Marketing intelligent decision-making system based on multi-modal learning
Through multimodal learning technology, combined with real-time market trend monitoring and in-depth psychological analysis, personalized incentives are designed to solve the problems of low decision efficiency and lack of personalization in the existing marketing intelligent decision-making system, and quickly respond to market changes and The goal of improving marketing effectiveness.
Patent Information
- Application Number
- CN202411369044.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-29
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing marketing intelligent decision-making system lacks real-time market trend monitoring and prediction, and it is difficult to respond quickly to market changes. The incentives lack personalization, resulting in low decision-making efficiency and poor marketing effectiveness.
Adopt a marketing intelligent decision-making system based on multimodal learning, including multimodal data acquisition, multimodal feature fusion, generative adversarial network, real-time market trend monitoring, in-depth psychological analysis, behavior prediction incentives and dynamic decision-making modules, through real-time monitoring Market trends and user behaviors, design personalized incentives, and quickly adjust marketing strategies.
It improves the real-time and accuracy of decision-making, enhances the timeliness of market trend forecasts and the accuracy of user behavior forecasts, ensures the flexibility and targetedness of marketing strategies, and maximizes marketing effectiveness.
Smart Images

Figure CN118917881A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of multimodal learning, and specifically refers to a marketing intelligent decision-making system based on multimodal learning. Background Art
[0002] With the continuous development of big data and artificial intelligence technology, the marketing field has gradually entered the era of intelligent decision-making. Traditional marketing decisions mainly rely on the experience and data analysis of market researchers, and there are problems such as low decision-making efficiency and lack of accuracy.
[0003] However, the existing marketing intelligent decision-making system still has certain defects. The existing marketing intelligent decision-making system lacks real-time market trend monitoring and prediction, and relies on relatively lagging data, resulting in poor timeliness of prediction results and difficulty in responding to market changes in a timely manner. When designing incentives, a more general method is adopted, ignoring the individual differences and demand diversity among different consumers, resulting in a significant reduction in the effectiveness of incentives. The lack of close integration with market trends and user behavior predictions makes the incentives out of touch with the actual market situation and difficult to achieve the expected marketing effect. The lack of sufficient flexibility and response speed makes it difficult to adjust marketing strategies in time when facing rapid market changes and sudden changes in user behavior, resulting in missed market opportunities. The decision-making process is too cumbersome or relies on manual judgment, making it difficult for the system to make fast and effective decisions when facing a complex and changing market environment. For this reason, a marketing intelligent decision-making system based on multimodal learning is proposed. Summary of the invention
[0004] The purpose of the present invention is to provide a marketing intelligent decision-making system based on multimodal learning to solve the problems raised in the above background technology.
[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a marketing intelligent decision-making system based on multimodal learning, including a multimodal data acquisition module, a multimodal feature fusion module, a generative adversarial network, a real-time market trend monitoring module, a deep psychological analysis module, a behavior prediction incentive module, a dynamic decision-making module, and a feedback optimization module;
[0006] The multimodal feature fusion module unifies the format of the data collected by the multimodal data collection module, extracts multimodal features, and fuses them to form multimodal fusion features;
[0007] The generative adversarial network is used to generate new data samples to simulate the behavior of real users and market scenarios;
[0008] Wherein, the real-time market trend monitoring module is used to monitor and predict market trends in real time based on real-time data and historical data;
[0009] Wherein, the deep psychological analysis module is used to analyze the user's psychological state;
[0010] The behavior prediction incentive module predicts user behavior and designs corresponding incentive measures based on market trend monitoring results and deep psychological analysis results;
[0011] The dynamic decision-making module formulates marketing strategies based on the real-time market trend information of the market trend monitoring module and the user behavior prediction results and incentive measures of the behavior prediction incentive module.
[0012] The multimodal feature fusion module extracts and fuses features based on the data collected by the multimodal data acquisition module. Assume that the collected data modalities include images, text and audio, and extract features from the data modules. The image feature extraction is , text feature extraction is , image feature extraction is ,in, Represent the feature vectors of images, texts and audios respectively, compare the feature vectors of multimodal data, minimize the difference between different modal features, and implement the formula as follows:
[0013] ,
[0014] In the formula, represents the alignment loss function, which minimizes the difference between different modal features, and g represents the alignment function, which measures the similarity between two features. The processed multimodal features are fused. Represents the calculation of alignment loss function The length of the vector output by each alignment function g in , after multimodal feature fusion, is expressed as , F represents the fused multimodal features, represents the weight coefficient, represents a non-linear activation function.
[0015] The generative adversarial network includes a generator G and a discriminator D. The generator receives a random noise z and a fused multimodal feature F, and generates a new sample x=G(z, F). The discriminator receives the input sample x and the fused multimodal feature F at the same time, which is expressed as D(x, F).
[0016] Generator, the loss function maximizes the discriminator's authenticity score for the generated samples, and the implementation formula is:
[0017] ,
[0018] In the formula, E represents the expected value, Represents the distribution of random noise z And the expected value of the fusion feature F, Represents the score of the discriminator D on the generated sample G(z,F) and the fused multimodal feature F;
[0019] The discriminator loss function maximizes the ability to distinguish between real data and generated data. The implementation formula is:
[0020] ,
[0021] In the formula, represents the real data distribution, E represents the expected value, Represents the real data distribution And the expected value of the fusion feature F, represents the logarithm of the score of the discriminator D on the real data x and the fusion feature F, Represents the distribution of random noise z And the expected value of the fusion feature F, Represents the fake data generated by the generator to the discriminator D The logarithm of the rating.
[0022] The real-time market trend monitoring module selects the feature X related to the market trend from the multimodal features fused by the multimodal feature fusion module, and the implementation formula is:
[0023] ,
[0024] In the formula, Indicates market characteristics The information entropy of Indicates market characteristics In the value , i represents the i-th market characteristic, j represents the market characteristic The jth possible value of Indicates market characteristics The number of all possible values of Select the first k features with the highest information entropy as important features;
[0025] The multimodal feature fusion module effectively reduces the deviation and noise brought by single-modal data by extracting and fusing features of different modalities, and improves the accuracy and robustness of feature representation. At the same time, the introduction of generative adversarial networks further enhances the diversity and authenticity of the data, making market trend predictions and user behavior predictions more accurate.
[0026] According to the obtained market characteristics, the current market trend is predicted, and the implementation formula is:
[0027] ,
[0028] In the formula, represents the predicted value at the current moment, c represents the constant term, and They represent the autoregressive and moving average coefficients respectively, p represents the order of the autoregressive term, q represents the order of the moving average term, i represents the lag order of the autoregressive part, from 1 to p, j represents the lag order of the moving average part, from 1 to q, Represents the white noise error term, and the predicted value at the current moment Put it into F to get a new feature set , according to the new set characteristics, the final forecast value of the market trend is obtained, and the implementation formula is:
[0029] ,
[0030] In the formula, Indicates the final forecast value of the market trend, represents a feature set, represents the weight of the mth market trend forecasting model, represents the predicted value of the mth market trend forecasting model, and M represents the number of market trend forecasting models.
[0031] Among them, the real-time market trend monitoring module updates the market trend prediction model according to the real-time data stream received by the multimodal feature fusion module, and the implementation formula is:
[0032] ,
[0033] In the formula, It represents the predicted value at the next time t+1. The final prediction model of the updated market trend is:
[0034] ,
[0035] The weight of the final prediction model of market trends is updated according to real-time data. The implementation formula is:
[0036] ,
[0037] In the formula, represents the weight of the m+1th prediction model, η represents the learning rate, represents the actual observed value, Represents the final forecast value of the market trend forecast model.
[0038] The real-time market trend monitoring module can monitor market changes in real time and make predictions based on historical data, ensuring the timeliness and effectiveness of decision-making.
[0039] The deep psychological analysis module extracts psychological state features from the multimodal feature fusion module. , each psychological state feature is assigned a weight , according to each psychological state feature and the corresponding weight, the comprehensive score of the psychological state is obtained, and the implementation formula is:
[0040] ,
[0041] In the formula, represents the comprehensive score of psychological state, k represents the number of psychological state characteristics, It represents the jth feature score on the i-th feature. The user's psychological state is analyzed according to the comprehensive score of the psychological state. The implementation formula is:
[0042] ,
[0043] In the formula, e represents the psychological state, and E represents the set of all possible psychological states. Indicates the psychological state e in the comprehensive score of psychological state The next score.
[0044] The behavior prediction incentive module predicts user behavior and designs corresponding incentive measures based on the market trend prediction results and psychological state analysis results. The comprehensive prediction implementation formula is:
[0045] ,
[0046] In the formula, Indicates that given the current market trend 、Future market trends Under the conditions of user psychological state e, the score of user behavior b is calculated, and corresponding measures are designed according to the comprehensive prediction score.
[0047] The dynamic decision-making module formulates marketing strategies based on the real-time market trend information of the market trend monitoring module and the user behavior prediction results and incentive measures of the behavior prediction incentive module. The optimal marketing strategy S is implemented as follows:
[0048] ,
[0049] In the formula, S represents the optimal marketing strategy, I(b) represents the incentives, and d represents the decision function, which is used to compare different marketing strategies. Indicates current market trends, It represents the future market trend, e represents the user's psychological state, and b represents the user's behavior.
[0050] The behavior prediction and incentive module comprehensively predicts user behavior based on market trends and psychological analysis results, and designs corresponding incentives to actively guide user behavior and maximize marketing effects.
[0051] The feedback optimization module collects the effect feedback and market data of marketing strategies in real time, and feeds the collected information back to the system for optimization.
[0052] Compared with the prior art, the present invention has the following beneficial effects:
[0053] 1. In order to solve the problem of inaccurate marketing decisions and weak real-time response capabilities, and to improve real-time performance, accuracy and effectiveness, the present invention uses a behavior prediction incentive module to comprehensively predict user behavior based on market trends and psychological analysis results, and designs corresponding incentive measures to actively guide user behavior and maximize marketing effects. The dynamic decision-making module can quickly formulate and adjust marketing strategies based on real-time market trends and user behavior prediction results;
[0054] 2. The present invention uses a real-time market trend monitoring module to monitor and predict market trends in real time based on historical and real-time data, which can quickly and accurately capture market changes and ensure the timeliness and accuracy of the prediction results;
[0055] 3. The present invention uses a behavior prediction incentive module, combined with market trend prediction and deep psychological analysis results, to accurately predict consumer behavior and formulate corresponding incentive measures. Through accurate prediction and personalized incentive measures, the effectiveness of marketing strategies is improved;
[0056] 4. The present invention uses a dynamic decision-making module to flexibly adjust marketing strategies according to market trends, real-time feedback of user behavior and incentives to ensure the pertinence and effectiveness of decisions. Through real-time feedback and optimization, the marketing strategy is continuously improved to enhance marketing effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 It is a structural diagram of the marketing intelligent decision-making system based on multimodal learning of the present invention;
[0058] Figure 2 It is an operation flow chart of the real-time market trend monitoring module of the marketing intelligent decision-making system based on multimodal learning of the present invention;
[0059] Figure 3 This is an operation flow chart of the behavior prediction incentive module of the marketing intelligent decision-making system based on multimodal learning of the present invention;
[0060] Figure 4 This is an operation flow chart of the dynamic decision module of the marketing intelligent decision system based on multimodal learning of the present invention. DETAILED DESCRIPTION
[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0062] Example
[0063] See also Figure 1-Figure 4 As shown, the present invention provides a technical solution: including a multimodal data acquisition module, a multimodal feature fusion module, a generative adversarial network, a real-time market trend monitoring module, a deep psychological analysis module, a behavior prediction incentive module, a dynamic decision module, and a feedback optimization module;
[0064] The multimodal feature fusion module unifies the format of the data collected by the multimodal data collection module, extracts multimodal features, and fuses them to form multimodal fusion features;
[0065] The generative adversarial network is used to generate new data samples to simulate the behavior of real users and market scenarios;
[0066] Wherein, the real-time market trend monitoring module is used to monitor and predict market trends in real time based on real-time data and historical data;
[0067] Wherein, the deep psychological analysis module is used to analyze the user's psychological state;
[0068] The behavior prediction incentive module predicts user behavior and designs corresponding incentive measures based on market trend monitoring results and deep psychological analysis results;
[0069] The dynamic decision-making module formulates marketing strategies based on the real-time market trend information of the market trend monitoring module and the user behavior prediction results and incentive measures of the behavior prediction incentive module.
[0070] The multimodal feature fusion module extracts and fuses features based on the data collected by the multimodal data acquisition module. Assume that the collected data modalities include images, text and audio, and extract features from the data modules. The image feature extraction is , text feature extraction is , image feature extraction is ,in, Represent the feature vectors of images, texts and audios respectively, compare the feature vectors of multimodal data, minimize the difference between different modal features, and implement the formula as follows:
[0071] ,
[0072] In the formula, represents the alignment loss function, which minimizes the difference between different modal features, and g represents the alignment function, which measures the similarity between two features. The processed multimodal features are fused. Represents the calculation of alignment loss function The length of the vector output by each alignment function g in , after multimodal feature fusion, is expressed as , F represents the fused multimodal features, represents the weight coefficient, represents a non-linear activation function.
[0073] The generative adversarial network includes a generator G and a discriminator D. The generator receives a random noise z and a fused multimodal feature F, and generates a new sample x=G(z, F). The discriminator receives the input sample x and the fused multimodal feature F at the same time, which is expressed as D(x, F).
[0074] Generator, the loss function maximizes the discriminator's authenticity score for the generated samples, and the implementation formula is:
[0075] ,
[0076] In the formula, E represents the expected value, Represents the distribution of random noise z And the expected value of the fusion feature F, Represents the score of the discriminator D on the generated sample G(z,F) and the fused multimodal feature F;
[0077] The discriminator loss function maximizes the ability to distinguish between real data and generated data. The implementation formula is:
[0078] ,
[0079] In the formula, represents the real data distribution, E represents the expected value, Represents the real data distribution And the expected value of the fusion feature F, represents the logarithm of the score of the discriminator D on the real data x and the fusion feature F, Represents the distribution of random noise z And the expected value of the fusion feature F, Represents the fake data generated by the generator to the discriminator D The logarithm of the rating.
[0080] The real-time market trend monitoring module selects the feature X related to the market trend from the multimodal features fused by the multimodal feature fusion module, and the implementation formula is:
[0081] ,
[0082] In the formula, Indicates market characteristics The information entropy of Indicates market characteristics In the value , i represents the i-th market characteristic, j represents the market characteristic The jth possible value of Indicates market characteristics The number of all possible values of Select the first k features with the highest information entropy as important features;
[0083] The multimodal feature fusion module effectively reduces the deviation and noise brought by single-modal data by extracting and fusing features of different modalities, and improves the accuracy and robustness of feature representation. At the same time, the introduction of generative adversarial networks further enhances the diversity and authenticity of the data, making market trend predictions and user behavior predictions more accurate.
[0084] According to the obtained market characteristics, the current market trend is predicted, and the implementation formula is:
[0085] ,
[0086] In the formula, represents the predicted value at the current moment, c represents the constant term, and They represent the autoregressive and moving average coefficients, respectively. p represents the order of the autoregressive term, q represents the order of the moving average term, i represents the lag order of the autoregressive part, from 1 to p, j represents the lag order of the moving average part, from 1 to q, ss Represents the white noise error term, and the predicted value at the current moment Put it into F to get a new feature set , according to the new set characteristics, the final forecast value of the market trend is obtained, and the implementation formula is:
[0087] ,
[0088] In the formula, Indicates the final forecast value of the market trend, represents a feature set, represents the weight of the mth market trend forecasting model, represents the predicted value of the mth market trend forecasting model, and M represents the number of market trend forecasting models.
[0089] Among them, the real-time market trend monitoring module updates the market trend prediction model according to the real-time data stream received by the multimodal feature fusion module, and the implementation formula is:
[0090] ,
[0091] In the formula, It represents the predicted value at the next time t+1. The final prediction model of the updated market trend is:
[0092] ,
[0093] The weight of the final prediction model of market trends is updated according to real-time data. The implementation formula is:
[0094] ,
[0095] In the formula, represents the weight of the m+1th prediction model, η represents the learning rate, represents the actual observed value, Represents the final forecast value of the market trend forecast model.
[0096] The real-time market trend monitoring module can monitor market changes in real time and make predictions based on historical data, ensuring the timeliness and effectiveness of decision-making.
[0097] The deep psychological analysis module extracts psychological state features from the multimodal feature fusion module. , each psychological state feature is assigned a weight , according to each psychological state feature and the corresponding weight, the comprehensive score of the psychological state is obtained, and the implementation formula is:
[0098] ,
[0099] In the formula, represents the comprehensive score of psychological state, k represents the number of psychological state characteristics, It represents the jth feature score on the i-th feature. The user's psychological state is analyzed according to the comprehensive score of the psychological state. The implementation formula is:
[0100] ,
[0101] In the formula, e represents the psychological state, and E represents the set of all possible psychological states. Indicates the psychological state e in the comprehensive score of psychological state The next score.
[0102] The behavior prediction incentive module predicts user behavior and designs corresponding incentive measures based on the market trend prediction results and psychological state analysis results. The comprehensive prediction implementation formula is:
[0103] ,
[0104] In the formula, Indicates that given the current market trend 、Future market trends Under the conditions of user psychological state e, the score of user behavior b is calculated, and corresponding measures are designed according to the comprehensive prediction score.
[0105] The dynamic decision-making module formulates marketing strategies based on the real-time market trend information of the market trend monitoring module and the user behavior prediction results and incentive measures of the behavior prediction incentive module. The optimal marketing strategy S is implemented as follows:
[0106] ,
[0107] In the formula, S represents the optimal marketing strategy, I(b) represents the incentives, and d represents the decision function, which is used to compare different marketing strategies. Indicates current market trends, It represents the future market trend, e represents the user's psychological state, and b represents the user's behavior.
[0108] The behavior prediction and incentive module comprehensively predicts user behavior based on market trends and psychological analysis results, and designs corresponding incentives to actively guide user behavior and maximize marketing effects.
[0109] The feedback optimization module collects the effect feedback and market data of marketing strategies in real time, and feeds the collected information back to the system for optimization.
[0110] In this example, specifically: the real-time market trend monitoring module selects the feature X related to the market trend from the multimodal features fused by the multimodal feature fusion module, and the implementation formula is:
[0111] ,
[0112] In the formula, Indicates market characteristics The information entropy of Indicates market characteristics In the value , i represents the i-th market characteristic, j represents the market characteristic The jth possible value of Indicates market characteristics The number of all possible values of Select the first k features with the highest information entropy as important features;
[0113] According to the obtained market characteristics, the current market trend is predicted, and the implementation formula is:
[0114] ,
[0115] In the formula, represents the predicted value at the current moment, c represents the constant term, and They represent the autoregressive and moving average coefficients respectively, p represents the order of the autoregressive term, q represents the order of the moving average term, i represents the lag order of the autoregressive part, from 1 to p, j represents the lag order of the moving average part, from 1 to q, s Represents the white noise error term, and the predicted value at the current moment Put it into F to get a new feature set , according to the new set characteristics, the final forecast value of the market trend is obtained, and the implementation formula is:
[0116] ,
[0117] In the formula, Indicates the final forecast value of the market trend, represents a feature set, represents the weight of the mth market trend forecasting model, represents the predicted value of the mth market trend forecasting model, and M represents the number of market trend forecasting models.
[0118] In this example, specifically: the real-time market trend monitoring module updates the market trend prediction model according to the real-time data stream received by the multimodal feature fusion module, and the implementation formula is:
[0119] ,
[0120] In the formula, It represents the predicted value at the next time t+1. The final prediction model of the updated market trend is:
[0121] ,
[0122] The weight of the final prediction model of market trends is updated according to real-time data. The implementation formula is:
[0123] ,
[0124] In the formula, represents the weight of the m+1th prediction model, η represents the learning rate, represents the actual observed value, Represents the final forecast value of the market trend forecast model.
[0125] In this example, specifically: the behavior prediction incentive module predicts user behavior and designs corresponding incentive measures based on the market trend prediction results and psychological state analysis results. The comprehensive prediction implementation formula is:
[0126] ,
[0127] In the formula, Indicates that given the current market trend 、Future market trends Under the conditions of user psychological state e, the score of user behavior b is calculated, and corresponding measures are designed according to the comprehensive prediction score.
[0128] In this example, specifically: the dynamic decision-making module formulates a marketing strategy based on the real-time market trend information of the market trend monitoring module and the user behavior prediction results and incentive measures of the behavior prediction incentive module. The optimal marketing strategy S is implemented as follows:
[0129] ,
[0130] In the formula, S represents the optimal marketing strategy, I(b) represents the incentives, and d represents the decision function, which is used to compare different marketing strategies. Indicates current market trends, It represents the future market trend, e represents the user's psychological state, and b represents the user's behavior.
[0131] Working principle: The multimodal data acquisition module collects multimodal data such as images, texts, and audio from various sources, processes the collected multimodal data into a unified format, and extracts the features of each modality through feature extraction technology. Then, by minimizing the differences between different modal features, these features are fused into a unified multimodal fusion feature.
[0132] Through the adversarial learning mechanism of the generator G and the discriminator D of the generative adversarial network, new data samples are generated and the generated samples are distinguished from the real samples. Through continuous adversarial training, the generative adversarial network can generate more real and useful data for enhancing multimodal fusion features. Features highly related to market trends are selected from the multimodal fusion features. Through time series analysis and prediction, real-time data and historical data are combined to monitor and predict market trends in real time. At the same time, the prediction model can be dynamically updated according to the real-time data stream. The deep psychological analysis module extracts features related to the user's psychological state from the multimodal fusion features, and analyzes the user's psychological state according to the psychological state comprehensive score by assigning weights to each psychological state feature and calculating the comprehensive score. The behavior prediction incentive module predicts user behavior based on the market trend prediction results and the user's psychological state analysis results, and designs corresponding incentives based on the prediction results. The dynamic decision-making module formulates the optimal marketing strategy based on the real-time information of market trends and the user's behavior prediction results and incentives. The feedback optimization module collects the effect feedback and market data after the implementation of the marketing strategy in real time, and feeds the collected information back to the system for optimizing various links such as multimodal feature fusion, market trend prediction, psychological analysis, behavior prediction and marketing strategy formulation.
[0133] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
[0134] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.
Claims
1. A marketing intelligent decision-making system based on multimodal learning, characterized by: It includes multimodal data acquisition module, multimodal feature fusion module, generative adversarial network, real-time market trend monitoring module, deep psychological analysis module, behavior prediction incentive module, dynamic decision-making module, and feedback optimization module; The multimodal feature fusion module unifies the format of the data collected by the multimodal data collection module, extracts multimodal features, and fuses them to form multimodal fusion features; The generative adversarial network is used to generate new data samples to simulate the behavior of real users and market scenarios; Wherein, the real-time market trend monitoring module is used to monitor and predict market trends in real time based on real-time data and historical data; Wherein, the deep psychological analysis module is used to analyze the user's psychological state; The behavior prediction incentive module predicts user behavior and designs corresponding incentive measures based on market trend monitoring results and deep psychological analysis results; The dynamic decision-making module formulates marketing strategies based on the real-time market trend information of the market trend monitoring module and the user behavior prediction results and incentive measures of the behavior prediction incentive module.
2. The marketing intelligent decision-making system based on multimodal learning according to claim 1, characterized in that: The multimodal feature fusion module extracts and fuses features based on the data collected by the multimodal data acquisition module. Assuming that the collected data modalities include images, texts and audios, and extracts features from the data modules, the image feature extraction is , text feature extraction is , image feature extraction is ,in, Represent the feature vectors of images, texts and audios respectively, compare the feature vectors of multimodal data, minimize the difference between different modal features, and implement the formula as follows: , In the formula, represents the alignment loss function, which minimizes the difference between different modal features, and g represents the alignment function, which measures the similarity between two features. The processed multimodal features are fused. Represents the calculation of alignment loss function The length of the vector output by each alignment function g in , after multimodal feature fusion, is expressed as , F represents the fused multimodal features, represents the weight coefficient, represents a non-linear activation function.
3. The marketing intelligent decision-making system based on multimodal learning according to claim 1, characterized in that: The generative adversarial network includes a generator G and a discriminator D. The generator receives a random noise z and a fused multimodal feature F, and generates a new sample x=G(z, F). The discriminator receives the input sample x and the fused multimodal feature F at the same time, which is expressed as D(x, F). Generator, the loss function maximizes the discriminator's authenticity score for the generated samples, and the implementation formula is: , In the formula, E represents the expected value, Represents the distribution of random noise z And the expected value of the fusion feature F, Represents the score of the discriminator D on the generated sample G(z,F) and the fused multimodal feature F; The discriminator loss function maximizes the ability to distinguish between real data and generated data. The implementation formula is: , In the formula, represents the real data distribution, E represents the expected value, Represents the real data distribution And the expected value of the fusion feature F, represents the logarithm of the score of the discriminator D on the real data x and the fusion feature F, Represents the distribution of random noise z And the expected value of the fusion feature F, Represents the fake data generated by the generator to the discriminator D The logarithm of the rating.
4. The marketing intelligent decision-making system based on multimodal learning according to claim 1, characterized in that: The real-time market trend monitoring module selects the feature X related to the market trend from the multimodal features fused by the multimodal feature fusion module, and the implementation formula is: , In the formula, Indicates market characteristics The information entropy of Indicates market characteristics In the value , i represents the i-th market characteristic, j represents the market characteristic The jth possible value of Indicates market characteristics The number of all possible values of Select the first k features with the highest information entropy as important features; According to the obtained market characteristics, the current market trend is predicted, and the implementation formula is: , In the formula, represents the predicted value at the current moment, c represents the constant term, and They represent the autoregressive and moving average coefficients respectively, p represents the order of the autoregressive term, q represents the order of the moving average term, i represents the lag order of the autoregressive part, from 1 to p, j represents the lag order of the moving average part, from 1 to q, Represents the white noise error term, and the predicted value at the current moment Put it into F to get a new feature set , according to the new set characteristics, the final forecast value of the market trend is obtained, and the implementation formula is: , In the formula, Indicates the final forecast value of the market trend, represents a feature set, represents the weight of the mth market trend forecasting model, represents the predicted value of the mth market trend forecasting model, and M represents the number of market trend forecasting models.
5. The marketing intelligent decision-making system based on multimodal learning according to claim 4 is characterized by: The real-time market trend monitoring module receives the real-time data stream from the multimodal feature fusion module and updates the market trend prediction model. The implementation formula is: , In the formula, It represents the predicted value at the next time t+1. The final prediction model of the updated market trend is: , The weight of the final prediction model of market trends is updated according to real-time data. The implementation formula is: , In the formula, represents the weight of the m+1th prediction model, η represents the learning rate, represents the actual observed value, Represents the final forecast value of the market trend forecast model.
6. The marketing intelligent decision-making system based on multimodal learning according to claim 1, characterized in that: The deep psychological analysis module extracts psychological state features from the multimodal feature fusion module , each psychological state feature is assigned a weight , according to each psychological state feature and the corresponding weight, the comprehensive score of the psychological state is obtained, and the implementation formula is: , In the formula, represents the comprehensive score of psychological state, k represents the number of psychological state characteristics, It represents the jth feature score on the i-th feature. The user's psychological state is analyzed according to the comprehensive score of the psychological state. The implementation formula is: , In the formula, e represents the psychological state, and E represents the set of all possible psychological states. Indicates the psychological state e in the comprehensive score of psychological state The next score.
7. The marketing intelligent decision-making system based on multimodal learning according to claim 1, characterized in that: The behavior prediction incentive module predicts user behavior and designs corresponding incentive measures based on the market trend prediction results and psychological state analysis results. The comprehensive prediction implementation formula is: , In the formula, Indicates that given the current market trend 、Future market trends Under the conditions of user psychological state e, the score of user behavior b is calculated, and corresponding measures are designed according to the comprehensive prediction score.
8. The marketing intelligent decision-making system based on multimodal learning according to claim 1, characterized in that: The dynamic decision-making module formulates marketing strategies based on the real-time market trend information of the market trend monitoring module and the user behavior prediction results and incentive measures of the behavior prediction incentive module. The optimal marketing strategy S is implemented as follows: , In the formula, S represents the optimal marketing strategy, I(b) represents the incentives, and d represents the decision function. Indicates current market trends, It represents the future market trend, e represents the user's psychological state, and b represents the user's behavior.
9. The marketing intelligent decision-making system based on multimodal learning according to claim 1, characterized in that: The feedback optimization module collects effect feedback and market data of marketing strategies in real time, and feeds the collected information back to the system for optimization.
Citation Information
Cited By
Service data management method and system for cross-border e-commerce
CN119558890A
Intelligent nutrition mode management method and system based on HAP multi-modal data
CN120448947A
