multiple equally divided logical grids. The starting row index, starting column index, number of rows occupied, number of columns occupied, and level depth of the title, image, button, and price tag in the grid coordinate system are defined to form a configuration list of page element arrangement and combination logic. For background music gene fragments, the audio signal is converted from the time domain to the frequency domain by fast Fourier transform, the average rhythm intensity is calculated and the basic frequency trend of the melody is identified, and the music is emotionally labeled by the audio classification module to generate a structured representation containing rhythm pulse sequence and frequency feature value.

3. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, In step 2, the specific process of initializing the creative population includes: The current computing power resource configuration is obtained in real time, including the available memory, graphics processor core load and network bandwidth margin, and the total number of individuals in the initial population is dynamically set according to the computing power resource configuration. Each individual advertising creative undergoes a business compliance constraint check. The verification module performs a logical consistency audit on the generated creatives. The audit content includes the matching degree between the product description in the copy and the image area size in the page layout, the consistency between the brand color scheme and the preset brand visual manual specifications, and the coverage relationship between the background music duration and the estimated advertising display cycle. If an individual fails any of the aforementioned business compliance constraints checks, the individual is marked and discarded, and a re-extraction logic is triggered until the number of compliant individuals generated meets the preset population size.

4. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, In step 3, the specific process of deploying the dynamic feedback evaluation mechanism includes: The system captures behavioral feedback from users through multi-channel integrated log stream technology. This behavioral feedback includes display exposure events, click interactions, page jump depth, and purchase conversion behavior. A sliding time window mechanism is established, which divides the preset observation period into multiple equal time slices and assigns corresponding attenuation weight coefficients according to the time distance between the time slice and the current time. The time slice that is closer to the current time has a higher attenuation weight coefficient. The click frequency and conversion frequency of each ad creative in each time slice are counted. The click frequency and conversion frequency are multiplied by the corresponding decay weight coefficient and then accumulated to calculate the real-time performance index of the ad creative. The fitness score of each creative is updated based on the real-time performance index.

5. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, In step 4, the specific process of performing natural selection includes: The short-term conversion efficiency and long-term user value are integrated to generate a comprehensive fitness score. The short-term conversion efficiency is composed of the weighted sum of the instant click-through rate and the instant conversion rate, and the long-term user value is composed of the weighted sum of the user retention probability predicted based on historical data and the repurchase intention score. The weighting coefficient between short-term conversion efficiency and long-term user value is dynamically configured according to the current marketing stage. During the brand promotion stage, the weight of click-through rate and dwell time is increased, and during the conversion harvesting stage, the weight of conversion rate and repurchase intention is increased. The sorting engine is invoked to rank the overall fitness scores of all individuals in the population from high to low. A selection threshold is set to filter the population. The selection threshold is a preset retention percentage or a dynamic boundary based on the mean fitness of the population. Individuals with scores higher than the selection threshold are retained, and the data structure of the eliminated individuals is transferred to the offline analysis module for failure mode mining.

6. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, In step 5, the specific process of implementing gene cross-recombination includes: In the parent population, a random sampling algorithm is used to select parental individuals to be paired up, and a mask vector containing multiple binary bits is generated. Each bit in the mask vector corresponds to a gene dimension, which includes text, color scheme, layout and background music. Iterate through each bit of the mask vector. When the mask bit is a preset first value, perform fragment-level swapping between parent individuals in the corresponding gene dimension. When the mask bit is the preset second value, the gene sequence of that gene dimension remains unchanged; during the process of performing fragment-level interchange, the cutting position in the gene sequence is determined by random number. For copywriting genes containing multiple phrases, the cutting position is set between word groups. By breaking the original gene combination logic, offspring individuals with new feature coupling points are generated.

7. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, In step 6, the specific process of introducing an environmental pressure-driven mutation mechanism includes: External environmental factor data is obtained through the real-time crawler interface of the environmental perception subsystem. The external environmental factor data includes the advertising density of competitors in media channels, the popularity value of hot topics on social platforms, seasonal consumption trend index, and macroeconomic operation fluctuation indicators. Semantic clustering is performed on the acquired text-based environmental data, and normalization is performed on the numerical environmental data. The processed data is then input into the mutation probability mapping function. When the fluctuation of external environmental factors increases, the mutation probability baseline value is automatically increased through the mutation probability mapping function, and the operation of randomly replacing gene fragment identifiers is performed at the binary stream level of offspring individuals, or random perturbation is performed on the parameters inside the gene fragments. The random perturbation includes changing the saturation value of the page background color and fine-tuning the rhythm and speed of the background music.

8. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, In step 7, the iterative evolution and population update process involves determining the termination condition, which includes one of the following situations: The current evolutionary generation has reached the preset maximum evolutionary generation threshold; The performance of the population tends to be stable, which is manifested in the fact that the increase in the average fitness score of the population is less than the preset minimum deviation over a continuous preset number of generations. If a major structural change in the market environment is detected, or if the environmental perception subsystem identifies an adjustment in industry policies or a sudden public event, the current iteration loop will be forcibly terminated and the population will be reinitialized.

9. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, The method also includes the archiving and application process of high-quality gene pools: The fitness score of each individual is monitored in real time. When the fitness score of an individual exceeds the preset proportion of the highest historical record, the gene fragments contained in that individual and the corresponding combination parameters are permanently stored in a high-performance read-only database to form a high-quality gene pool. In the subsequent new population initialization process or mutation operation, the evolution engine is configured to extract gene fragments from the high-quality gene pool with a preset priority probability. By introducing an experience memory mechanism, the performance optimization cycle after the new advertising material is launched is shortened and the evolution convergence speed is accelerated.

10. The method for optimizing advertising performance based on big data according to claim 1, characterized in that, The method also includes a secondary verification and anomaly monitoring process before deployment: Before the final advertising file is distributed to the distribution node, the compliance secondary verification module is called to link with the legal database in real time to check whether the advertising copy contains updated prohibited words and to verify whether the image content complies with the copyright policy of the target platform. Run an independent anomaly monitoring thread to continuously scan the characteristic distribution of individuals in the population. When a gene fragment is found to have a logical break, illegal characters, or garbled text due to mutation, immediately trigger the circuit breaker mechanism, revoke the deployment permission of the corresponding individual, and record the anomaly log. For dynamic page ads that include interactive logic, the layout gene fragments are expanded into interactive genes that include event triggering logic. Animation feedback is defined when the user clicks on different screen areas, and the interactive genes are made to participate in the evolutionary process of crossover, mutation and selection in sync.

Images