Intelligent vending machine marketing system based on large model

Abstract

The invention discloses an intelligent vending machine marketing system based on a large model, which successfully solves a plurality of problems mentioned in the background technology through the technical means of multi-modal interaction, dynamic commodity recommendation, emotional marketing strategy, real-time payment and inventory synchronization and the like. Firstly, by adopting the technologies of voice recognition, voiceprint recognition, natural language understanding and the like, the system breaks through the problem that a traditional vending machine is single in interaction, a user can conduct multi-round dialogue with the system through voice, commodity selection is rapidly completed, identity verification and personalized service are supported, and the interaction experience of the user is remarkably improved. Secondly, a commodity recommendation module of the system combines a user portrait, real-time inventory information and an emotional state, and adopts a mixed recommendation model and real-time inventory linkage, so that recommendation logic is more intelligent and dynamic, and the problem of static commodity recommendation in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent marketing of vending machines based on large models, and in particular to an intelligent marketing system for vending machines based on large models. Background Art

[0002] With the rapid development of artificial intelligence and the Internet of Things (IoT), vending machines have gradually become a part of people's daily lives as a convenient self-service shopping method. However, existing vending machines still have many technical deficiencies in terms of user interaction, product recommendations, and marketing strategies, making them unable to meet the demands of modern consumers for intelligent, personalized, and efficient interactions.

[0003] Currently, there are several obvious problems with the technical implementation of vending machines: First, traditional vending machines usually rely on physical buttons or touch screens to select and pay for goods. This single interaction method is cumbersome, inefficient, and lacks sufficient user-friendliness; second, the recommendation systems of most vending machines are based on fixed rules and lack dynamic adaptation to user interests and emotional changes. Therefore, the recommended goods are often not personalized and timely; third, the marketing strategies of vending machines are mostly inflexible and cannot be adjusted in real time according to changes in the external environment (such as weather, time, etc.), resulting in poor marketing effects and user experience; in addition, the payment process is relatively complicated and usually requires separate payment each time, which increases the cumbersomeness of the operation; finally, traditional vending machines cannot be optimized based on the user's voice emotions or identity characteristics, resulting in a lack of emotional and personalized interaction.

[0004] Therefore, existing technologies have not effectively addressed issues such as how to accurately understand and dynamically respond to user behavior through multimodal interaction and large-scale modeling, how to optimize recommendation strategies based on user emotions and external environmental changes, and how to enhance user experience through voice and sentiment analysis. These issues urgently need to be addressed. Summary of the Invention

[0005] This application provides a large-scale model-based intelligent vending machine marketing system, which aims to solve the problem of relying on physical buttons or touch screens to select and pay for goods in traditional vending machines. This single interaction method is cumbersome and inefficient, and lacks sufficient user-friendliness.

[0006] A large-scale model-based intelligent vending machine marketing system, comprising:

[0007] Multimodal interaction module, which realizes user voice input, identity recognition and intent recognition through speech recognition, voiceprint recognition and natural language understanding technologies;

[0008] The product recommendation module dynamically generates personalized product recommendation lists based on user portraits, real-time inventory data, and user emotional status;

[0009] The marketing strategy generation module combines external environment data to generate personalized promotion strategies in real time and outputs matching promotional slogans;

[0010] The payment module supports one-time payment for batches of goods, and uses aggregated QR code generation and payment API to enable users to pay for multiple goods;

[0011] The data synchronization module synchronizes inventory data in real time through the WebSocket protocol and ensures that inventory status is synchronized with user information.

[0012] In the above solution, optionally, the multimodal interaction module includes:

[0013] The speech recognition submodule uses a Transformer-based speech recognition model to convert user audio signals into text;

[0014] The voiceprint recognition submodule uses the DeepSpeaker model to extract the user's voiceprint features for identity verification, avoiding the influence of interfering sounds;

[0015] The natural language understanding submodule uses the BERT model to perform intent recognition and slot filling for user input voice or text.

[0016] In the above solution, optionally, the product recommendation module further includes:

[0017] The recommendation algorithm submodule uses a hybrid recommendation model, including the Wide & Deep model and collaborative filtering algorithm, to dynamically generate product recommendation scores based on user behavior data and product features;

[0018] The inventory linkage sub-module adjusts the recommendation weight of the product based on real-time inventory data. When the inventory level is lower than the set threshold, the recommendation weight of the product will automatically decrease.

[0019] In the above solution, optionally, the marketing strategy generation module includes:

[0020] The external data adapter module dynamically adjusts recommended products and marketing strategies based on real-time weather, time, and holiday information by accessing external data sources such as weather APIs and holiday calendars;

[0021] The sales pitch generation sub-module uses a large model to generate personalized, scenario-based promotional sales pitches, and combines them with emotional voice output to improve user experience.

[0022] In the above solution, optionally, the payment module includes:

[0023] The payment interface submodule uses the Alipay / WeChat payment API to generate aggregated QR codes, supporting one-time payment of multiple products;

[0024] The payment synchronization submodule updates inventory data in real time through the WebSocket protocol after successful payment to ensure that the inventory status is consistent with the actual goods.

[0025] In the above solution, optionally, the data synchronization module synchronizes the payment result with the inventory data in real time through the WebSocket protocol to ensure that the inventory information is updated in a timely manner.

[0026] In the above solution, optionally, the multimodal interaction module further includes:

[0027] The semantic sentiment analysis sub-module analyzes the user's emotional state through the RoBERTa model and dynamically adjusts the recommendation logic and promotional language based on the user's emotions.

[0028] In the above solution, optionally, the product recommendation module further includes:

[0029] The emotion adaptation recommendation submodule combines the results of semantic sentiment analysis to automatically adjust recommended products under different emotional states, giving priority to recommending products that match the emotional state.

[0030] In the above solution, optionally, the marketing strategy generation module further includes:

[0031] The emotional voice output sub-module uses speech synthesis technology to generate emotionally charged promotional slogans. It adjusts the voice output parameters such as timbre and speaking speed according to factors such as the user's emotional state, weather and time, thereby improving the user experience.

[0032] In the above solution, optionally, the product recommendation module further includes:

[0033] The user behavior analysis submodule divides users into groups through the K-means clustering algorithm, and optimizes recommendation strategies and replenishment forecasts based on data such as user purchase frequency, category preferences, and time period distribution.

[0034] Compared with the prior art, this application has at least the following beneficial effects:

[0035] This application is based on further analysis and research of existing technical problems, and recognizes that in the use of traditional vending machines, the interaction method is single, the product recommendation accuracy is insufficient, and the marketing strategy lacks flexibility. Through technical means such as multimodal interaction, dynamic product recommendation, emotional marketing strategy, and real-time payment and inventory synchronization, it successfully solves many of the problems mentioned in the background technology. First, by adopting technologies such as voice recognition, voiceprint recognition, and natural language understanding, the system breaks through the problem of single interaction of traditional vending machines. Users can conduct multiple rounds of dialogue with the system through voice and quickly complete product selection. It also supports identity authentication and personalized services, which significantly improves the user's interactive experience. Secondly, the product recommendation module of the system combines user portraits, real-time inventory information and emotional status, and adopts a hybrid recommendation model and real-time inventory linkage, making the recommendation logic more intelligent and dynamic, solving the problem of static product recommendations in the existing technology. Each recommendation of the user is adjusted in real time according to his or her historical behavior, emotional changes, and inventory status, ensuring the personalization and real-time nature of the recommendation. In terms of marketing strategies, by introducing external data sources (such as weather, time periods, and holidays) and promotional scripts generated by large models, the system can generate adaptive marketing strategies based on real-time environmental changes, and interact with users through emotional voice output, thus avoiding the disadvantage of traditional vending machine marketing strategies being out of touch with actual scenarios. Finally, the payment module simplifies the cumbersome payment process of traditional vending machines and improves payment efficiency by supporting one-time payments for bulk goods. At the same time, it synchronizes payment information and inventory data in real time through the WebSocket protocol, ensuring the accuracy and timely updating of inventory. Overall, the present invention not only greatly improves the user experience, but also improves the operational efficiency of vending machines by dynamically adjusting recommendations and marketing strategies, solving key problems in the background technology such as single interaction, static recommendations, disconnected marketing, and redundant payment processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of a module of a vending machine intelligent marketing system based on a large model provided in one embodiment of the present application;

[0037] Figure 2 A schematic diagram of a usage scenario of a large-model-based vending machine intelligent marketing method provided in one embodiment of the present application;

[0038] Figure 3 A schematic diagram of the method architecture of a large-model-based vending machine intelligent marketing method provided in one embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0040] In one embodiment, Figure 1 、 Figure 2 and Figure 3 As shown, a smart vending machine marketing system based on a large model is provided, the system comprising:

[0041] Multimodal interaction module, which realizes user voice input, identity recognition and intent recognition through speech recognition, voiceprint recognition and natural language understanding technologies;

[0042] The product recommendation module dynamically generates personalized product recommendation lists based on user portraits, real-time inventory data, and user emotional status;

[0043] The marketing strategy generation module combines external environment data to generate personalized promotion strategies in real time and outputs matching promotional slogans;

[0044] The payment module supports one-time payment for batches of goods, and uses aggregated QR code generation and payment API to enable users to pay for multiple goods;

[0045] The data synchronization module synchronizes inventory data in real time through the WebSocket protocol and ensures that inventory status is synchronized with user information.

[0046] The large-scale model-based smart vending machine marketing system described in this embodiment consists of multiple core modules, primarily a multimodal interaction module, a product recommendation module, a marketing strategy generation module, a payment module, and a data synchronization module. These modules, through their organic integration and collaborative operation, enable efficient and intelligent user interaction, product recommendations, marketing strategy optimization, and payment process optimization, ultimately improving the user experience and operational efficiency of the smart vending machine.

[0047] In this system, the multimodal interaction module simplifies the user operation path and enables intelligent voice conversations by adopting advanced speech recognition, voiceprint recognition, and natural language understanding (NLU) technologies. The implementation process of this module is as follows:

[0048] Speech Recognition Submodule: This submodule uses a Transformer-based speech recognition model, such as the Conformer model. Users input product requirements or instructions via voice. The audio signal is extracted using Mel-spectrogram features and then converted to text using a self-attention mechanism. This process effectively improves recognition accuracy and speed, reducing user wait time.

[0049] Voiceprint Recognition Submodule: To identify users and eliminate background noise, the voiceprint recognition submodule uses the DeepSpeaker model. It extracts the user's voiceprint characteristics and matches them with the characteristics of registered users in the database to confirm the user's identity. This technology helps enhance the system's personalized experience and prevent misoperation.

[0050] Natural Language Understanding Submodule: This submodule uses the BERT model to identify user intent and fill in slots based on voice or text input. For example, if a user inputs "I want a low-sugar drink" via voice, the system can identify the user's purchase intent as "buy a drink" and extract the slot information "low sugar."

[0051] The product recommendation module combines user portraits, real-time inventory information, and user emotional state, and uses advanced recommendation algorithms to generate personalized recommendation lists. The specific implementation method is as follows:

[0052] Hybrid Recommendation Model: The system utilizes a hybrid model that combines the Wide & Deep model with a collaborative filtering algorithm. The Wide & Deep model generates a basic recommendation based on user and product characteristics, while the collaborative filtering algorithm optimizes recommendations based on user history and product similarities. Dynamically adjusting recommendation scores ensures that recommendations align with user interests and needs.

[0053] Real-time inventory linkage: The recommendation module monitors inventory levels in real time and dynamically adjusts the weight of recommended products. When the inventory level of a product falls below a preset threshold, the recommendation weight of the product will be automatically reduced, ensuring that the recommended product is always available for purchase, thereby improving the user experience.

[0054] The marketing strategy generation module of this system can automatically generate real-time promotion strategies based on external data sources (such as weather, time periods, holidays, etc.), and generate relevant promotional words. The specific implementation method is as follows:

[0055] External Data Adaptation: This module accesses external data sources like weather APIs and holiday calendars to obtain real-time weather, date, and time information, dynamically adjusting marketing strategies based on this information. For example, the system might recommend hot drinks based on rainy weather, with promotional messages like "Warm your stomach on rainy days, hot coffee half price."

[0056] Large-scale model-generated promotional pitches: We use large models like GPT-3.5 to generate personalized promotional pitches based on real-time contextual data. These pitches incorporate user sentiment, time of day, weather, and other factors to ensure the most compelling promotional information is delivered to users at the right time.

[0057] The payment module can support one-time payment for batches of goods, improving the convenience of shopping for users. The specific implementation method is as follows:

[0058] Payment interface: This module generates an aggregated QR code by integrating payment APIs such as Alipay and WeChat Pay. Users can pay for multiple items at once by scanning the QR code, avoiding the tedious operation of paying for each item.

[0059] Payment synchronization function: After the payment is completed, the system synchronizes the payment information to the inventory management system in real time through the WebSocket protocol to ensure real-time update of inventory data.

[0060] The data synchronization module uses the WebSocket protocol to synchronize with the inventory management system and user behavior analysis module in real time to ensure the accuracy and timeliness of all data. The specific implementation is as follows:

[0061] Inventory synchronization: When a user makes a payment or an item is shipped, the system immediately updates inventory data to ensure that the inventory status is consistent with the actual situation. Real-time synchronization via the WebSocket protocol effectively avoids stock shortages or overstocking caused by delayed inventory information.

[0062] The intelligent vending machine marketing system of the present invention can effectively solve several major problems mentioned in the background technology through the deep integration of various modules:

[0063] Solving the problem of single interaction: Through multimodal interaction technology, the system supports voice recognition, voiceprint recognition, and natural language understanding, allowing users to interact with the system through voice. Users can not only quickly complete product selection, but also conduct multiple rounds of dialogue within the system, receiving more intelligent and personalized services, significantly improving the interactive experience.

[0064] Solving the problem of static recommendations: Traditional vending machine recommendation systems are often based on fixed rules and cannot dynamically adapt to changing user needs. By introducing a hybrid recommendation model and real-time inventory linkage, the recommendation module of this invention can dynamically adjust the recommendation list based on the user's interests, historical behavior, and emotional state. This allows each product recommendation to be optimized according to the user's actual needs, avoiding the limitations of static recommendations.

[0065] Solve the disconnect between marketing and user scenarios: Traditional vending machine marketing strategies are often preset and fixed, unable to adapt to external environmental changes (such as weather and holidays). However, the marketing strategy generation module of this invention can generate promotional strategies based on external data in real time and, combined with a large model, generate personalized promotional messages. This ensures that marketing strategies are closely integrated with user scenarios, effectively improving marketing effectiveness and user purchasing desire.

[0066] This solution addresses payment process redundancy by supporting one-time payment for bulk purchases, eliminating the need for users to pay for each item individually. This reduces operational complexity and improves payment efficiency. Furthermore, real-time synchronization of payment information and inventory status via the WebSocket protocol ensures timely updates of payments and inventory, avoiding user experience issues caused by payment delays or insufficient inventory.

[0067] Through the above technical solutions, the present invention effectively solves a series of problems in the prior art and improves the user experience and operational efficiency of vending machines through intelligent recommendation and marketing strategies.

[0068] In this embodiment, the multimodal interaction module includes:

[0069] The speech recognition submodule uses a Transformer-based speech recognition model to convert user audio signals into text;

[0070] The voiceprint recognition submodule uses the DeepSpeaker model to extract the user's voiceprint features for identity verification, avoiding the influence of interfering sounds;

[0071] The natural language understanding submodule uses the BERT model to perform intent recognition and slot filling for user input voice or text.

[0072] In this embodiment, the product recommendation module further includes:

[0073] The recommendation algorithm submodule uses a hybrid recommendation model, including the Wide & Deep model and collaborative filtering algorithm, to dynamically generate product recommendation scores based on user behavior data and product features;

[0074] The inventory linkage sub-module adjusts the recommendation weight of the product based on real-time inventory data. When the inventory level is lower than the set threshold, the recommendation weight of the product will automatically decrease.

[0075] In this embodiment, the marketing strategy generation module includes:

[0076] The external data adapter module dynamically adjusts recommended products and marketing strategies based on real-time weather, time, and holiday information by accessing external data sources such as weather APIs and holiday calendars;

[0077] The sales pitch generation sub-module uses a large model to generate personalized, scenario-based promotional sales pitches, and combines them with emotional voice output to improve user experience.

[0078] In this embodiment, the payment module includes:

[0079] The payment interface submodule uses the Alipay / WeChat payment API to generate aggregated QR codes, supporting one-time payment of multiple products;

[0080] The payment synchronization submodule updates inventory data in real time through the WebSocket protocol after successful payment to ensure that the inventory status is consistent with the actual goods.

[0081] In this embodiment, the data synchronization module synchronizes the payment results with the inventory data in real time through the WebSocket protocol to ensure that the inventory information is updated in a timely manner.

[0082] In this embodiment, the multimodal interaction module further includes:

[0083] The semantic sentiment analysis sub-module analyzes the user's emotional state through the RoBERTa model and dynamically adjusts the recommendation logic and promotional language based on the user's emotions.

[0084] In this embodiment, the product recommendation module further includes:

[0085] The emotion adaptation recommendation submodule combines the results of semantic sentiment analysis to automatically adjust recommended products under different emotional states, giving priority to recommending products that match the emotional state.

[0086] In this embodiment, the marketing strategy generation module further includes:

[0087] The emotional voice output sub-module uses speech synthesis technology to generate emotionally charged promotional slogans. It adjusts the voice output parameters such as timbre and speaking speed according to factors such as the user's emotional state, weather and time, thereby improving the user experience.

[0088] In this embodiment, the product recommendation module further includes:

[0089] The user behavior analysis submodule divides users into groups through the K-means clustering algorithm, and optimizes recommendation strategies and replenishment forecasts based on data such as user purchase frequency, category preferences, and time period distribution.

[0090] The aforementioned large-scale model-based vending machine intelligent marketing system incorporates AI large-scale models and natural language processing technology to enable multiple rounds of user interaction with the vending machine through voice, quickly understanding user needs and lowering the barrier to entry for traditional interaction methods, allowing users to complete product selection and purchase through natural conversation. Using small industry models to analyze users' voice needs, shopping preferences, and historical behavior data, combined with the vending machine's real-time inventory dynamics, it provides more accurate and personalized product recommendations, improving product matching efficiency and increasing user purchase conversion rates. Leveraging the combined power of large and small models, it dynamically generates promotional or marketing strategies based on different scenarios (such as time, holidays, and weather). By delivering contextualized prompts or customized recommendations to users, product sales performance is improved. It supports users with multiple product recommendations and payments at once, streamlining the entire process from product selection to purchase and significantly improving user shopping efficiency. The backend management system collects real-time user behavior and sales data, enabling the operations team to fine-tune product inventory, recommendation strategies, and marketing campaigns.

[0091] In this embodiment, the AI large model interaction realizes multi-round dialogue interaction through voice wake-up and natural language understanding technology to accurately analyze user needs.

[0092] Industry small model recommendation: Combining the semantic parsing results provided by the large model and the vertical field optimization capabilities of the industry small model, it screens products that meet user needs from the current inventory and dynamically generates a recommendation list.

[0093] Intelligent sales control: Through the linkage of payment integration, inventory management and shipping systems, a complete closed loop from product recommendation to purchase is achieved.

[0094] Users can wake up the vending machine AI system through voice (such as saying "'Xiao Qu Xiao Qu'"), and the system will enter interactive mode to support users in making shopping requests.

[0095] Based on the keywords entered by the user (such as "low-sugar drinks"), eligible products are filtered from the current inventory in the machine, and up to three are displayed for the user to choose from.

[0096] After the user selects the product, the system generates a payment QR code. After the payment is completed, the product is automatically shipped and the inventory status is synchronized.

[0097] Humanistic care and marketing tips: During the recommendation process, the system dynamically displays promotional information or humanistic care tips (such as "The weather is dry, drink more water") based on time, festivals or user data.

[0098] This embodiment significantly improves the performance of traditional vending machines by introducing an intelligent marketing method based on a large model and a small industry model, effectively solving multiple problems mentioned in the background technology. First, the interaction method of traditional vending machines is single and the operation is cumbersome. Through the voice wake-up and multi-round voice interaction functions of the present invention, users can have intelligent conversations with the vending machine through natural language, greatly simplifying the operation process and improving the interactive experience. Users no longer rely on buttons or touch screens, but can complete product selection and payment through simple voice commands, lowering the operation threshold and improving user convenience and shopping efficiency. Secondly, the personalized recommendation system of the present invention can dynamically adjust product recommendation content based on multi-dimensional data such as user historical behavior, shopping preferences, real-time inventory, and external environment. Compared with the fixed product display method of traditional vending machines, the present invention can accurately match user needs, ensuring that each recommendation meets the user's personalized needs, improving product matching efficiency and purchase conversion rate, thereby solving the problem that traditional vending machines have single recommendation content and are difficult to meet personalized needs. In addition, the present invention introduces intelligent payment methods, including QR code payment, virtual currency payment, facial recognition payment and other payment methods. Users can complete payments quickly and easily, which greatly simplifies the payment process and solves the problem of lengthy payment processes in traditional vending machines. By linking with the inventory management system, the system can update inventory information in real time, avoid inventory errors or out-of-stock situations, and ensure the accurate delivery of goods. Furthermore, the dynamic marketing strategy generation module of the present invention can intelligently adjust marketing strategies and push promotional information according to external factors such as time, weather, and festivals, thereby improving the flexibility and timeliness of marketing and solving the problem of limited marketing capabilities and lack of flexible adjustment mechanisms in traditional vending machines. In general, the present invention comprehensively improves the user experience, operational efficiency and marketing capabilities of vending machines through intelligent interaction, accurate recommendations, simplified payment and efficient inventory management, solves many pain points in traditional technologies, and has significant innovation and practical application value.

[0099] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. An intelligent vending machine marketing system based on a large model, characterized by: The system comprises: Multimodal interaction module, which realizes user voice input, identity recognition and intent recognition through speech recognition, voiceprint recognition and natural language understanding technologies; The product recommendation module dynamically generates personalized product recommendation lists based on user portraits, real-time inventory data, and user emotional status; The marketing strategy generation module combines external environment data to generate personalized promotion strategies in real time and outputs matching promotional slogans; The payment module supports one-time payment for batches of goods, and uses aggregated QR code generation and payment API to enable users to pay for multiple goods; The data synchronization module synchronizes inventory data in real time through the WebSocket protocol and ensures that inventory status is synchronized with user information.

2. The intelligent vending machine marketing system according to claim 1, characterized in that: The multimodal interaction module includes: The speech recognition submodule uses a Transformer-based speech recognition model to convert user audio signals into text; The voiceprint recognition submodule uses the DeepSpeaker model to extract the user's voiceprint features for identity verification, avoiding the influence of interfering sounds; The natural language understanding submodule uses the BERT model to perform intent recognition and slot filling for user input voice or text.

3. The intelligent vending machine marketing system according to claim 1, characterized in that: The product recommendation module further includes: The recommendation algorithm submodule uses a hybrid recommendation model, including the Wide & Deep model and collaborative filtering algorithm, to dynamically generate product recommendation scores based on user behavior data and product features; The inventory linkage sub-module adjusts the recommendation weight of the product based on real-time inventory data. When the inventory level is lower than the set threshold, the recommendation weight of the product will automatically decrease.

4. The intelligent vending machine marketing system according to claim 1, characterized in that: The marketing strategy generation module includes: The external data adapter module dynamically adjusts recommended products and marketing strategies based on real-time weather, time, and holiday information by accessing external data sources such as weather APIs and holiday calendars; The sales pitch generation sub-module uses a large model to generate personalized, scenario-based promotional sales pitches, and combines them with emotional voice output to improve user experience.

5. The intelligent vending machine marketing system according to claim 1, characterized in that: The payment module includes: The payment interface submodule uses the Alipay / WeChat payment API to generate aggregated QR codes, supporting one-time payment of multiple products; The payment synchronization submodule updates inventory data in real time through the WebSocket protocol after successful payment to ensure that the inventory status is consistent with the actual goods.

6. The intelligent vending machine marketing system according to claim 1, characterized in that: The data synchronization module synchronizes payment results with inventory data in real time through the WebSocket protocol to ensure that inventory information is updated in a timely manner.

7. The intelligent vending machine marketing system according to claim 1, characterized in that: The multimodal interaction module further includes: The semantic sentiment analysis sub-module analyzes the user's emotional state through the RoBERTa model and dynamically adjusts the recommendation logic and promotional language based on the user's emotions.

8. The intelligent vending machine marketing system according to claim 1, characterized in that: The product recommendation module also includes: The emotion adaptation recommendation submodule combines the results of semantic sentiment analysis to automatically adjust recommended products under different emotional states, giving priority to recommending products that match the emotional state.

9. The intelligent vending machine marketing system according to claim 1, characterized in that: The marketing strategy generation module also includes: The emotional voice output sub-module uses speech synthesis technology to generate emotionally charged promotional slogans. It adjusts the voice output parameters such as timbre and speaking speed according to factors such as the user's emotional state, weather and time, thereby improving the user experience.

10. The intelligent vending machine marketing system according to claim 1, characterized in that: The product recommendation module also includes: The user behavior analysis submodule divides users into groups through the K-means clustering algorithm, and optimizes recommendation strategies and replenishment forecasts based on data such as user purchase frequency, category preferences, and time period distribution.

Images