Marketing data analysis method and system based on AI large model

Abstract

The invention discloses a marketing data analysis method and system based on an AI large model, and belongs to the technical field of data processing. Marketing data analysis requirements customized by a worker are obtained, then the AI large model is optimized through data in the marketing data analysis requirements, and then the optimized AI large model is obtained; finally, the marketing data can be automatically analyzed through the optimized AI large model, the processing accuracy of the marketing data can be effectively improved, workers can be assisted in improving the processing efficiency of the marketing data, large-scale marketing data analysis can be effectively achieved, and the method has wide application prospects.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a marketing data analysis method and system based on an AI large model. Background Art

[0002] Marketing data refers to various data generated by enterprises during market promotion and sales processes, including consumer behavior, market trends, competition status, etc. Specifically, it involves multi-dimensional information such as sales performance, user feedback, advertisement click-through rate, and social media interaction. In the digital age, marketing data has become an important basis for enterprise decision-making. The marketing data contains rich market information and consumer behavior patterns, which is of great significance for enterprises to formulate marketing strategies and optimize products and services. However, traditional data analysis methods often struggle to handle such large-scale and complex data, resulting in the inability to fully exploit the data value. Summary of the Invention

[0003] The present invention provides a marketing data analysis method and system based on an AI large model to solve the problem that traditional data analysis methods often struggle to handle such large-scale and complex data, resulting in the inability to fully exploit the data value.

[0004] On the one hand, the present invention provides a marketing data analysis method based on an AI large model, including: Obtaining marketing data analysis requirements; wherein, the marketing data analysis requirements include historical marketing data samples specified by staff and marketing data analysis tags corresponding to the historical marketing data samples; Constructing an AI large model, and based on the marketing data analysis requirements, optimizing the model parameters of the AI large model using an AI model optimization algorithm to determine the optimized AI large model; Deploying the optimized AI large model in a cloud computing center, and collecting real-time marketing data through the cloud computing center, scheduling the deployed AI large model to analyze the real-time marketing data to determine the marketing data analysis result.

[0005] Further, constructing an AI large model includes: using a CNN large model or an RNN large model as the AI large model.

[0006] Further, based on the marketing data analysis requirements, optimizing the model parameters of the AI large model using an AI model optimization algorithm to determine the optimized AI large model, including: Initializing and encoding the model parameters of the AI large model using a solution space uniform distribution initialization strategy to determine multiple parameter encodings; Use the data in the marketing data analysis requirements as the optimization data, obtain the loss function value corresponding to each parameter encoding, and determine the parameter encoding with the smallest loss function value as the optimal parameter encoding; According to the optimal parameter encoding, use a large-range search strategy to quickly search the parameter encoding and obtain the parameter encoding after the quick search; Use a pivot reflection search strategy to perform a local search on the parameter encoding after the quick search and obtain the parameter encoding after the local search; Use an adaptive adjustment search strategy to perform an environmental impact search on the parameter encoding after the local search and obtain the parameter encoding after the environmental impact search; Use a global search strategy to perform a global search on the parameter encoding after the environmental impact search and obtain the parameter encoding after the global search; Repeat the execution of the large-range search strategy, pivot reflection search strategy, adaptive adjustment search strategy, and global search strategy until the optimization end condition is met. Re-determine the optimal parameter encoding according to the parameter encoding after the global search, and obtain the optimized AI large model according to the re-determined optimal parameter encoding.

[0007] Further, use a solution space uniform distribution initialization strategy to initialize and encode the model parameters of the AI large model, and determine multiple parameter encodings, including: For the model parameters of the AI large model, perform random initialization between the upper and lower limits of the model parameters, and encode the initialized model parameters into vectors to obtain the initial parameter encoding; Based on the initial parameter encoding, use a chaotic mapping initialization method to generate multiple different parameter encodings.

[0008] Further, according to the optimal parameter encoding, use a large-range search strategy to quickly search the parameter encoding and obtain the parameter encoding after the quick search, including: Use a Brownian motion function to generate a random influence factor, and generate a quick search quantity according to the random influence factor and the optimal parameter encoding; Based on the current optimization times, obtain an adaptive step size control factor, and use the adaptive step size control factor to control the quick search quantity to obtain the controlled quick search quantity; According to the optimal parameter encoding and the controlled quick search quantity, quickly search the parameter encoding to obtain the parameter encoding after the quick search.

[0009] Further, use a pivot reflection search strategy to perform a local search on the parameter encoding after the quick search and obtain the parameter encoding after the local search, including: Based on the current number of optimization times, combine with the sine function to generate the first reflection search factor, the second reflection search factor, and the third reflection search factor; Based on the current number of optimization times, generate the first fulcrum reflection search quantity, and after arranging the parameter encodings after the fast search in ascending order according to the loss function value, divide the parameter encodings into the first target encoding and the second target encoding according to the first fulcrum reflection search quantity; For the first target encoding, taking the optimal parameter encoding as the fulcrum, perform local search on the first target encoding according to the first reflection search factor, the second reflection search factor, and the third reflection search factor to obtain the first parameter encoding after local search; For the second target encoding, taking a random parameter encoding as the fulcrum, perform local search on the second target encoding according to the first reflection search factor, the second reflection search factor, and the third reflection search factor to obtain the second parameter encoding after local search.

[0010] Further, adopt an adaptive adjustment search strategy to perform environmental impact search on the parameter encodings after local search to obtain the parameter encodings after environmental impact search, including: According to the parameter encodings after local search, determine the worst parameter encoding, and determine the environmental impact amount according to the worst parameter encoding; Based on the current number of optimization times, obtain the adaptive environmental regulation factor, and use the adaptive environmental regulation factor to regulate the environmental impact amount to determine the regulated environmental impact amount; According to the regulated environmental impact amount and the historical parameter encoding state in the previous optimization process, perform environmental impact search on the parameter encodings after local search to obtain the parameter encodings after environmental impact search.

[0011] Further, adopt a global search strategy to perform global search on the parameter encodings after environmental impact search to obtain the parameter encodings after global search, including: For the parameter encodings after environmental impact search, generate the reverse solutions corresponding to the parameter encodings; Based on the reverse solutions corresponding to the parameter encodings and the worst parameter encoding, perform global search on the parameter encodings after environmental impact search to obtain the parameter encodings after global search.

[0012] Further, repeatedly execute the large - range search strategy, the fulcrum reflection search strategy, the adaptive adjustment search strategy, and the global search strategy until the optimization end condition is met. Re - determine the optimal parameter encoding according to the parameter encodings after global search, and obtain the optimized AI large model according to the re - determined optimal parameter encoding, including: Determine whether the current number of optimization times is less than or equal to the preset maximum number of optimization times. If so, repeatedly execute the large-scale search strategy, the pivot reflection search strategy, the adaptive adjustment search strategy, and the global search strategy. Otherwise, re-determine the optimal parameter encoding based on the parameter encoding after the global search; Use the model parameters included in the re-determined optimal parameter encoding as the final hyperparameters of the AI large model to obtain the optimized AI large model.

[0013] On the other hand, the present invention provides a marketing data analysis system based on an AI large model, including: a historical data acquisition module, a historical data learning module, and a marketing data analysis module; The historical data acquisition module is used to obtain marketing data analysis requirements; wherein, the marketing data analysis requirements include historical marketing data samples specified by staff and marketing data analysis labels corresponding to the historical marketing data samples; The historical data learning module is used to construct an AI large model, and based on the marketing data analysis requirements, use an AI model optimization algorithm to optimize the model parameters of the AI large model to determine the optimized AI large model; The marketing data analysis module is used to deploy the optimized AI large model to a cloud computing center, collect real-time marketing data through the cloud computing center, and schedule the deployed AI large model to analyze the real-time marketing data to determine the marketing data analysis result.

[0014] A marketing data analysis method and system based on an AI large model provided by the present invention, by obtaining marketing data analysis requirements customized by staff, then optimizing the AI large model through the data in the marketing data analysis requirements to obtain the optimized AI large model, and finally can automatically analyze marketing data through the optimized AI large model, which can not only effectively improve the processing accuracy of marketing data, but also assist staff in improving the processing efficiency of marketing data, and can effectively realize large-scale marketing data analysis, with broad application prospects. Brief Description of the Drawings

[0015] The drawings here are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0016] Figure 1 It is a flowchart of a marketing data analysis method based on an AI large model provided by an embodiment of the present invention.

[0017] Figure 2 It is a schematic structural diagram of a marketing data analysis system based on an AI large model provided by an embodiment of the present invention.

[0018] Among them, there are 201 - historical data acquisition module, 202 - historical data learning module, and 203 - marketing data analysis module.

[0019] Through the above - mentioned accompanying drawings, specific embodiments of the present invention have been shown, and there will be a more detailed description hereinafter. These drawings and text descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. Detailed implementation manners

[0020] Here, exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0021] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] As Figure 1 shown, an embodiment of the present invention provides a marketing data analysis method based on an AI large - model, including: S101. Obtain marketing data analysis requirements; among them, the marketing data analysis requirements include historical marketing data samples specified by staff and marketing data analysis tags corresponding to the historical marketing data samples; The data type of the historical marketing data samples can be specified by the staff themselves, so as to realize customized marketing data analysis tasks. For example, the historical marketing data samples can be set as image sample data. Suppose that during the process of marketing to users, it is necessary to collect the psychological types of users, and in the prior art, action types are usually used to analyze psychological types. Then, it is necessary to classify the image sample data first. Therefore, the marketing data analysis tag can be the action type of the user. After learning the data relationship between the image sample data and the action type, automatic analysis can be realized during the marketing process, achieving refined marketing. However, it should be noted that the above - mentioned marketing data analysis requirements are only an example, and the marketing data analysis requirements can also be set as other types of data, subject to actual needs, to assist the staff in improving the efficiency of marketing data analysis.

[0023] S102. Construct an AI (Artificial Intelligence) large - model, and based on the marketing data analysis requirements, use an AI model optimization algorithm to optimize the model parameters of the AI large - model, and determine the optimized AI large - model; The AI large model can be an artificial intelligence large model capable of analyzing historical marketing data samples. Generally, marketing data may be divided into image data or text data. Therefore, in the embodiments of the present invention, a convolutional neural network large model can be preferably used as the AI large model to realize the analysis of marketing data. However, it should be noted that before the data is input, it should be converted into the data format required by the AI large model.

[0024] S103. Deploy the optimized AI large model in the cloud computing center, collect real-time marketing data through the cloud computing center, and schedule the deployed AI large model to analyze the real-time marketing data to determine the marketing data analysis result.

[0025] A marketing data analysis method based on an AI large model provided by the present invention obtains the marketing data analysis requirements customized by the staff, then optimizes the AI large model through the data in the marketing data analysis requirements to obtain the optimized AI large model. Finally, the optimized AI large model can be used to automatically analyze the marketing data, which can not only effectively improve the processing accuracy of the marketing data, but also assist the staff in improving the processing efficiency of the marketing data, and can effectively realize large-scale marketing data analysis, having broad application prospects.

[0026] In the embodiments of the present invention, building an AI large model includes: using a CNN (convolutional neural network) large model or an RNN (recurrent neural network) large model as the AI large model. It should be noted that the above AI large model is only an example in the embodiments of the present invention, and other AI large models can also be used for data analysis.

[0027] In the embodiments of the present invention, based on the marketing data analysis requirements, using an AI model optimization algorithm to optimize the model parameters of the AI large model to determine the optimized AI large model includes: Initializing and encoding the model parameters of the AI large model using a solution space uniform distribution initialization strategy to determine multiple parameter encodings; Using the data in the marketing data analysis requirements as optimization data, obtaining the loss function value (such as the root mean square loss function value) corresponding to each parameter encoding, and determining the parameter encoding with the smallest loss function value as the optimal parameter encoding; According to the optimal parameter encoding, using a large range search strategy to quickly search the parameter encodings to obtain the parameter encodings after the quick search; Using a pivot reflection search strategy to perform local search on the parameter encodings after the quick search to obtain the parameter encodings after the local search; An adaptive adjustment search strategy is used to perform an environmental impact search on the parameter encoding after local search to obtain the parameter encoding after the environmental impact search; A global search strategy is used to perform a global search on the parameter encoding after the environmental impact search to obtain the parameter encoding after the global search; The large-range search strategy, the pivot reflection search strategy, the adaptive adjustment search strategy, and the global search strategy are repeatedly executed until the optimization end condition is met. The optimal parameter encoding is re-determined according to the parameter encoding after the global search, and the optimized AI large model is obtained according to the re-determined optimal parameter encoding.

[0028] Optionally, after each strategy is executed, out-of-limit processing can also be performed on the parameter encoding to ensure that the parameter encoding is always between the preset upper and lower limits.

[0029] In the process of optimizing the model parameters of the AI large model in the prior art, it is easy to fall into the local optimum in the solution space, resulting in the inability of the optimized AI large model to effectively analyze marketing data. Therefore, the embodiment of the present invention provides an AI model optimization algorithm to optimize the model parameters of the AI large model to improve the analysis accuracy of marketing data and assist staff in improving the analysis efficiency of marketing data.

[0030] In the embodiment of the present invention, the model parameters of the AI large model are initialized and encoded by using a solution space uniform distribution initialization strategy to determine multiple parameter encodings, including: For the model parameters (such as connection weights) of the AI large model, random initialization is performed between the upper and lower limits of the model parameters, and the initialized model parameters are encoded as vectors to obtain the initial parameter encoding; Based on the initial parameter encoding, a chaotic mapping initialization method is used to generate multiple different parameter encodings.

[0031] The solution space uniform distribution initialization strategy provided by the embodiment of the present invention can effectively improve the distribution uniformity of the initial solution in the solution space, increase the possibility of finding the global optimal solution, and improve the algorithm optimization speed.

[0032] In the embodiment of the present invention, according to the optimal parameter encoding, a large-range search strategy is used to quickly search the parameter encoding to obtain the parameter encoding after the quick search, including: A Brownian motion function is used to generate a random influence factor, and according to the random influence factor and the optimal parameter encoding, a quick search quantity is generated as:

[0033] where represents the quick search quantity corresponding to the mth parameter encoding, It represents the encoding of the m-th parameter during the t-th optimization process. It represents the encoding of the optimal parameter. It represents the random influence factor generated by the Brownian motion function; m = 1, 2, …, M, where M represents the total number of parameter encodings. Based on the current optimization times, the adaptive step size control factor is obtained as follows:

[0034] Among them, It represents the adaptive step size control factor. It represents the preset maximum number of optimization times. The adaptive step size control factor is used to control the rapid search quantity, and the rapid search quantity after control is obtained as follows: ; Among them, It represents the information learning factor, which is generally set as a constant (such as 0.02, 0.05, etc., which can be defined by oneself). Based on the optimal parameter encoding and the rapid search quantity after control, a rapid search is performed on the parameter encoding, and the parameter encoding after rapid search is obtained as follows:

[0035] Among them, It represents the encoding of the m-th parameter after rapid search.

[0036] The large-range search strategy provided by the embodiments of the present invention can perform a large-range search on the entire area based on the optimal parameter encoding, and at the same time introduce the Brownian motion function to make the search process fluctuate, thereby improving the global search ability. At the later stage of the algorithm, all parameter encodings search around the optimal position, improving the search accuracy.

[0037] In the embodiments of the present invention, a fulcrum reflection search strategy is used to perform a local search on the parameter encoding after rapid search to obtain the parameter encoding after local search, including: Based on the current optimization times, the first reflection search factor, the second reflection search factor, and the third reflection search factor are generated in combination with the sine function as follows:

[0038]

[0039]

[0040] Among them, It represents the first reflection search factor during the t-th optimization process. It represents the second reflection search factor during the t-th optimization process. represents the third reflection search factor in the t-th optimization process, represents the third reflection search factor in the (t - 1)-th optimization process, represents the pi, represents the coefficient of the Sine mapping (which can be set to 4). The initial values of the first reflection search factor, the second reflection search factor, and the third reflection search factor are all randomly generated within (0, 1); Based on the current optimization times, the number of first pivot point reflection searches generated is:

[0041] where, represents the number of first pivot point reflection searches, represents the rounding function, represents the preset maximum number of optimizations, represents the preset maximum number of first pivot point reflection searches (which can be set to M / 10), represents the preset minimum number of first pivot point reflection searches (which can be set to / 2); After arranging the parameter encodings after the fast search in ascending order of the loss function values, according to the number of first pivot point reflection searches, the parameter encodings are divided into the first target encoding and the second target encoding; For example, based on the number of first pivot point reflection searches, N parameter encodings after sorting can be taken as the first target encoding, and the remaining parameter encodings can be taken as the second target encoding; For the first target encoding, with the optimal parameter encoding as the pivot point, local search is performed on the first target encoding according to the first reflection search factor, the second reflection search factor, and the third reflection search factor, and the first parameter encoding after local search is obtained as:

[0042]

[0043] where, represents the i -th first target encoding in the t-th optimization process, i = 1, 2, …, N, represents the i -th first parameter encoding after local search, represents the optimal parameter encoding, represents the reflection search factor, represents the maximum value of the reflection search factor (which can be set to 2), represents the minimum value of the reflection search factor (which can be set to 1); For the second target encoding, with the random parameter encoding as the fulcrum, local search is performed on the second target encoding according to the first reflection search factor, the second reflection search factor, and the third reflection search factor, and the second parameter encoding after local search is obtained as:

[0044] Wherein, represents the j -th second target encoding in the t-th optimization process, j = 1, 2, …, M - N, represents the j -th second parameter encoding after local search, represents the random parameter encoding, which is selected from the parameter encodings after fast search.

[0045] For the fulcrum reflection search strategy provided by the embodiments of the present invention, in the early stage of algorithm optimization, the number of the first fulcrum reflection searches is small, ensuring that most of the parameter encodings among the worst N parameter encodings are reflected around with other random parameter encodings as the fulcrum, enhancing the global exploration ability and ensuring the diversity of the population. As the number of optimizations increases, the value of the number of the first fulcrum reflection searches gradually increases, and more and more parameter encodings among the N parameter encodings with the largest loss function value are reflected around with the currently found optimal solution as the fulcrum, so as to enhance the local mining ability of these parameter encodings around the optimal solution. Because the parameter encodings are reflected and explored around with the optimal solution as the fulcrum, it can help the population still have the opportunity and ability to jump out of the local optimum in the later stage of iteration.

[0046] In the embodiments of the present invention, an adaptive adjustment search strategy is adopted to perform environmental impact search on the parameter encoding after local search, and the parameter encoding after environmental impact search is obtained, including: Determine the worst parameter encoding according to the parameter encoding after local search, and determine the environmental impact amount according to the worst parameter encoding as:

[0047] Wherein, represents the t -th parameter encoding after local search in the k -th optimization process, represents the worst parameter encoding, represents the environmental impact amount corresponding to the k -th parameter encoding after local search, k = 1, 2, …, M; | | represents the symbol for taking the absolute value of the elements in the parameter encoding; Based on the current number of optimizations, obtain the adaptive environmental regulation factor as:

[0048] Among them, represents the adaptive environment regulation factor, represents the maximum value of the adaptive environment regulation factor (which can be set to 0.2), represents the minimum value of the adaptive environment regulation factor (which can be set to 0); The adaptive environment regulation factor is used to regulate the environmental impact amount, and the environmental impact amount after regulation is determined as: ; among them, represents the first random number between (0, 1); According to the environmental impact amount after the regulation and the historical parameter coding state in the previous optimization process, the parameter coding after local search is subjected to environmental impact search, and the parameter coding after environmental impact search is obtained as:

[0049] Among them, represents the t th parameter coding after the k -1st local search in the optimization process (i.e., the historical parameter coding state), represents the parameter coding after the kth environmental impact search, represents the second random number that is randomly -1 or 1, represents the historical state memory parameter between (0, 0.2).

[0050] The adaptive adjustment search strategy provided by the embodiments of the present invention can adaptively adjust the search speed according to the overall state of the current search, and at the same time memorize the historical parameter coding state, which can not only effectively improve the local search ability, but also increase the ability to jump out of the local optimum and improve the parameter optimization effect.

[0051] In the embodiments of the present invention, a global search strategy is adopted to perform a global search on the parameter coding after environmental impact search, and the parameter coding after global search is obtained, including: For the parameter coding after environmental impact search, a reverse solution corresponding to the parameter coding is generated; Based on the reverse solution corresponding to the parameter coding and the worst parameter coding, a global search is performed on the parameter coding after environmental impact search, and the parameter coding after global search is obtained as:

[0052] Among them, represents the t th reverse solution corresponding to the n th parameter coding in the optimization process, represents then Parameter encoding after a global search Represents the first differential evolution factor (which can be set as a constant term between (0, 0.1) or (0, 0.2)), Represents the second differential evolution factor (which can be set as a constant term between (0, 0.1) or (0, 0.2)), Represents the third random number between (0, 1), Represents the fourth random number between (0, 1), Represents the newly generated parameter encoding randomly.

[0053] The global search strategy provided by the embodiments of the present invention can adopt a fusion strategy of reverse search, differential evolution, and exclusive search, which can effectively improve the global search ability of the algorithm and assist the algorithm to jump out of the local optimal solution.

[0054] Optionally, the greedy principle or the simulated annealing algorithm can also be used to control the global search strategy to ensure the optimization speed of the algorithm.

[0055] Through the mutual cooperation among the large-scale search strategy, the fulcrum reflection search strategy, the adaptive adjustment search strategy, and the global search strategy in the embodiments of the present invention, the global search ability and search accuracy of the algorithm are effectively improved, the optimization effect of the algorithm is assisted to be improved, the data recognition performance of the AI large model is improved, and finally the analysis accuracy of marketing data is improved.

[0056] In the embodiments of the present invention, the large-scale search strategy, the fulcrum reflection search strategy, the adaptive adjustment search strategy, and the global search strategy are repeatedly executed until the optimization end condition is met. The optimal parameter encoding is re-determined according to the parameter encoding after the global search, and the optimized AI large model is obtained according to the re-determined optimal parameter encoding, including: Judge whether the current number of optimizations is less than or equal to the preset maximum number of optimizations. If so, repeatedly execute the large-scale search strategy, the fulcrum reflection search strategy, the adaptive adjustment search strategy, and the global search strategy. Otherwise, re-determine the optimal parameter encoding according to the parameter encoding after the global search; Use the model parameters included in the re-determined optimal parameter encoding as the final hyperparameters of the AI large model to obtain the optimized AI large model.

[0057] As Figure 2 shown, the present invention provides a marketing data analysis system based on an AI large model, including: a historical data acquisition module 201, a historical data learning module 202, and a marketing data analysis module 203; The historical data acquisition module 201 is configured to obtain marketing data analysis requirements; wherein, the marketing data analysis requirements include historical marketing data samples specified by staff and marketing data analysis tags corresponding to the historical marketing data samples. The historical data learning module 202 is configured to build an AI large model, and based on the marketing data analysis requirements, optimize the model parameters of the AI large model by using an AI model optimization algorithm to determine the optimized AI large model. The marketing data analysis module 203 is configured to deploy the optimized AI large model to a cloud computing center, collect real-time marketing data through the cloud computing center, and schedule the deployed AI large model to analyze the real-time marketing data to determine the marketing data analysis result.

[0058] A marketing data analysis system based on an AI large model provided by an embodiment of the present invention can execute the above marketing data analysis method, and its principle and beneficial effects are similar, which will not be elaborated here.

[0059] Those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not disclosed by the present invention. It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A marketing data analysis method based on AI big model, characterized in that: include: Obtaining marketing data analysis requirements; wherein the marketing data analysis requirements include historical marketing data samples specified by the staff and marketing data analysis tags corresponding to the historical marketing data samples; Construct an AI big model, and based on the marketing data analysis requirements, use an AI model optimization algorithm to optimize the model parameters of the AI ​​big model, and determine the optimized AI big model; The optimized AI big model is deployed in the cloud computing center, and real-time marketing data is collected through the cloud computing center. The deployed AI big model is scheduled to analyze the real-time marketing data to determine the marketing data analysis results.

2. The marketing data analysis method based on AI big model according to claim 1 is characterized in that: Constructing an AI big model, including: using a CNN big model or an RNN big model as the AI ​​big model.

3. The marketing data analysis method based on AI big model according to claim 1 is characterized in that: Based on the marketing data analysis requirements, the model parameters of the AI ​​big model are optimized using an AI model optimization algorithm to determine the optimized AI big model, including: Initialize and encode the model parameters of the AI ​​large model using a solution space uniform distribution initialization strategy to determine multiple parameter encodings; Using the data in the marketing data analysis requirements as optimization data, obtaining the loss function value corresponding to each parameter encoding, and determining the parameter encoding with the smallest loss function value as the optimal parameter encoding; According to the optimal parameter code, a large-scale search strategy is used to quickly search the parameter code to obtain the parameter code after the quick search; A pivot reflection search strategy is used to perform a local search on the parameter coding after the fast search to obtain the parameter coding after the local search; Adopting an adaptive adjustment search strategy to perform an environmental impact search on the parameter coding after the local search, and obtaining the parameter coding after the environmental impact search; A global search strategy is used to perform a global search on the parameter coding after the environmental impact search, and the parameter coding after the global search is obtained; Repeat the large-range search strategy, fulcrum reflection search strategy, adaptive adjustment search strategy and global search strategy until the optimization end conditions are met, redetermine the optimal parameter encoding based on the parameter encoding after the global search, and obtain the optimized AI large model based on the redetermined optimal parameter encoding.

4. The marketing data analysis method based on AI big model according to claim 3 is characterized in that: The model parameters of the AI ​​large model are initialized and encoded using a solution space uniform distribution initialization strategy to determine multiple parameter encodings, including: For the model parameters of the AI ​​large model, randomly initialize them between the upper limit and the lower limit of the model parameters, and encode the initialized model parameters into a vector to obtain an initial parameter code; Based on the initial parameter code, a chaotic mapping initialization method is used to generate a plurality of different parameter codes.

5. The marketing data analysis method based on AI big model according to claim 3 is characterized in that: According to the optimal parameter code, a large-range search strategy is used to quickly search the parameter code to obtain the parameter code after the quick search, including: A Brownian motion function is used to generate a random impact factor, and a fast search volume is generated according to the random impact factor and the optimal parameter encoding; Based on the current optimization times, an adaptive step size control factor is obtained, and the fast search amount is controlled by using the adaptive step size control factor to obtain the fast search amount after the control; According to the optimal parameter code and the fast search amount after the control, the parameter code is quickly searched to obtain the parameter code after the fast search.

6. The marketing data analysis method based on AI big model according to claim 5 is characterized in that: The pivot reflection search strategy is used to perform local search on the parameter encoding after the quick search to obtain the parameter encoding after the local search, including: Based on the current optimization times, a first reflection search factor, a second reflection search factor and a third reflection search factor are generated in combination with a sine function; Based on the current optimization times, the number of first-point reflection searches is generated, and the parameter codes after the quick search are arranged in ascending order of the loss function value, and then the parameter codes are divided into first target codes and second target codes according to the number of first-point reflection searches; For the first target code, taking the optimal parameter code as a fulcrum, performing a local search on the first target code according to the first reflection search factor, the second reflection search factor and the third reflection search factor, and obtaining a first parameter code after the local search; For the second target code, with the random parameter code as a fulcrum, a local search is performed on the second target code according to the first reflection search factor, the second reflection search factor and the third reflection search factor to obtain the second parameter code after the local search.

7. The marketing data analysis method based on AI big model according to claim 6 is characterized in that: Adopting an adaptive adjustment search strategy to perform environmental impact search on the parameter coding after the local search, and obtaining the parameter coding after the environmental impact search, including: Determining the worst parameter coding according to the parameter coding after the local search, and determining the environmental impact amount according to the worst parameter coding; Based on the current number of optimizations, an adaptive environmental control factor is obtained, and the environmental impact amount is controlled by the adaptive environmental control factor to determine the environmental impact amount after control; According to the environmental impact amount after the regulation and the historical parameter coding state in the last optimization process, an environmental impact search is performed on the parameter coding after the local search to obtain the parameter coding after the environmental impact search.

8. The marketing data analysis method based on AI big model according to claim 7 is characterized in that: A global search strategy is used to perform a global search on the parameter coding after the environmental impact search to obtain the parameter coding after the global search, including: For the parameter coding after the environmental impact search, generate the reverse solution corresponding to the parameter coding; Based on the reverse solution corresponding to the parameter code and the worst parameter code, a global search is performed on the parameter code after the environmental impact search to obtain the parameter code after the global search.

9. The marketing data analysis method based on AI big model according to claim 3 is characterized in that: Repeat the large-scale search strategy, fulcrum reflection search strategy, adaptive adjustment search strategy and global search strategy until the optimization end condition is met, redetermine the optimal parameter encoding according to the parameter encoding after the global search, and obtain the optimized AI large model according to the redetermined optimal parameter encoding, including: Determine whether the current number of optimizations is less than or equal to the preset maximum number of optimizations. If so, repeat the large-scale search strategy, the pivot reflection search strategy, the adaptive adjustment search strategy, and the global search strategy. Otherwise, re-determine the optimal parameter encoding based on the parameter encoding after the global search. The model parameters contained in the re-determined optimal parameter encoding are used as the final hyperparameters of the AI ​​big model to obtain the optimized AI big model.

10. A marketing data analysis system based on AI big model, characterized in that: include: Historical data acquisition module, historical data learning module and marketing data analysis module; The historical data acquisition module is used to acquire marketing data analysis requirements; wherein the marketing data analysis requirements include historical marketing data samples specified by the staff and marketing data analysis tags corresponding to the historical marketing data samples; The historical data learning module is used to build an AI big model, and based on the marketing data analysis requirements, use an AI model optimization algorithm to optimize the model parameters of the AI ​​big model to determine the optimized AI big model; The marketing data analysis module is used to deploy the optimized AI big model in the cloud computing center, collect real-time marketing data through the cloud computing center, schedule the deployed AI big model to analyze the real-time marketing data, and determine the marketing data analysis results.

Images