Marketing strategy intelligent analysis system based on multi-modal learning

Abstract

The invention discloses a marketing strategy intelligent analysis system based on multi-modal learning, and belongs to the technical field of multi-modal learning. The system comprises a time feature function, a spatial feature function, a modal feature function, a heterogeneous spatial-temporal feature fusion module, a dynamic weight distribution module, a user portrait generation module, a personalized strategy generation module, a marketing strategy analysis and evaluation module and a feedback optimization module. The time-space characteristics and modal characteristics of the data are fully considered, deep fusion is performed on the multi-modal data, the fused characteristics can more comprehensively and accurately reflect the actual situation of the market, the fusion strategy can be flexibly adjusted according to different business requirements, and the adaptability and analysis ability to the complex market environment are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of multimodal learning, and specifically refers to an intelligent analysis system for marketing strategies based on multimodal learning. Background Art

[0002] In today's digital marketing era, the market environment is complex and ever-changing, and consumer demands are increasingly diversified and personalized. The traditional way of formulating marketing strategies mainly relies on manual experience and limited data statistical analysis, and there are many problems;

[0003] However, there are still certain defects in the existing intelligent analysis of marketing strategies. The existing intelligent analysis of marketing strategies relies on a single type of data, fails to make full use of the advantages of multimodal data, ignores the information in the time and space dimensions, resulting in the inability to comprehensively reflect the dynamic changes of the market and regional differences. Usually, a fixed weight allocation scheme is adopted, which cannot be adjusted according to real-time market changes and business requirements, and there is no effective feedback mechanism to adjust the weight in real time, resulting in an inflexible decision-making process and difficulty in coping with the rapidly changing market environment. Usually, a general marketing strategy is used, which fails to be customized according to the characteristics of different user groups. Therefore, an intelligent analysis system for marketing strategies based on multimodal learning is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent analysis system for marketing strategies based on multimodal learning to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solutions: An intelligent analysis system for marketing strategies based on multimodal learning, including a multimodal data collection and preprocessing module, a heterogeneous spatio-temporal feature fusion module, a dynamic weight allocation module, a user portrait generation module, a personalized strategy generation module, a marketing strategy analysis and evaluation module, and a feedback optimization module;

[0006] The multimodal data collection and preprocessing module is used to obtain multimodal data from multiple channels through web crawlers and preprocess the multimodal data;

[0007] The heterogeneous spatio-temporal feature fusion module is used to fuse the multimodal data features from different sources and different time points after preprocessing;

[0008] The dynamic weight allocation module is used to adjust the importance weights of the fused heterogeneous spatio-temporal features in real time according to different business requirements and market environments;

[0009] The user portrait generation module is used to construct a user portrait according to the obtained multimodal data and the heterogeneous spatio-temporal features optimized by dynamic weight allocation;

[0010] The personalized strategy generation module is used to formulate personalized marketing strategies by combining the optimized heterogeneous spatio-temporal features with dynamic weight allocation and the user profile;

[0011] The marketing strategy analysis and evaluation module is used to conduct comprehensive analysis, evaluation and recommendation based on the formulated personalized marketing strategies;

[0012] The feedback optimization module is used to obtain in real time the feedback data for the marketing strategies and optimize the marketing strategies.

[0013] The multi-modal data acquisition and preprocessing module is wirelessly connected to the heterogeneous spatio-temporal feature fusion module and the user profile generation module. The heterogeneous spatio-temporal feature fusion module is wirelessly connected to the dynamic weight allocation module. The heterogeneous spatio-temporal feature fusion module is wirelessly connected to the user profile generation module and the personalized strategy generation module. The user profile generation module is wirelessly connected to the personalized strategy generation module. The personalized strategy generation module is wirelessly connected to the marketing strategy analysis and evaluation module. The marketing strategy analysis and evaluation module is wirelessly connected to the feedback optimization module. The feedback optimization module is wirelessly connected to the dynamic weight allocation module. The dynamic weight allocation module is wirelessly connected to the multi-modal data acquisition and preprocessing module.

[0014] Among them, the multi-modal data acquisition and preprocessing module obtains multi-modal data from multiple channels through web crawlers and preprocesses the multi-modal data; text data information, image information, audio information and video information are obtained from multiple channels through web crawlers, and data cleaning is performed on different types of data. Data cleaning includes removing duplicate data items, identifying and deleting low-quality content, error correction, data annotation and format standardization.

[0015] Among them, the heterogeneous spatio-temporal feature fusion module fuses the multi-modal data features from different sources and different time points after preprocessing; there are N data sources and M modal types, and the fused heterogeneous spatio-temporal feature is F. The implementation formula is:

[0016]

[0017] In the formula, F represents the fused heterogeneous spatio-temporal feature, represents the importance weight calculated for each data source and modality, represents the data feature vector of the j-th modality of the i-th source at time point t, represents the fusion function, which integrates the data features of different modalities by combining time features and space features.

[0018] Among them, the multi-modal data acquisition and preprocessing module, the fusion function is expressed as:

[0019]

[0020] In the formula, ψ(T i ) represents the time feature function, which extracts the time features of the data, and γ(S j ) represents the space feature function, which extracts the space features of the data. represents the modal feature function, which processes the specific data feature vectors. α, β, and δ represent the weighting coefficients used to balance the importance of time, space, and modal features.

[0021] The fusion function of the heterogeneous spatio-temporal feature fusion module combines the time feature function, the space feature function, and the modal feature function, fully considering the spatio-temporal characteristics and modal characteristics of the data, and performs deep fusion on the multi-modal data. The fused features can more comprehensively and accurately reflect the actual market situation, and can flexibly adjust the fusion strategy according to different business requirements, improving the adaptability and analysis ability to complex market environments.

[0022] Among them, the dynamic weight allocation module adjusts the importance weights of the fused heterogeneous spatio-temporal features in real time according to different business requirements and market environments, and obtains the data feature vectors and the fused heterogeneous spatio-temporal feature F, and performs dynamic weight adjustment according to business requirements, market changes, and real-time feedback data. The implementation formula is:

[0023]

[0024] In the formula, represents the data feature vector of the j-th modality from the i-th source at time point t. is the weight of represents the updated weight at time point t + 1. η represents the speed of controlling the weight update. represents the error adjustment factor, which reflects the performance error under the current weight. represents the adjustment factor based on historical performance. represents the feedback adjustment factor. α, β, and γ represent the weighting coefficients used to balance the importance of error, historical performance, and feedback.

[0025] The dynamic weight allocation module adjusts the importance weights of the fused heterogeneous spatio-temporal features in real time according to different business requirements, market changes, and real-time feedback data. The dynamic adjustment mechanism can ensure that in different scenarios, each feature can be accurately adjusted according to the actual situation. By introducing the error adjustment factor, the historical performance adjustment factor, and the feedback adjustment factor, it ensures that the weight allocation is more scientific and reasonable, thus improving the decision-making accuracy.

[0026] Among them, the user portrait generation module constructs a user portrait based on the acquired multimodal data and the heterogeneous spatio-temporal features optimized by dynamic weight allocation; obtains the heterogeneous spatio-temporal features F and data feature vectors optimized by dynamic weight allocation Construct a user portrait vector. The implementation formula for constructing the user portrait vector is as follows:

[0027]

[0028] In the formula, U represents the user portrait vector, N represents the number of data sources, M represents the number of modality types, represents the weight at the updated time point t + 1, represents the processing of the data feature vector, represents the data feature vector of the j-th modality of the i-th source at time point t, ψ(F (t) ) represents the processing of the optimized heterogeneous spatio-temporal features, and λ1, λ2 represent weighting coefficients to balance the importance of the data vector features and the optimized heterogeneous spatio-temporal features.

[0029] The user portrait generation module constructs a user portrait based on the heterogeneous spatio-temporal features optimized by dynamic weight allocation, comprehensively considers the number of data sources, the number of modality types, and the weights of each feature, and generates a comprehensive and accurate user portrait vector, which can more accurately capture the user's preferences and behavior patterns.

[0030] Among them, the personalized strategy generation module formulates a personalized marketing strategy according to the heterogeneous spatio-temporal features optimized by dynamic weight allocation in combination with the user portrait; obtains the constructed user portrait vector U and the heterogeneous spatio-temporal features F optimized by dynamic weight allocation, and generates a personalized marketing strategy according to the constructed user portrait vector U and the heterogeneous spatio-temporal features F optimized by dynamic weight allocation. The implementation formula is as follows:

[0031]

[0032] In the formula, S represents the finally generated personalized marketing strategy, N represents the number of features, represents the importance weight calculated for each feature at time point t, U represents the user portrait vector, F i (t) represents the i-th type of optimized heterogeneous spatio-temporal feature data, ρ represents the degree of association between the user portrait and the optimized heterogeneous spatio-temporal feature data, Ω represents the optimization function, generates the optimal strategy according to the user portrait and all heterogeneous spatio-temporal feature data in combination with the business objective, and μ1, μ2 represent weighting coefficients, and G represents the rule set.

[0033] The personalized marketing strategy generation combines the user profile and the optimized heterogeneous spatio-temporal features with dynamic weight allocation. The personalized strategy generation module can formulate a more accurate and personalized marketing strategy, fully considering the individual differences of users and market characteristics, and can formulate exclusive strategies for different user groups, improving the marketing effect.

[0034] Among them, the marketing strategy analysis and evaluation module conducts comprehensive analysis, evaluation and recommendation according to the formulated personalized marketing strategy; obtains the formulated personalized marketing strategy S, sets evaluation indicators, quantitatively evaluates the effects of each strategy, calculates the actual values of each indicator, and compares them with the expected goals, identifies the potential risk factors during the strategy implementation process, and evaluates their impact degrees. Ranks all strategies according to the evaluation results, gives priority to recommending the best-performing strategy, and recommends the most suitable personalized marketing strategy according to different user groups.

[0035] The marketing strategy analysis and evaluation module quantitatively evaluates the effect of the personalized marketing strategy by setting evaluation indicators, compares it with the expected goal, identifies the risk factors and their impact degrees, ranks and recommends the strategy according to the evaluation result, and generates the optimal strategy according to the optimization function, effectively improving the effect and conversion rate of the marketing activity.

[0036] Among them, the feedback optimization module obtains in real time the feedback data for optimizing the marketing strategy; obtains in real time the data information of the user's feedback on the marketing strategy, conducts a preliminary analysis of the collected feedback data, identifies the obvious trends, segments the user groups according to different dimensions, analyzes the different reactions of different groups to the marketing strategy, evaluates the effect of the marketing strategy according to the feedback data, compares the gap between the actual performance and the expected goal, deeply analyzes the strategy problems, recalculates the importance weights of each data source and feature through the feedback data, increases the weights of relevant features for the strategies with good performance; otherwise, reduces the weights, and feeds the new weight values back to the dynamic weight allocation module.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] 1. Through the fusion function of the heterogeneous spatio-temporal feature fusion module of the present invention, combining the time feature function, the space feature function and the modal feature function, fully considering the spatio-temporal characteristics and modal characteristics of the data, deeply fusing the multi-modal data, the fused features can more comprehensively and accurately reflect the actual situation of the market, and can flexibly adjust the fusion strategy according to different business requirements, improving the adaptability and analysis ability to the complex market environment;

[0039] 2. The present invention adjusts the importance weights of the fused heterogeneous spatio-temporal features in real time through a dynamic weight allocation module according to different business requirements, market changes, and real-time feedback data. The dynamic adjustment mechanism can ensure that in different scenarios, each feature can be accurately adjusted according to the actual situation. By introducing an error adjustment factor, a historical performance adjustment factor, and a feedback adjustment factor, the weight allocation is ensured to be more scientific and reasonable, thereby improving the decision-making accuracy.

[0040] 3. The present invention generates a personalized marketing strategy by combining the user portrait and the optimized heterogeneous spatio-temporal features with dynamic weight allocation. The personalized strategy generation module can formulate a more accurate and personalized marketing strategy, fully considering the individual differences of users and market characteristics, and can formulate exclusive strategies for different user groups, improving the marketing effect.

[0041] 4. The present invention evaluates the effect of the personalized marketing strategy quantitatively through a marketing strategy analysis and evaluation module by setting evaluation indicators, compares it with the expected goal, identifies risk factors and their influencing degrees, ranks and recommends the strategy according to the evaluation results, and generates the optimal strategy according to the optimization function, effectively improving the effect and conversion rate of the marketing activity. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic structural diagram of the intelligent marketing strategy analysis system based on multi-modal learning of the present invention;

[0043] Figure 2 It is the operation process of the intelligent marketing strategy analysis system based on multi-modal learning of the present invention Figure 1 ;

[0044] Figure 3 It is the operation process of the intelligent marketing strategy analysis system based on multi-modal learning of the present invention Figure 2 ;

[0045] Figure 4 It is the operation process of the intelligent marketing strategy analysis system based on multi-modal learning of the present invention Figure 3 . DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] Embodiment

[0048] Please refer to Figures 1 - 4As shown in the figure, the present invention provides a technical solution, including a multi-modal data acquisition and preprocessing module, a heterogeneous spatio-temporal feature fusion module, a dynamic weight allocation module, a user profile generation module, a personalized strategy generation module, a marketing strategy analysis and evaluation module, and a feedback optimization module;

[0049] The multi-modal data acquisition and preprocessing module is used to obtain multi-modal data from multiple channels through web crawlers and preprocess the multi-modal data;

[0050] The heterogeneous spatio-temporal feature fusion module is used to fuse the features of multi-modal data from different sources and different time points after preprocessing;

[0051] The dynamic weight allocation module is used to adjust the importance weights of the fused heterogeneous spatio-temporal features in real time according to different business requirements and market environments;

[0052] The user profile generation module is used to construct a user profile according to the obtained multi-modal data and the heterogeneous spatio-temporal features optimized by dynamic weight allocation;

[0053] The personalized strategy generation module is used to formulate a personalized marketing strategy according to the heterogeneous spatio-temporal features optimized by dynamic weight allocation in combination with the user profile;

[0054] The marketing strategy analysis and evaluation module is used to conduct comprehensive analysis, evaluation and recommendation according to the formulated personalized marketing strategy;

[0055] The feedback optimization module is used to obtain in real time the feedback data for the marketing strategy to optimize the marketing strategy.

[0056] The multi-modal data acquisition and preprocessing module is wirelessly connected to the heterogeneous spatio-temporal feature fusion module and the user profile generation module. The heterogeneous spatio-temporal feature fusion module is wirelessly connected to the dynamic weight allocation module. The heterogeneous spatio-temporal feature fusion module is wirelessly connected to the user profile generation module and the personalized strategy generation module. The user profile generation module is wirelessly connected to the personalized strategy generation module. The personalized strategy generation module is wirelessly connected to the marketing strategy analysis and evaluation module. The marketing strategy analysis and evaluation module is wirelessly connected to the feedback optimization module. The feedback optimization module is wirelessly connected to the dynamic weight allocation module. The dynamic weight allocation module is wirelessly connected to the multi-modal data acquisition and preprocessing module.

[0057] Among them, the multi-modal data acquisition and preprocessing module obtains multi-modal data from multiple channels through web crawlers and preprocesses the multi-modal data; obtains text data information, image information, audio information and video information from multiple channels through web crawlers, and conducts data cleaning on different types of data. Data cleaning includes removing duplicate data items, identifying and deleting low-quality content, error correction, data annotation and format standardization.

[0058] Among them, the heterogeneous spatio-temporal feature fusion module fuses the multi-modal data features from different sources and at different time points after preprocessing; there are N data sources and M modal types, and the fused heterogeneous spatio-temporal feature is F. The implementation formula is:

[0059]

[0060] In the formula, F represents the fused heterogeneous spatio-temporal feature, represents the importance weight calculated for each data source and modality, represents the data feature vector of the j-th modality at the i-th source at time point t, represents the fusion function, which integrates the data features of different modalities by combining the time feature and the space feature.

[0061] Among them, the multi-modal data acquisition and preprocessing module, the fusion function is expressed as:

[0062]

[0063] In the formula, ψ(T i ) represents the time feature function, which extracts the time feature of the data, and γ(S j ) represents the space feature function, which extracts the space feature of the data, represents the modality feature function, which processes the specific data feature vector, and α, β, δ represent the weighting coefficients, which are used to balance the importance of the time, space, and modality features.

[0064] The fusion function of the heterogeneous spatio-temporal feature fusion module combines the time feature function, the space feature function, and the modality feature function, fully considers the spatio-temporal characteristics and modality characteristics of the data, deeply fuses the multi-modal data, and the fused features can more comprehensively and accurately reflect the actual market situation, and can flexibly adjust the fusion strategy according to different business requirements, improving the adaptability and analysis ability to the complex market environment

[0065] Among them, the dynamic weight allocation module adjusts the importance weight of the fused heterogeneous spatio-temporal feature in real time according to different business requirements and market environments; obtains the data feature vector and the fused heterogeneous spatio-temporal feature F, and performs dynamic weight adjustment according to business requirements, market changes, and real-time feedback data. The implementation formula is:

[0066]

[0067] In the formula, represents the data feature vector of the j-th modality of the i-th source at time point t The weight, represents the weight at the updated time point t+1, and η represents the speed of controlling the weight update. represents the error adjustment factor, reflecting the performance error under the current weight. represents the adjustment factor based on historical performance. represents the adjustment factor of feedback. α, β, and γ represent the weighting coefficients, which are used to balance the importance of error, historical performance, and feedback.

[0068] The dynamic weight allocation module, according to different business requirements, market changes, and real-time feedback data, dynamically adjusts the importance weights of the fused heterogeneous spatio-temporal features by real-time adjustment. The dynamic adjustment mechanism can ensure that in different scenarios, each feature can be accurately adjusted according to the actual situation. By introducing the error adjustment factor, historical performance adjustment factor, and feedback adjustment factor, it ensures that the weight allocation is more scientific and reasonable, thereby improving the decision-making accuracy.

[0069] Among them, the user profile generation module constructs a user profile according to the obtained multi-modal data and the heterogeneous spatio-temporal features optimized by dynamic weight allocation; obtains the heterogeneous spatio-temporal features F and data feature vectors optimized by dynamic weight allocation. Constructs a user profile vector. The formula for constructing the user profile vector is:

[0070]

[0071] In the formula, U represents the user profile vector, N represents the number of data sources, and M represents the number of modality types. represents the weight at the updated time point t+1. represents the processing of the data feature vector. represents the data feature vector of the j-th modality of the i-th source at time point t, and ψ(F (t) ) represents the processing of the optimized heterogeneous spatio-temporal features. λ1 and λ2 represent the weighting coefficients, balancing the importance of the data vector features and the optimized heterogeneous spatio-temporal features.

[0072] The user profile generation module constructs a user profile according to the heterogeneous spatio-temporal features optimized by dynamic weight allocation, comprehensively considering the number of data sources, the number of modality types, and the weights of each feature, and generates a comprehensive and accurate user profile vector, which can more accurately capture the user's preferences and behavior patterns.

[0073] Among them, the personalized strategy generation module formulates a personalized marketing strategy according to the optimized heterogeneous spatio-temporal features combined with the user portrait through dynamic weight allocation; obtains the constructed user portrait vector U and the optimized heterogeneous spatio-temporal features F after dynamic weight allocation, and generates a personalized marketing strategy according to the constructed user portrait vector U and the optimized heterogeneous spatio-temporal features F after dynamic weight allocation. The implementation formula is as follows:

[0074]

[0075] In the formula, S represents the finally generated personalized marketing strategy, N represents the number of features, represents the importance weight calculated for each feature at time point t, U represents the user portrait vector, and F i (t) represents the i-th optimized heterogeneous spatio-temporal feature data, ρ represents the degree of association between the user portrait and the optimized heterogeneous spatio-temporal feature data, Ω represents the optimization function, generates the optimal strategy according to the user portrait and all heterogeneous spatio-temporal feature data combined with the business objective, μ1 and μ2 represent the weighting coefficients, and G represents the rule set.

[0076] The generation of the personalized marketing strategy combines the user portrait and the optimized heterogeneous spatio-temporal features after dynamic weight allocation. The personalized strategy generation module can formulate a more accurate and personalized marketing strategy, fully considering the individual differences of users and market characteristics, and can formulate exclusive strategies for different user groups, improving the marketing effect.

[0077] Among them, the marketing strategy analysis and evaluation module conducts comprehensive analysis, evaluation and recommendation according to the formulated personalized marketing strategy; obtains the formulated personalized marketing strategy S, sets evaluation indicators, quantitatively evaluates the effect of each strategy, calculates the actual values of each indicator, compares them with the expected goals, identifies the possible risk factors in the process of strategy implementation, and evaluates their impact degrees. Rank all strategies according to the evaluation results, and give priority to recommending the best-performing strategies. According to different user groups, recommend the most suitable personalized marketing strategy.

[0078] The marketing strategy analysis and evaluation module quantitatively evaluates the effect of the personalized marketing strategy by setting evaluation indicators, compares it with the expected goal, identifies the risk factors and their impact degrees, ranks and recommends the strategies according to the evaluation results, and generates the optimal strategy according to the optimization function, effectively improving the effect and conversion rate of marketing activities.

[0079] Among them, the feedback optimization module obtains in real time the feedback data for optimizing the marketing strategy; obtains in real time the data information of the user's feedback on the marketing strategy, conducts a preliminary analysis of the collected feedback data, identifies obvious trends, segments the user groups according to different dimensions, analyzes the different responses of different groups to the marketing strategy, evaluates the effect of the marketing strategy according to the feedback data, compares the gap between the actual performance and the expected goal, deeply analyzes the strategy problems, recalculates the importance weights of each data source and feature through the feedback data, for the strategies with good performance, increases the weights of relevant features; otherwise, reduces the weights, and feeds back the new weight values to the dynamic weight allocation module.

[0080] Working principle: The multi-modal data acquisition and preprocessing module obtains multi-modal data from multiple channels through web crawler technology. These data may include text, images, audio, and video, and preprocesses the obtained multi-modal data, including data cleaning, data annotation, and format standardization.

[0081] The heterogeneous spatio-temporal feature fusion module fuses the features of the preprocessed multi-modal data, considering the features of different data sources and time points, and integrates the data features of different modalities to form heterogeneous spatio-temporal features by setting a fusion function and combining time features and space features. The dynamic weight allocation module adjusts the importance weights of the fused heterogeneous spatio-temporal features in real time according to business requirements and market environment. The dynamic weight adjustment updates the weights of the feature vectors according to business requirements, market changes, and real-time feedback data factors. The user portrait generation module constructs a user portrait by using the multi-modal data obtained and the heterogeneous spatio-temporal features optimized by the dynamic weight allocation. Through the user portrait vector construction formula, combining the data feature vector and the optimized heterogeneous spatio-temporal features, a user portrait is formed. The personalized strategy generation module formulates a personalized marketing strategy according to the heterogeneous spatio-temporal features optimized by the dynamic weight allocation and the user portrait. Through the personalized strategy generation formula, combining factors such as the user portrait vector, the optimized heterogeneous spatio-temporal feature data, and the feature importance weights, the final personalized marketing strategy is generated. The marketing strategy analysis and evaluation module comprehensively analyzes and evaluates the formulated personalized marketing strategies, sets evaluation indicators, quantifies the effects of each strategy, identifies risk factors and evaluates their impact degrees, ranks all strategies according to the evaluation results, preferentially recommends the best-performing strategies, and recommends the most suitable personalized marketing strategies for different user groups. The feedback optimization module obtains in real time the feedback data of the user on the marketing strategy, conducts a preliminary analysis, evaluates the effect of the marketing strategy according to the feedback data, compares the gap between the actual performance and the expected goal, recalculates the importance weights of each data source and feature through the feedback data, and conducts feedback optimization on the dynamic weight allocation module according to the new weight values.

[0082] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

[0083] The above describes the present invention and its embodiments. Such a description is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In general, if those of ordinary skill in the art are inspired by it and, without departing from the purpose of the present invention, design similar structural forms and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention.

Claims

1. The marketing strategy intelligent analysis system based on multimodal learning is characterized by: It includes multimodal data collection preprocessing module, heterogeneous spatiotemporal feature fusion module, dynamic weight allocation module, user portrait generation module, personalized strategy generation module, marketing strategy analysis and evaluation module and feedback optimization module; The multimodal data acquisition preprocessing module is used to obtain multimodal data from multiple channels through a crawler and preprocess the multimodal data; The heterogeneous spatiotemporal feature fusion module is used to fuse the pre-processed multimodal data features from different sources and at different time points; The dynamic weight allocation module is used to adjust the importance weights of the fused heterogeneous spatiotemporal features in real time according to different business requirements and market environments; The user portrait generation module is used to construct a user portrait based on the acquired multimodal data and the heterogeneous spatiotemporal features after dynamic weight allocation optimization; The personalized strategy generation module is used to formulate personalized marketing strategies based on the heterogeneous spatiotemporal features optimized by dynamic weight allocation and combined with user portraits; The marketing strategy analysis and evaluation module is used to conduct comprehensive analysis and evaluation and make recommendations based on the formulated personalized marketing strategies; The feedback optimization module is used to obtain feedback data for marketing strategies in real time to optimize the marketing strategies.

2. The marketing strategy intelligent analysis system based on multimodal learning according to claim 1 is characterized by: The multimodal data acquisition preprocessing module obtains multimodal data from multiple channels through a crawler and preprocesses the multimodal data; obtains text data information, image information, audio information and video information from multiple channels through a crawler, and performs data cleaning on different types of data, wherein the data cleaning includes removing duplicate data items, identifying and deleting low-quality content, error correction, data annotation and format standardization.

3. The marketing strategy intelligent analysis system based on multimodal learning according to claim 1 is characterized by: The heterogeneous spatiotemporal feature fusion module fuses the preprocessed multimodal data features from different sources and at different time points. Suppose there are N data sources and M modality types, the fused heterogeneous spatiotemporal feature is F, and the implementation formula is: In the formula, F represents the fused heterogeneous spatiotemporal features. represents the importance weight calculated for each data source and modality, represents the data feature vector of the jth mode of the ith source at time point t, Represents the fusion function, which combines temporal and spatial features to integrate data features of different modalities.

4. The marketing strategy intelligent analysis system based on multimodal learning according to claim 2 is characterized by: The multimodal data acquisition preprocessing module, fusion function It is expressed as: In the formula, ψ(T i ) represents the time characteristic function, extracting the time characteristics of the data, γ(S j ) represents the spatial feature function, which extracts the spatial features of the data. It represents the modal characteristic function, which processes the specific data eigenvector. α, β, and δ represent weighting coefficients, which are used to balance the importance of time, space, and modal characteristics.

5. The marketing strategy intelligent analysis system based on multimodal learning according to claim 1 is characterized by: The dynamic weight allocation module adjusts the importance weights of the fused heterogeneous spatiotemporal features in real time according to different business needs and market environments; obtains the data feature vector The fused heterogeneous spatiotemporal features F are dynamically weighted according to business needs, market changes, and real-time feedback data. The implementation formula is: In the formula, Represents the data feature vector of the jth mode of the ith source at time t The weight of represents the weight at the updated time point t+1, η represents the speed of controlling the weight update, Represents the error adjustment factor, reflecting the performance error under the current weight, represents the adjustment factor based on historical performance, represents the adjustment factor of feedback, and α, β, and γ represent weighting coefficients used to balance the importance of error, historical performance, and feedback.

6. The marketing strategy intelligent analysis system based on multimodal learning according to claim 1 is characterized by: The user portrait generation module constructs a user portrait based on the acquired multimodal data and the heterogeneous spatiotemporal features optimized by dynamic weight allocation; Obtain the heterogeneous spatiotemporal feature structure F and data feature vector after dynamic weight allocation optimization Construct a user portrait vector. The formula for constructing a user portrait vector is: In the formula, U represents the user portrait vector, N represents the number of data sources, and M represents the number of modal types. represents the weight at the updated time point t+1, Indicates processing of data feature vectors. represents the data feature vector of the jth mode of the ith source at time point t, ψ(F (t) ) represents the processing of the optimized heterogeneous spatiotemporal features, λ1 and λ2 represent weighting coefficients, balancing the importance of data vector features and the optimized heterogeneous spatiotemporal features.

7. The marketing strategy intelligent analysis system based on multimodal learning according to claim 1 is characterized by: The personalized strategy generation module formulates personalized marketing strategies based on the heterogeneous spatiotemporal features optimized by dynamic weight allocation and combined with user portraits; Obtain the constructed user portrait vector U and the heterogeneous spatiotemporal feature structure F after dynamic weight allocation optimization, and generate a personalized marketing strategy based on the constructed user portrait vector U and the heterogeneous spatiotemporal feature structure F after dynamic weight allocation optimization. The implementation formula is: In the formula, S represents the personalized marketing strategy finally generated, and N represents the number of features. represents the importance weight of each feature calculated at time point t, U represents the user portrait vector, and F i (t) represents the i-th optimized heterogeneous spatiotemporal feature data, ρ represents the degree of association between the user profile and the optimized heterogeneous spatiotemporal feature data, Ω represents the optimization function, which generates the optimal strategy based on the user profile and all heterogeneous spatiotemporal feature data combined with the business goal, μ1, μ2 represent weighting coefficients, and G represents the rule set.

8. The marketing strategy intelligent analysis system based on multimodal learning according to claim 1 is characterized by: The marketing strategy analysis and evaluation module conducts comprehensive analysis and evaluation based on the formulated personalized marketing strategy and makes recommendations; obtains the formulated personalized marketing strategy S, sets evaluation indicators, conducts quantitative evaluation on the effect of each strategy, calculates the actual value of each indicator, and compares it with the expected target, identifies the risk factors that may arise during the implementation of the strategy, and evaluates their impact, ranks all strategies according to the evaluation results, prioritizes the best performing strategies, and recommends the most suitable personalized marketing strategies according to different user groups.

9. The marketing strategy intelligent analysis system based on multimodal learning according to claim 1 is characterized by: The feedback optimization module obtains feedback data for marketing strategies in real time to optimize marketing strategies; obtains data information on user feedback on marketing strategies in real time, performs preliminary analysis on the collected feedback data, identifies obvious trends, segments user groups according to different dimensions, analyzes different responses of different groups to marketing strategies, evaluates the effectiveness of marketing strategies based on feedback data, compares the gap between actual performance and expected goals, deeply analyzes strategic issues, recalculates the importance weights of various data sources and features through feedback data, and increases the weights of relevant features for strategies that perform well; Otherwise, the weight is reduced and the new weight value is fed back to the dynamic weight allocation module.

Images