Map shopping method based on AI voice interaction

Abstract

The invention relates to the technical field of shopping, and discloses a map shopping method based on AI voice interaction. Through deep fusion of AI voice interaction and a map visualization technology, the user experience and the operation efficiency of a geographic shopping scene are remarkably improved. According to the method, a real-time association system of the map POI and the commodity inventory is constructed, multi-modal interaction of voice instructions, dragging operation and cross-shop settlement is supported, and intelligent complementation of fuzzy instructions and dynamic distribution route optimization are realized by adopting BERT semantic understanding and a Q-learning path planning algorithm, so that the order response speed is improved, and the distribution distance is optimal; voiceprint DTW verification and Redis distributed inventory management are integrated, and real-time inventory scheduling in a 3-kilometer business district is realized while payment security is guaranteed; full-link state visual management is established through SSE push and WebSocket positioning technologies, and closed-loop experience of'search-purchase-payment-tracking 'is formed in combination with a thermodynamic layer recommendation system, so that operation is convenient.

Description

Technical Field

[0001] The present invention relates to the technical field of shopping, and specifically provides a map shopping method based on AI voice interaction. Background Art

[0002] With the rapid development of mobile Internet and artificial intelligence technologies, the deep integration of online shopping and offline scenarios has become a trend. Traditional e-commerce platforms mostly rely on text search or category navigation to complete the shopping process, lacking intelligent response to the user's geographical location and real-time scenarios, resulting in low shopping efficiency. Although existing map applications can display information about surrounding shops, they cannot directly complete the one-stop operations of product selection, order placement, and delivery. Users need to jump to other platforms to complete transactions, leading to a fragmented experience.

[0003] In the field of voice interaction, existing technologies mostly focus on smart speakers or simple command control scenarios (such as playing music, querying the weather), and have not been deeply integrated with map navigation, real-time inventory management, and multi-modal interaction. In addition, the traditional shopping cart function is limited to a single platform or store, and cannot support cross-store settlement, dynamic inventory synchronization, and location-based freight calculation, resulting in cumbersome operations for users when shopping across multiple channels.

[0004] The deficiencies of the existing technologies are mainly reflected in:

[0005] Fragmentation between map and shopping scenarios: Map applications only provide location display, lacking a full-process closed loop for product browsing, order placement, and delivery.

[0006] Single voice interaction function: Voice technology is mostly used for simple commands and has not been deeply integrated with complex scenarios such as map operations (such as clicking on a POI), inventory query, and multi-round conversations.

[0007] Insufficient multi-modal interaction experience: Users cannot flexibly switch operations through multiple methods such as voice, map clicking, and manual input, and lack real-time order status push.

[0008] Therefore, there is an urgent need for a technical solution that integrates AI voice interaction, dynamic map rendering, multi-store inventory synchronization, and intelligent delivery scheduling to improve user shopping efficiency and experience. Summary of the Invention

[0009] In view of the deficiencies of the existing technologies, the present invention provides a map shopping method based on AI voice interaction to solve the problems in the background art.

[0010] To achieve the above object, the present invention provides the following technical solution: A map shopping method based on AI voice interaction, comprising the following steps:

[0011] Step 1: Render a visual map of the user's current location or a specified area through the Map API, mark the POI information of surrounding stores, where the POI information is associated with the product inventory, price, and geographical coordinates in the product database, and support map zooming and dragging interactions;

[0012] Step 2: The voice interaction module receives the user's voice command and performs the following operations:

[0013] Convert the voice signal into text using end-to-end speech recognition technology, and parse the user's intention through a natural language processing model, including product name, quantity, delivery address, and time;

[0014] Combine with the context dialogue state management module to support multi-round interaction to correct the command;

[0015] Convert the system response into voice feedback through text-to-speech technology.

[0016] Step 3: Embed a dynamic shopping cart in the map interface, and support at least one of the following operations to trigger product addition or deletion:

[0017] Trigger product addition or deletion through voice commands;

[0018] Manually drag the map POI icon to the shopping cart;

[0019] After clicking on the POI, select the product to add to the shopping cart.

[0020] Step 4: When generating an order, calculate the shipping fee based on the delivery distance d between the product geographical coordinates and the user's address, where d is calculated through the Haversine algorithm, and complete the payment through a third-party payment interface.

[0021] Step 5: Call the path planning algorithm to generate the optimal delivery route, dynamically adjust the rider's position in combination with real-time road conditions, and map the delivery person's coordinates (x r , y r ) to the user's map in real time through WebSocket.

[0022] Step 6: After the payment is completed, send an order status update to the user side through Server-Sent Events (SSE), including packing, in-delivery, and signed-for notifications.

[0023] Preferably, the natural language processing model in Step 2 adopts the BERT architecture. When the user's command is ambiguous, calculate the semantic similarity to match the historical order data:

[0024]

[0025] If the similarity exceeds the threshold θ, automatically complete the missing information.

[0026] Preferably, in the third step, the shopping cart supports cross-store settlement, and real-time distributed inventory deduction is achieved through Redis. If the inventory of a certain commodity is insufficient, the system will automatically trigger a query for the inventory of the same commodity in other stores within 3 kilometers nearby.

[0027] Preferably, in the fifth step, the Q-learning algorithm is used to optimize the route for path planning, and the reward function is defined as:

[0028] r(s,a) = -(α · time cost + β · road congestion coefficient)

[0029] where α and β are dynamic weight parameters.

[0030] Preferably, the specific formula for calculating d in the haversine algorithm in the fourth step is expressed as:

[0031]

[0032] where R is the radius of the earth, φ is the latitude, and λ is the longitude.

[0033] Preferably, the payment process integrates voiceprint dynamic time warping (DTW) verification, and the matching degree between the voice feature sequence V = {V 1 , V 2 ,..., V n} input by the user and the registered voiceprint feature sequence V ′ = {V 1 ′ , V 2 ′ ,..., V m ′} is calculated, where:

[0034] V i represents the Mel-frequency cepstral coefficient (MFCC) feature vector of the i-th frame of the input voice;

[0035] V j ′ represents the MFCC feature vector of the j-th frame of the registered voiceprint;

[0036] The DTW distance calculation formula is:

[0037]

[0038] where π is the alignment path. If the distance is less than the preset threshold, payment is authorized.

[0039] Preferably, it also includes predicting the preference score of user u for commodity i using a matrix factorization model based on the user's historical behavior data The calculation formula is:

[0040]

[0041] u u is the potential feature vector of user u;

[0042] v i is the potential feature vector of product i;

[0043] b u is the bias term of user u;

[0044] b i is the bias term of product i.

[0045] The recommended results are rendered on the map interface through a heat map layer, highlighting the product-related POIs with scores associated.

[0046] Preferably, the user interface of this method uses the React Native framework to achieve multi-terminal adaptation and uniformly processes the following interaction logics:

[0047] Map rendering is accelerated through WebGL;

[0048] Voice commands are captured through the Web Speech API or native SDK;

[0049] Order status is synchronized through a cross-platform push channel.

[0050] Preferably, the data processing and business logic part of this method adopts a microservices architecture, which is split into independent map services, voice services, order services, and delivery services, and processes high-concurrency order requests through the Kafka message queue.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] Through the deep integration of AI voice interaction and map visualization technology, the present invention significantly improves the user experience and operation efficiency in the geographical shopping scenario. The present invention constructs a real-time association system between map POIs and product inventories, supports multi-modal interactions such as voice commands, drag operations, and cross-store settlement, adopts BERT semantic understanding and Q-learning path planning algorithms to achieve intelligent completion of fuzzy commands and optimization of dynamic delivery routes, improves the order response speed, and makes the delivery distance optimal; integrates voiceprint DTW verification and Redis distributed inventory management to achieve real-time inventory scheduling within a 3-kilometer business district while ensuring payment security; establishes full-link status visualization management through SSE push and WebSocket positioning technology, combined with a heat map layer recommendation system, to form a closed-loop experience of "search - purchase - payment - tracking" and optimize operations. This solution effectively solves the technical pain points of fragmented scenarios and single interactions in traditional e-commerce platforms.

[0053] Other features and advantages of the present invention will be described in the subsequent specification, and in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures pointed out in the specification, claims, and drawings. Brief Description of the Drawings

[0054] Figure 1 It is a flowchart of the map shopping method based on AI voice interaction of the present invention. Detailed Embodiments

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art in the technical field without creative efforts shall fall within the protection scope of the present invention.

[0056] Please refer to Figure 1 , a map shopping method based on AI voice interaction in the present invention, which is used to realize map shopping in the AI voice interaction mode, specifically includes the following content:

[0057] 1. System Architecture and Operating Environment

[0058] This embodiment is based on the mobile Internet platform. Users can realize the map shopping function through the application (App) installed on smart phones, tablets or car machines. The system architecture is divided into the following modules:

[0059] User Interaction Interface: Developed using the React Native framework, supporting unified interaction logic for iOS, Android, and Web terminals. The interaction logic processed includes:

[0060] Map rendering is accelerated through WebGL;

[0061] Voice commands are captured through the Web Speech API or native SDK;

[0062] Order status is synchronized through a cross-platform push channel.

[0063] Data Processing and Business Logic: Adopting a microservices architecture, including map services, voice services, order services, and delivery services.

[0064] AI Model Service: NLP models, recommendation models, and path planning models deployed in the cloud.

[0065] Infrastructure: Based on the AWS cloud platform, using Nginx load balancing, Kafka message queue, and Redis cache.

[0066] This method specifically includes the following steps:

[0067] Step 1:

[0068] 1. Map rendering and POI annotation

[0069] After the user starts the App, the system calls the API of AutoNavi Map or other App maps, and renders the map of the current area according to the GPS positioning (coordinates (x u , y u )) and displays the user's address (such as "Sanlitun, Chaoyang District, Beijing") through reverse geocoding.

[0070] Retrieve the POI information of shops within 5 kilometers around from the commodity database (MySQL) (including shop names, commodity lists, inventory S i , price P i and geographical coordinates (x i , y i )) and mark them as interactive icons on the map.

[0071] The POI click event triggers a commodity list pop-up window, and the user can manually select commodities to add to the shopping cart;

[0072] 2. Dynamic shopping cart implementation

[0073] The shopping cart interface floats in the upper right corner of the map, and the selected commodity list and total price are displayed in real time.

[0074] Total price calculation formula:

[0075]

[0076] Among them, F(d) is the freight, which is calculated using the haversine algorithm based on the delivery distance d, and the formula is expressed as:

[0077]

[0078] Among them, R = 6371 km (radius of the earth), is the latitude, λ is the longitude, Δ = 2 - 1 , Δλ = λ 2 - λ 1 .

[0079] Step 2:

[0080] 1. Voice interaction and command parsing

[0081] The user long-presses the voice button to input a command (such as "buy three cases of mineral water in this store"), and the voice signal V is converted into text T through an end-to-end ASR model:

[0082] T = ASR(V) (The ASR model is an LSTM network based on the CTC loss)

[0083] The text T is input into the BERT model for intent classification, and the user intent II and structured parameters are output:

[0084] I = argmax(f θ (T)) (The classification loss function is the cross-entropy loss)

[0085] If a fuzzy instruction is recognized (such as "deliver to the old place"), calculate the semantic similarity to match the historical address:

[0086]

[0087] When the similarity exceeds the threshold θ = 0.85, the address is automatically filled.

[0088] 2. Multi-turn dialogue management

[0089] Use a finite state machine (FSM) to track the dialogue state C t . For example:

[0090] User: "Buy two cases of mineral water" → System: "Please select the delivery time"

[0091] User: "Tomorrow at 5 pm" → The status is updated to:

[0092] C t+1 = {Commodity: Mineral water, Quantity: 2, Time: 17:00}

[0093] Step 3:

[0094] 1. Real-time inventory management

[0095] When the user adds a commodity, call the Redis atomic operation to reduce the inventory:

[0096] DECRBY stock:shop1:item123 2 # Reduce the inventory by 2 pieces.

[0097] If the inventory is insufficient (returns a negative value), trigger a query for nearby shops, and only retrieve shops within 3 kilometers of the user.

[0098] 2. Order generation and payment

[0099] The order data includes: user ID, commodity list, total price, delivery address (x u , y u ).

[0100] When making a payment, the user needs to read aloud a random string of numbers (such as "6583") for voiceprint verification:

[0101] Extract the MFCC features of the voice V = {V 1 , V 2 ,..., V n}, and calculate the DTW distance with the registered voiceprint V':

[0102]

[0103] If the distance is less than the threshold τ = 10τ = 10, authorize the payment.

[0104] Step Four:

[0105] 1. Rider scheduling algorithm

[0106] Adopt Q-learning to optimize the delivery route. The state s is the road network node, and the action a is to select the next road segment:

[0107]

[0108] The reward function r(s, a) is defined as:

[0109] r(s, a) = -(0.6·t(s, a) + 0.4·c(s, a))

[0110] where t(s, a) is the road segment time, and c(s, a) is the congestion coefficient (0 - 1).

[0111] Step Five:

[0112] Product recommendation system

[0113] Predict the preference score of user u for product i based on the matrix factorization model:

[0114]

[0115] where is the latent feature vector, and it is trained by minimizing the following loss function:

[0116]

[0117] The recommendation results are rendered through a heat map layer, and the product POIs with a score are highlighted.

[0118] This embodiment discloses a map shopping method based on AI voice interaction. By integrating technologies such as map rendering, speech recognition, real-time inventory management, path planning, and personalized recommendation, a full-process closed loop from product browsing, voice ordering to delivery tracking is realized. The system adopts a microservice architecture, combines Redis caching, Q-learning algorithm, and matrix factorization model, significantly improving the interaction real-time performance and user experience, and providing an efficient and intelligent solution for online and offline integrated shopping.

Claims

1. A map shopping method based on AI voice interaction, characterized in that: The following steps are involved: Step 1: Render a visual map of the user's current location or specified area through the map API, mark the POI information of surrounding shops, and associate the POI information with the product inventory, price and geographic coordinates in the product database, and support map zooming and dragging interaction; Step 2: The voice interaction module receives the user's voice command and performs the following operations: End-to-end speech recognition technology is used to convert speech signals into text, and the user's intention is analyzed through the natural language processing model, including product name, quantity, delivery address and time; Combined with the contextual dialogue state management module, it supports multiple rounds of interactive correction instructions; The system response is converted into speech feedback through speech synthesis technology. Step 3: Embed a dynamic shopping cart in the map interface, and support at least one of the following operations to trigger product addition or deletion: Voice commands trigger product addition or deletion; Manually drag the map POI icon to the shopping cart; Click on the POI and select the product to add to the shopping cart. Step 4: When an order is generated, the shipping fee is calculated based on the delivery distance d between the product's geographic coordinates and the user's address, where d is calculated using the Haversine algorithm, and payment is completed through a third-party payment interface. Step 5: Call the path planning algorithm to generate the optimal delivery route, dynamically adjust the rider's position based on real-time traffic conditions, and send the delivery person's coordinates (x r ,y r ) is mapped to the user-side map in real time. Step 6: After payment is completed, the order status update is sent to the user through the server push event (SSE), including packaging, delivery and receipt notifications.

2. A map shopping method based on AI voice interaction according to claim 1, characterized in that: The natural language processing model in step 2 adopts the BERT architecture. When the user instruction is ambiguous, the historical order data is matched by calculating the semantic similarity: If the similarity exceeds the threshold θ, the missing information is automatically completed.

3. A map shopping method based on AI voice interaction according to claim 1, characterized in that: In step three, the shopping cart supports cross-store settlement and implements real-time deduction of distributed inventory through Redis. If the inventory of a certain product is insufficient, it will automatically trigger inventory queries of the same product in other stores within 3 kilometers.

4. The map shopping method based on AI voice interaction according to claim 1, characterized in that: In step 5, the path planning uses the Q-learning algorithm to optimize the route, and the reward function is defined as: r(s,a)=-(α·time cost+β·road congestion coefficient) Among them, α and β are dynamic weight parameters.

5. The map shopping method based on AI voice interaction according to claim 1, characterized in that: The formula for calculating d using the haversine algorithm in step 4 is expressed as: Where R is the radius of the earth, is the latitude, and λ is the longitude.

6. The map shopping method based on AI voice interaction according to claim 1, characterized in that: The payment process integrates voiceprint dynamic time warping (DTW) verification and calculates the user input voice feature sequence V = {V1, V2, ..., V n } and the registered voiceprint feature sequence V ′ = {V1 ′ ,V2 ′ ,...,V m ′ }, where: V i Represents the Mel-frequency cepstral coefficient (MFCC) feature vector of the i-th frame of the input speech; V j ′ Represents the MFCC feature vector of the jth frame of the registered voiceprint; The DTW distance calculation formula is: Where π is the alignment path, and if the distance is less than the preset threshold, the payment is authorized.

7. The map shopping method based on AI voice interaction according to claim 1, characterized in that: It also includes using a matrix decomposition model to predict user u’s preference score for product i based on user historical behavior data. The calculation formula is: u u is the potential feature vector of user u; v i is the potential feature vector of product i; b u is the bias term of user u; b i is the bias term for product i. The recommended results are rendered on the map interface through the heat map layer, and the ratings are highlighted. The product associated POI.

8. The map shopping method based on AI voice interaction according to claim 1, characterized in that: The user interaction interface of this method uses the React Native framework to achieve multi-terminal adaptation and uniformly handle the following interaction logic: Map rendering is accelerated via WebGL; Voice commands are captured via the Web Speech API or native SDK; Order status is synchronized through cross-platform push channels.

9. The map shopping method based on AI voice interaction according to claim 1, characterized in that: The data processing and business logic parts of this method adopt a microservice architecture, which is split into independent map services, voice services, order services, and delivery services, and processes high-concurrency order requests through Kafka message queues.

Images