Food sales website promotion platform and method based on dynamic network intention analysis

Abstract

The invention discloses a food sales website promotion platform and method based on dynamic network intention analysis, and relates to the technical field of food sales. The method comprises the steps of collecting multi-source user data through a three-level authorization mechanism, extracting a feature vector at a local device, generating a desensitization label, destroying original data, constructing a four-dimensional dynamic network based on the desensitization label, dynamically updating a node weight, carrying out intention analysis based on the four-dimensional dynamic network and the feature vector, generating a final intention label, and carrying out intention analysis based on the final intention label. And collecting competing product data, generating a multi-modal promotion material in combination with the final intention label, constructing a promotion decision matrix, adaptively pushing the multi-modal promotion material to a user, obtaining user interaction data, extracting feedback characteristics, and optimizing the four-dimensional dynamic network and the promotion decision matrix. According to the method, the user characteristics can be obtained on the premise of ensuring the privacy security of the user, the food consumption intention of the user is captured, the multi-modal promotion material is generated in combination with the competitive product data, and the promotion efficiency and conversion rate of the food sales website are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of food sales, and in particular to a food sales website promotion platform and method based on dynamic network intention analysis. BACKGROUND

[0002] In the current rapid development of Internet technology, the competition of food sales websites is becoming increasingly fierce. How to accurately grasp user needs and effectively promote has become the key. Traditional food sales website promotion methods are mostly based on static data and simple rules, which are difficult to capture the changing consumption intentions of users in real time. Not only is the matching degree of promotion materials and user needs low, causing resource waste, but the user privacy protection mechanism is weak, and the original data storage has security risks. At the same time, there is a lack of effective use of user feedback, which cannot realize dynamic optimization of promotion strategies, and it is difficult to meet the dynamic and personalized promotion needs of the food sales market. SUMMARY

[0003] The present application provides a food sales website promotion platform based on dynamic network intention analysis, comprising: A multi-source user data acquisition and processing module acquires multi-source user data through a three-level authorization mechanism, extracts feature vectors on a local device, generates desensitization labels, and destroys original data; A four-dimensional dynamic network construction and update module constructs a four-dimensional dynamic network based on the desensitization labels and dynamically updates node weights; An intention analysis module performs intention analysis based on the four-dimensional dynamic network and the feature vectors to generate final intention labels; A multi-modal promotion material acquisition module acquires competitor data and generates multi-modal promotion materials in combination with the final intention labels; A material pushing module constructs a promotion decision matrix and adaptively pushes multi-modal promotion materials to users; A network and matrix optimization module acquires user interaction data, extracts feedback features, and optimizes the four-dimensional dynamic network and the promotion decision matrix.

[0004] The food sales website promotion platform based on dynamic network intention analysis as described above, wherein the multi-source user data acquisition and processing module specifically comprises: A source user data acquisition sub-module acquires multi-source user data through three-level authorization mechanism data layer authorization of users; A vector extraction and desensitization label generation sub-module pre-processes the acquired multi-source user data through federated learning technology, extracts feature vectors on a local device, generates desensitization labels, and triggers random noise signals to destroy original data.

[0005] The food sales website promotion platform based on dynamic network intention analysis as described above, wherein the four-dimensional dynamic network construction and update module specifically comprises: A four-dimensional dynamic network construction submodule constructs a four-dimensional dynamic network of "user-scene-food attribute-intention" based on the desensitization label; A node weight updating submodule dynamically updates the node weights of the four-dimensional dynamic network through a multi-dimensional weight updating algorithm.

[0006] The food sales website promotion platform based on dynamic network intention analysis as described above, wherein the intention analysis module specifically comprises: A double-channel intention analysis submodule obtains multi-hop associated intentions and real-time important features through double-channel intention analysis based on the four-dimensional dynamic network and the extracted feature vectors; A final intention label generation submodule generates a final intention label through a contradiction fusion algorithm based on the multi-hop associated intentions and the real-time important features.

[0007] The food sales website promotion platform based on dynamic network intention analysis as described above, wherein the multi-modal promotion material acquisition module specifically comprises: A competitor data acquisition submodule crawls competitor public data by simulating real behaviors through distributed proxy nodes and filters false information through an adversarial denoising network; A promotion material generation submodule generates multi-modal promotion materials through an "attribute-sentiment-visual" joint mapper based on the final intention label and the competitor public data.

[0008] The food sales website promotion platform based on dynamic network intention analysis as described above, wherein the material pushing module specifically comprises: A promotion decision matrix construction submodule acquires promotion adaptation factors and constructs a promotion decision matrix; An adaptive material pushing submodule adaptively pushes multi-modal promotion materials to users based on the promotion decision matrix.

[0009] The food sales website promotion platform based on dynamic network intention analysis as described above, wherein the network and matrix optimization module specifically comprises: An interaction data acquisition submodule acquires interaction data of users and multi-modal promotion materials through a three-level authorization mechanism; An optimization submodule extracts feedback features based on the interaction data and optimizes the four-dimensional dynamic network and the promotion decision matrix.

[0010] The application further provides a food sales website promotion method based on dynamic network intention analysis, comprising: Step S1: Collecting multi-source user data through a three-level authorization mechanism, extracting feature vectors on a local device, generating desensitization labels, and destroying original data; Step S2: Constructing a four-dimensional dynamic network based on the desensitization labels and dynamically updating node weights; Step S3, intent analysis based on four-dimensional dynamic network and feature vector, generating final intent label; Step S4, collecting competitive product data, combining final intent label to generate multi-modal promotion materials; Step S5, constructing promotion decision matrix, and adaptively pushing multi-modal promotion materials to users; Step S6, obtaining user interaction data, extracting feedback features, optimizing four-dimensional dynamic network and promotion decision matrix.

[0011] The food sales website promotion method based on dynamic network intent analysis as described above, wherein, further, multi-source user data is collected through a three-level authorization mechanism, and feature vectors are extracted on local devices to generate desensitization labels and destroy original data, including the following sub-steps: Step S11, obtaining user data hierarchical authorization through a three-level authorization mechanism, and collecting multi-source user data; Step S12, pre-processing the collected multi-source user data through federated learning technology, extracting feature vectors on local devices, and generating desensitization labels to trigger random noise signals to destroy original data.

[0012] The food sales website promotion method based on dynamic network intent analysis as described above, wherein, further, a four-dimensional dynamic network is constructed based on the desensitization label, and the node weights are dynamically updated, including the following sub-steps: Step S21, constructing a four-dimensional dynamic network of "user-scene-food attribute-intent" based on the desensitization label; Step S22, dynamically updating the node weights of the four-dimensional dynamic network through a multi-dimensional weight updating algorithm.

[0013] The food sales website promotion method based on dynamic network intent analysis as described above, wherein, further, intent analysis is performed based on the four-dimensional dynamic network and the feature vector to generate a final intent label, including the following sub-steps: Step S31, obtaining multi-hop associated intent and real-time important features through double-channel intent analysis based on the four-dimensional dynamic network and the extracted feature vector; Step S32, generating a final intent label based on multi-hop associated intent and real-time important features through a contradiction fusion algorithm.

[0014] The food sales website promotion method based on dynamic network intent analysis as described above, wherein, further, competitive product data is collected, and multi-modal promotion materials are generated in combination with the final intent label, including the following sub-steps: Step S41, simulating real behavior through distributed proxy nodes to crawl competitive public data, and filtering false information through an adversarial denoising network; Step S42: Generate multimodal promotional materials based on the final intent label and competitor public data through the "attribute-emotion-vision" joint mapper.

[0015] The above-mentioned food sales website promotion method based on dynamic network intent analysis, wherein further, constructing a promotion decision matrix and adaptively pushing multimodal promotional materials to users includes the following sub-steps: Step S51: Obtain promotion adaptation factors and construct a promotion decision matrix; Step S52: Based on the promotion decision matrix, adaptively push multimodal promotion materials to the user.

[0016] The above-mentioned method for promoting a food sales website based on dynamic network intent analysis further includes the following sub-steps: obtaining user interaction data, extracting feedback features, and optimizing the four-dimensional dynamic network and promotion decision matrix: Step S61: Obtaining interaction data between the user and the multimodal promotional material through a three-level authorization mechanism; Step S62: Extract feedback features based on the interaction data and optimize the four-dimensional dynamic network and promotion decision matrix.

[0017] The beneficial effects achieved by the present invention are as follows: the present invention can obtain user characteristics while ensuring user privacy and security, accurately capture users' ever-changing food consumption intentions, improve the accuracy of intention judgment, and generate multimodal promotional materials in combination with competitor data, thereby enhancing the attractiveness and competitiveness of promotional content, effectively improving users' shopping experience, and improving the promotion efficiency and conversion rate of food sales websites. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0019] Figure 1 This is a schematic diagram of a food sales website promotion platform based on dynamic network intent analysis provided in Example 1 of the present application; Figure 2 This is a flow chart of the food sales website promotion method based on dynamic network intent analysis provided in Example 2 of the present application. DETAILED DESCRIPTION

[0020] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0021] Example 1 like Figure 1 As shown, the first embodiment of the present application provides a food sales website promotion platform based on dynamic network intent analysis, including: The multi-source user data collection and processing module 11 collects multi-source user data through a three-level authorization mechanism, extracts feature vectors on the local device, generates desensitization labels, and destroys the original data; Furthermore, the multi-source user data collection and processing module 11 includes the following submodules: The multi-source user data collection submodule 111 obtains the user's data layered authorization through a three-level authorization mechanism and collects multi-source user data; Specifically, the three-level authorization mechanism obtains user hierarchical authorization by obtaining user data authorization through a visual pop-up interface. When the user visits for the first time, the first-level authorization mechanism is activated to obtain the default authorization for basic behavior data. When the user triggers deep interaction, the second-level authorization mechanism is activated to obtain authorization for biometric data and environmental data. When the user is in the second-level authorization state and an intent conflict is detected, the third-level authorization mechanism is activated to obtain authorization for intent data, and multi-source user data is collected based on the user's authorization.

[0022] The vector extraction desensitization label generation submodule 112 pre-processes the collected multi-source user data through federated learning technology, extracts feature vectors on the local device, generates desensitization labels, and triggers random noise signals to destroy the original data; Specifically, the collected data is preprocessed on the local device through federated learning technology, non-reversible feature vectors are extracted, and formulas are generated through desensitized labels. Get the desensitized label of the processed data, where: For the generated A desensitizing label, The value range is , is the number of desensitized labels, is the irreversible chaos processing factor, is the feature compression function, is a k-bit locality sensitive hash function, For the feature vectors, is bitwise exclusive OR, is the chaotic mapping scrambling function, a locally generated random noise vector, a federated gradient feature processing factor, a gradient quantization encoding function, a parameter trained based on federated learning and the gradient calculation function of the th feature vector, a food industry attribute mask factor, an attribute mask result transformation function, an element-wise multiplication, a food industry attribute mask vector. After obtaining the desensitized label, the generation of the random noise signal is automatically triggered to cover the original user data for destruction.

[0023] The four-dimensional dynamic network construction and update module 12 constructs a four-dimensional dynamic network based on the desensitized label and dynamically updates the node weights; Further, the four-dimensional dynamic network construction and update module 12 includes the following sub-modules: The four-dimensional dynamic network construction sub-module 121 constructs a four-dimensional dynamic network of “user-scene-food attribute-intention” based on the desensitized label; Specifically, the desensitized labels corresponding to the user-food interaction time axis dimension, the scene dimension, the food attribute dimension, and the intention dimension are extracted from the desensitized label, the desensitized labels of different dimensions are aligned through the TransE algorithm, a four-dimensional dynamic network is constructed, and the initial weights of each node of the four-dimensional dynamic network are generated according to the correlation degree of each desensitized label.

[0024] The node weight update sub-module 122 dynamically updates the node weights of the four-dimensional dynamic network through a multi-dimensional weight update algorithm; Specifically, when the input feedback feature or the desensitized label is updated, the data is de-dimensioned, the updated weights of each node are calculated through a multi-dimensional weight update algorithm and the node weights of the four-dimensional dynamic network are dynamically updated, wherein, is the updated weight of the th node, the value range of , is the number of nodes of the four-dimensional dynamic network, is a historical weight decay factor, is the weight of the th node before update, is a label update coefficient, is the number of desensitized labels constituting the th node, is the th desensitized label constituting the a desensitization label before update, a first updated desensitization label of the first node, a first updated desensitization label of the first node, a desensitization label before update, a similarity recognition index before and after label update, a user feedback coefficient, a user feedback feature value of the first node, a user feedback feature value of the first node, a competitor influence coefficient, a competitor influence value of the first node.

[0025] The intent analysis module 13 performs intent analysis based on the four-dimensional dynamic network and the feature vector to generate a final intent label; Further, the intent analysis module 13 includes the following sub-modules: The dual-channel intent analysis sub-module 131 obtains multi-hop associated intent and real-time important features through a dual-channel intent analysis based on the four-dimensional dynamic network and the extracted feature vector; Specifically, the dual-channel intent analysis includes a graph neural network channel and a vector direct reasoning channel. The multi-hop associated intent is captured through the graph neural network channel based on the four-dimensional dynamic network, and the real-time important features are obtained through the vector direct reasoning channel based on the extracted feature vector.

[0026] The final intent label generation sub-module 132 generates a final intent label through a contradiction fusion algorithm based on the multi-hop associated intent and the real-time important features; Specifically, the final intent label is obtained through a contradiction fusion algorithm based on the network associated intent and the real-time important features wherein, the final intent label, the adaptive fusion coefficient of the real-time important features, the de-dimensioned multi-hop associated intent vector, the de-dimensioned real-time important feature vector, the contradiction balancing factor.

[0027] The multi-modal promotion material acquisition module 14 collects competitor data and generates multi-modal promotion materials in combination with the final intent label; Further, the multi-modal promotion material acquisition module 14 includes the following sub-modules: The competitor data acquisition sub-module 141 crawls competitor public data through a distributed agent node simulating real behavior and filters false information through an adversarial denoising network; ​Specifically, a distributed agent node is deployed to dynamically generate a click path through reinforcement learning technology, simulate real user behavior to crawl public data of a competitor, and identify and filter false information in the public data of the competitor through a denoising filter network to ensure the authenticity of the collected data.

[0028] The promotion material generation submodule 142 generates multi-modal promotion materials based on the final intent label and the public data of the competitor through an "attribute-emotion-visual" joint mapper. Specifically, the advantage points of the food are extracted from the attribute comparison matrix according to the final intent label and the public data of the competitor, the emotion guide is generated by mining the emotion word library from the local food library and calibrating the emotion intensity, and the multi-modal promotion material is generated according to the advantage points of the food and the emotion guide.

[0029] The material pushing module 15 constructs a promotion decision matrix and adaptively pushes the multi-modal promotion material to the user. Further, the material pushing module 15 includes the following submodules: The promotion decision matrix construction submodule 151 acquires a promotion adaptation factor and constructs a promotion decision matrix. Specifically, the promotion adaptation factor includes but is not limited to a user state factor, a content adaptation factor, an environment adaptation factor, and a competition intensity factor, and the promotion decision matrix is constructed based on the promotion adaptation factor and the real-time updated factor weight.

[0030] The adaptive material pushing submodule 152 adaptively pushes the multi-modal promotion material to the user based on the promotion decision matrix. Specifically, the adaptive pushing matching algorithm is used to calculate the real-time pushable value of the multi-modal promotion material based on the promotion decision matrix. The real-time pushable value of the multi-modal promotion material is calculated, wherein, is the pushable value of the multi-modal promotion material, is the number of promotion adaptation factors, is the first is the push matching coefficient of each promotion adaptation factor, is the first promotion adaptation factor, is the material adaptation influence index, is the amount of material data in the multi-modal promotion material, is the first material data adaptation influence weight, is the first material data push matching value. When the pushable value of the multi-modal promotion material reaches the preset pushable threshold, the adaptive selection matching push window is selected to push the multi-modal promotion material to the user.

[0031] Network and matrix optimization module 16, acquires user interaction data, extracts feedback features, and optimizes the four-dimensional dynamic network and promotion decision matrix; Furthermore, the network and matrix optimization module 16 includes the following submodules: The interaction data acquisition submodule 161 acquires the interaction data between the user and the multimodal promotional material through a three-level authorization mechanism; Specifically, user authorization is obtained through a three-level authorization mechanism to extract user interaction data, negative feedback data, cross-model correlation data, and other interaction data between users and multimodal promotional materials.

[0032] Optimization submodule 162 extracts feedback features based on interaction data and optimizes the four-dimensional dynamic network and promotion decision matrix; Specifically, based on the interaction data, the feedback features of edge nodes are extracted through federated learning technology, the four-dimensional dynamic network and promotion decision matrix are optimized, and dual-module collaborative optimization is achieved.

[0033] Example 2 like Figure 2 As shown, the second embodiment of the present application provides a food sales website promotion method based on dynamic network intent analysis, which includes the following steps: Step S1: Collect multi-source user data through a three-level authorization mechanism, extract feature vectors on the local device, generate desensitization labels, and destroy the original data; Furthermore, collecting multi-source user data through a three-level authorization mechanism, extracting feature vectors on the local device, generating desensitizing labels, and destroying the original data include the following sub-steps: Step S11: Obtain user data layered authorization through a three-level authorization mechanism and collect multi-source user data; Step S12: Preprocess the collected multi-source user data using federated learning technology, extract feature vectors on the local device, generate desensitization labels, and trigger a random noise signal to destroy the original data; Step S2: construct a four-dimensional dynamic network based on the desensitized labels and dynamically update the node weights; Furthermore, constructing a four-dimensional dynamic network based on the desensitized labels and dynamically updating the node weights includes the following sub-steps: Step S21: constructing a four-dimensional dynamic network of "user-scenario-food attribute-intention" based on the desensitized labels; Step S22: Dynamically update the node weights of the four-dimensional dynamic network using a multi-dimensional weight update algorithm; Step S3: perform intent analysis based on the four-dimensional dynamic network and feature vector to generate the final intent label; Furthermore, intent parsing based on the four-dimensional dynamic network and feature vectors to generate the final intent label includes the following sub-steps: Step S31, obtaining multi-hop associated intent and real-time important features through double-channel intent analysis based on the four-dimensional dynamic network and the extracted feature vector; Step S32, generating a final intent label through a contradiction fusion algorithm based on the multi-hop associated intent and the real-time important features; Step S4, collecting competitive product data, and generating multi-modal promotion materials in combination with the final intent label; Further, collecting competitive product data, and generating multi-modal promotion materials in combination with the final intent label includes the following sub-steps: Step S41, simulating real behavior to crawl competitive public data through a distributed agent node, and filtering false information through an adversarial denoising network; Step S42, generating multi-modal promotion materials through an "attribute-sentiment-visual" joint mapper based on the final intent label and the competitive public data; Step S5, constructing a promotion decision matrix, and adaptively pushing multi-modal promotion materials to users; Further, constructing a promotion decision matrix, and adaptively pushing multi-modal promotion materials to users includes the following sub-steps: Step S51, obtaining a promotion adaptation factor, and constructing a promotion decision matrix; Step S52, adaptively pushing multi-modal promotion materials to users based on the promotion decision matrix; Step S6, obtaining user interaction data, extracting feedback features, and optimizing the four-dimensional dynamic network and the promotion decision matrix; Further, obtaining user interaction data, extracting feedback features, and optimizing the four-dimensional dynamic network and the promotion decision matrix includes the following sub-steps: Step S61, obtaining interaction data of users and multi-modal promotion materials through a three-level authorization mechanism; Step S62, extracting feedback features based on the interaction data, and optimizing the four-dimensional dynamic network and the promotion decision matrix; Corresponding to the above-mentioned embodiments, the embodiments of the present application provide a computer storage medium, comprising: at least one memory and at least one processor; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute the food sales website promotion method based on dynamic network intent analysis.

[0034] Corresponding to the above-mentioned embodiments, the embodiments of the present application provide a computer readable storage medium, and the computer storage medium contains one or more program instructions, which are used for the food sales website promotion method based on dynamic network intent analysis by the processor.

[0035] The embodiment disclosed by the application provides a computer readable storage medium, wherein computer program instructions are stored in the computer readable storage medium, and when the computer program instructions run on a computer, the computer program instructions make the computer execute the food sales website promotion method based on dynamic network intention analysis.

[0036] In the embodiment of the application, the processor can be an integrated circuit chip with processing capability of signals. The processor can be a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0037] The disclosed methods, steps and logic block diagrams in the embodiments of the application can be implemented or executed. The general purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the application can be directly embodied as hardware code processor execution or executed by a combination of hardware and software modules in the code processor. The software module can be located in a random memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The processor reads the information in the storage medium and combines the hardware to complete the steps of the above method.

[0038] The storage medium can be a memory, for example, can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories.

[0039] Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory.

[0040] The volatile memory can be Random Access Memory (RAM), used as external cache memory. By way of example, and not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The below-described embodiments do not imply that a given embodiment is necessary or necessary to practice the application in its application.

[0041] The storage media described in the embodiments of the present application is intended to include, but not be limited to, these and any other suitable types of memory.

[0042] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the present application can be implemented in combination with hardware and software. When the software is applied, the corresponding functions can be stored in a computer readable medium or transmitted as one or more instructions or codes on a computer readable medium. The computer readable medium includes computer storage medium and communication medium, wherein the communication medium includes any medium facilitating the transmission of computer programs from one place to another. The storage medium can be any available medium accessible by a general-purpose or special-purpose computer.

[0043] The above detailed description of the specific implementation of the present application further describes the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific implementation of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made on the basis of the technical solutions of the present application should be included in the protection scope of the present application.

Claims

1. A food sales website promotion platform based on dynamic network intent analysis, characterized by: include: The multi-source user data collection and processing module collects multi-source user data through a three-level authorization mechanism, extracts feature vectors on the local device, generates desensitizing labels, and destroys the original data; A four-dimensional dynamic network construction and update module builds a four-dimensional dynamic network based on desensitized labels and dynamically updates node weights; The intent parsing module performs intent parsing based on a four-dimensional dynamic network and feature vectors to generate the final intent label; The multimodal promotion material acquisition module collects competitor data and generates multimodal promotion materials based on the final intent tag. The material push module builds a promotion decision matrix and adaptively pushes multi-modal promotional materials to users; The network and matrix optimization module obtains user interaction data, extracts feedback features, and optimizes the four-dimensional dynamic network and promotion decision matrix.

2. The food sales website promotion platform based on dynamic network intent analysis according to claim 1, characterized in that: Multi-source user data collection and processing module, specifically including: The multi-source user data collection submodule obtains user data layer authorization through a three-level authorization mechanism and collects multi-source user data; The vector extraction and desensitization label generation sub-module pre-processes the collected multi-source user data through federated learning technology, extracts feature vectors on the local device, generates desensitization labels, and triggers random noise signals to destroy the original data.

3. The food sales website promotion platform based on dynamic network intent analysis according to claim 1, characterized in that: The four-dimensional dynamic network construction and update module includes: The four-dimensional dynamic network construction submodule builds a four-dimensional dynamic network of "user-scenario-food attribute-intention" based on desensitized labels; The node weight update submodule dynamically updates the node weights of the four-dimensional dynamic network through a multi-dimensional weight update algorithm.

4. The food sales website promotion platform based on dynamic network intent analysis according to claim 1, characterized in that: Intent parsing module, specifically including: The dual-channel intent parsing submodule obtains multi-hop associated intent and real-time important features through dual-channel intent parsing based on a four-dimensional dynamic network and extracted feature vectors; The final intent label generation submodule generates the final intent label based on the multi-hop associated intent and real-time important features through the contradiction fusion algorithm.

5. The food sales website promotion platform based on dynamic network intent analysis as claimed in claim 1, characterized in that: The material push module specifically includes: The promotion decision matrix construction submodule obtains the promotion adaptation factor and constructs the promotion decision matrix; The adaptive material push sub-module adaptively pushes multimodal promotional materials to users based on the promotion decision matrix.

6. A food sales website promotion method based on dynamic network intention analysis, characterized in that: include: Step S1: Collect multi-source user data through a three-level authorization mechanism, extract feature vectors on the local device, generate desensitization labels, and destroy the original data; Step S2: construct a four-dimensional dynamic network based on the desensitized labels and dynamically update the node weights; Step S3: perform intent analysis based on the four-dimensional dynamic network and feature vector to generate the final intent label; Step S4: Collect competitor data and generate multimodal promotional materials based on the final intent tag; Step S5: Construct a promotion decision matrix and adaptively push multimodal promotional materials to users; Step S6: Obtain user interaction data, extract feedback features, and optimize the four-dimensional dynamic network and promotion decision matrix.

7. The method for promoting a food sales website based on dynamic network intent analysis according to claim 6, characterized in that: Furthermore, collecting multi-source user data through a three-level authorization mechanism, extracting feature vectors on the local device, generating desensitizing labels, and destroying the original data include the following sub-steps: Step S11: Obtain user data layered authorization through a three-level authorization mechanism and collect multi-source user data; Step S12: Preprocess the collected multi-source user data through federated learning technology, extract feature vectors on the local device, generate desensitization labels, and trigger random noise signals to destroy the original data.

8. The method for promoting a food sales website based on dynamic network intent analysis according to claim 6, wherein: Furthermore, constructing a four-dimensional dynamic network based on the desensitized labels and dynamically updating the node weights includes the following sub-steps: Step S21: construct a four-dimensional dynamic network of "user-scenario-food attribute-intention" based on the desensitized labels; Step S22: Dynamically update the node weights of the four-dimensional dynamic network through a multi-dimensional weight update algorithm.

9. The method for promoting a food sales website based on dynamic network intent analysis according to claim 6, wherein: Furthermore, intent parsing based on the four-dimensional dynamic network and feature vectors to generate the final intent label includes the following sub-steps: Step S31: Obtain multi-hop associated intent and real-time important features through dual-channel intent parsing based on the four-dimensional dynamic network and the extracted feature vectors; Step S32: Generate a final intent label based on the multi-hop association intent and real-time important features through a contradiction fusion algorithm.

10. The method for promoting a food sales website based on dynamic network intent analysis according to claim 6, wherein: Furthermore, constructing a promotion decision matrix to adaptively push multimodal promotional materials to users includes the following sub-steps: Step S51: Obtain promotion adaptation factors and construct a promotion decision matrix; Step S52: Based on the promotion decision matrix, adaptively push multimodal promotion materials to the user.

Images