Interactive shopping experience system in virtual reality

Abstract

The invention discloses an interactive shopping experience system in virtual reality, and the system comprises the following modules: a virtual reality shopping environment construction module which is used for generating a three-dimensional virtual store scene; the man-machine interaction control module is used for realizing natural interaction operation of a user in a virtual space; the commodity information management module is used for carrying out real-time docking and synchronization with back-end data of the e-commerce platform; the transaction processing module is used for executing encrypted transmission after the user completes commodity selection; the user behavior analysis module is used for collecting and analyzing a behavior track of a user in a virtual scene, constructing a user preference model based on behavior data and driving recommendation logic; and the commodity recommendation result rendering module is used for dynamically generating a recommended commodity list according to the user behavior analysis result. According to the invention, virtual reality and an attention recommendation mechanism are fused, and an immersive interactive shopping experience system is constructed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of virtual reality and human-computer interaction, and particularly relates to an interactive shopping experience system in virtual reality. BACKGROUND

[0002] Under the background of the rapid development of current e-commerce platforms and consumer technology, the demand of users for shopping experience gradually changes from traditional text and image display and click interaction to a more immersive, real-time and personalized way. Especially under the driving of the accelerated popularization of virtual reality (VR) devices, building a virtual shopping scene with immersive visual environment and natural interaction ability has become an important direction to improve user experience and consumer conversion efficiency. However, the degree of integration of virtual reality and e-commerce systems in the existing technical system is still low, and a perfect interactive closed loop and recommendation logic chain has not yet been formed, which limits the landing ability of virtual shopping in actual scenarios.

[0003] Currently, some platforms have tried to introduce three-dimensional product models or virtual showrooms into online shopping processes, but most of them only stay at the level of static display or local product rotation browsing, lacking immersive reconstruction of the entire shopping process. Users in such systems usually cannot achieve multi-modal interaction control such as natural language instructions, gestures or gaze focus, and product information cannot be dynamically bound to the real e-commerce backend system, resulting in delayed product attributes, non-real-time inventory information, and seriously affecting the reliability and smoothness of user operations. At the same time, most existing solutions do not introduce deep user behavior modeling mechanisms, and lack intelligent recommendation methods based on semantic intent recognition and interaction context analysis, resulting in a low degree of disconnection between recommended content and user real needs.

[0004] In addition, in terms of transaction process handling, traditional VR shopping systems generally fail to connect the complete link from product selection, payment verification to result confirmation, and the coupling degree of payment methods and user account systems is not high, often causing payment response lag or identity verification chain breakage, which cannot guarantee the continuity of interaction and data security. At the same time, the collection and feedback utilization of user behavior data in the system are not sufficient, and there is a lack of unified behavior log structure and iterative user portrait mechanism, resulting in the inability of the recommendation strategy to adaptively optimize and the difficulty of realizing the strategy update closed loop for long-term interaction evolution.

[0005] The interactive shopping experience system in virtual reality provided by the present application is just to solve the above problems, and systematically introduces a three-dimensional virtual scene construction module, a human-computer control module supporting natural interaction, a real-time linkage commodity information management module, and a behavior analysis and intelligent recommendation system throughout the whole process. By constructing a complete data link from intention recognition, personalized recommendation to transaction confirmation, and fusing the Transformer architecture and multi-channel attention mechanism for intention modeling and recommendation optimization, the present application effectively improves the immersion, accuracy and closed-loop transaction capability in the virtual shopping process, and solves the key technical bottlenecks such as lack of immersion, interaction fault, rough recommendation and non-closed transaction chain in the prior art.

[0006] Therefore, how to provide an interactive shopping experience system in virtual reality is a problem that those skilled in the art need to solve. SUMMARY

[0007] One object of the present application is to provide an interactive shopping experience system in virtual reality, which integrates virtual reality modeling, multi-modal human-computer interaction and attention mechanism recommendation algorithm, and constructs an immersive and intelligent virtual shopping experience system. Through three-dimensional scene generation, semantic intention recognition and real-time synchronization of goods, the user's immersion and interaction freedom are improved, accurate recommendation and transaction closed loop are realized, and the technical bottlenecks of traditional e-commerce in terms of single experience, rigid interaction and rough recommendation are solved.

[0008] According to an embodiment of the present application, an interactive shopping experience system in virtual reality comprises the following modules:

[0009] A virtual reality shopping environment construction module is used to generate a three-dimensional virtual store scene, which includes a commodity display area, a navigation area and an interaction interface. The module constructs shelves, commodity models and scene light and shadow effects in the virtual space based on three-dimensional modeling technology, and is used to present a shopping environment with immersion and spatial sense, and supports user walking, observation and purchase in the virtual space.

[0010] A human-computer interaction control module is used to realize natural interaction operations of the user in the virtual space, including commodity selection, information calling, trial starting and payment confirmation. The human-computer interaction control module includes a gesture recognition unit, a voice recognition unit and an eye tracking unit, which are respectively used to recognize hand movements, voice instructions and visual focus behaviors of the user, and map the interaction intention to system control instructions.

[0011] A commodity information management module is used to real-time interface and synchronize with the back-end data of the e-commerce platform. The module accesses the commodity SKU information library, and calls commodity images, prices, inventory states and attribute parameter information, binds the above information to the corresponding three-dimensional commodity model, and dynamically loads and displays when the user interacts.

[0012] a transaction processing module for executing order generation, account verification, payment operation and encrypted transmission of payment data after the user completes the product selection, the transaction processing module comprising a virtual payment interface, an order generation unit and a payment authentication unit, the virtual payment interface supporting digital wallet operation, virtual keyboard input and voice payment control, the order generation unit generating structured order information according to the interaction record of the user in the virtual scene, and the payment authentication unit performing user identity verification and payment result confirmation;

[0013] a user behavior analysis module for collecting and analyzing the behavior trajectory of the user in the virtual scene, including the browsing path, the gaze dwell time, the interaction click frequency and the product trial record, and constructing a user preference model based on the behavior data for driving subsequent recommendation logic;

[0014] a product recommendation result rendering module for dynamically generating a recommended product list according to the user behavior analysis result and visually labeling in the virtual environment in a highlighted display, interface top or prominent label manner, so that the user obtains personalized guidance and intelligent push in the shopping process.

[0015] An interactive shopping experience method in virtual reality according to an embodiment of the application comprises the following steps:

[0016] S1, initializing a virtual reality shopping environment construction module, loading a virtual scene template and a three-dimensional product model resource, and constructing a three-dimensional virtual shopping scene containing space layout, interaction hotspot area and path navigation;

[0017] S2, starting a human-computer interaction control module, binding a multi-modal input signal channel, respectively connecting a gesture recognition unit, a voice recognition unit and an eye tracking unit, for capturing natural actions of the user and generating interaction control instructions;

[0018] S3, calling a product information management module, synchronizing product meta information, inventory status and real-time price from a product database, and mapping and binding the product information to the corresponding three-dimensional model node in the virtual scene;

[0019] S4, during the user browsing process, collecting continuous user behavior sequences including browsing path, operation action, gaze focus and voice feedback, and constructing a behavior log sequence input tensor;

[0020] S5, inputting the behavior log sequence into a user behavior intention recognition network based on a multi-layer Transformer structure, the network being composed of a position encoding layer, a self-attention layer and a feedforward network layer, and outputting a user current behavior intention classification result;

[0021] S6, input the behavior intention classification result and the user historical preference model into a multi-channel attention recommendation algorithm, the algorithm fuses the commodity label vector, the scene interaction context and the user behavior vector, calculates the matching score of each candidate commodity and generates a recommended commodity set;

[0022] S7, deliver the recommended commodity set to a commodity recommendation result rendering module, combine the matching score ranking result, adjust the display priority of the commodity, and highlight the display in the virtual reality scene in the form of label marking, interface highlighting and visual focusing;

[0023] S8, when the user selects a target commodity to initiate a transaction request, a structured order data is generated by a transaction processing module, an account verification process is executed and a virtual payment interface is called to complete the transaction process submission, and the payment process supports virtual keyboard input and voice confirmation interaction;

[0024] S9, after the user completes the transaction or exits the scene, the behavior and recommendation data of this round are archived, the user preference model is updated, and is used to optimize the subsequent recommendation strategy and display configuration.

[0025] Optionally, the S2 specifically comprises:

[0026] S21, start a man-machine interaction control module, access a multi-modal input signal channel connected with a virtual reality terminal device, and complete device state detection and initialization configuration;

[0027] S22, load a gesture recognition unit, use a spatial positioning sensor or a body sensing capture device to collect dynamic trajectories of the user's upper limbs and hands, recognize standard gesture instructions such as clicking, grabbing, sliding and rotating, and map the instructions to interaction events of corresponding commodity nodes in the scene;

[0028] S23, load a speech recognition unit, activate a microphone array, collect user speech data in real time, perform speech signal coding processing and command word extraction operations, and extract keyword instructions for commodity search, information callout, filtering switching and transaction confirmation control;

[0029] S24, load an eye movement tracking unit, complete initial calibration of the user eye movement tracking system, collect user binocular gaze position and gaze time data, determine the user's current attention commodity area, and use it to perform commodity focusing, information loading and automatic confirmation operations;

[0030] S25, establish a multi-channel fusion strategy between the gesture recognition unit, the speech recognition unit and the eye movement tracking unit, when two or more input signals are triggered at the same time, generate a unified control instruction based on a priority judgment or a fusion intention mechanism;

[0031] S26, input the unified control instruction to the commodity information management module, drive the corresponding interactive response in the three-dimensional scene, including commodity highlight display, detail window expansion, try-on interface loading and payment button activation.

[0032] Optionally, the S3 specifically comprises:

[0033] S31, start the commodity information management module, establish a network connection with the backend database of the e-commerce platform, and initialize the commodity information synchronization process;

[0034] S32, pull the commodity meta information associated with the current virtual scene from the database, the commodity meta information including commodity name, SKU code, main picture resource, price data, inventory status, attribute parameter and sales label;

[0035] S33, the mapping relationship between the commodity meta information and the three-dimensional commodity model according to the unique SKU identifier is established, and the binding configuration item is generated, which is used to drive the information loading and display synchronization in the subsequent interactive response;

[0036] S34, load the commodity main picture resource to the visual map area of the virtual scene corresponding to the commodity node, and the visual map area is a replaceable display area specified for the surface of the three-dimensional model;

[0037] S35, configure the interactive feature information of the commodity according to the commodity attribute parameters, including rotatable viewing, try-on simulation and multi-style switching operation options, and register in the response event table of the human-computer interaction control module;

[0038] S36, real-time synchronization monitoring of inventory status and price data, periodic access to e-commerce platform database or cache server, getting the latest value and refreshing the commodity state information displayed in the current scene;

[0039] S37, when the user performs commodity interactive operation, according to the user request to call commodity detailed information, load and display commodity introduction text, high-definition atlas, parameter explanation, user evaluation and related recommendation link, and configure different display layout formats according to the type of commodity.

[0040] Optionally, the S4 specifically comprises:

[0041] S41, after the user enters the virtual shopping scene, behavior data collection is carried out, gesture track recording channel, voice instruction listening channel, gaze tracking channel and location path tracking channel are opened;

[0042] S42, real-time record the moving path, stay area, commodity proximity distance and stay time of the user in the scene, generate the space-time behavior track of the user in the virtual space;

[0043] S43: Collect user gesture interaction action sequences on the product, including grabbing, rotating, scaling, and clicking actions, and mark them as structured interaction event data;

[0044] S44: Collect user voice input information, extract command sentences, question sentences, and descriptive sentences, and record the corresponding timestamps and product interaction context;

[0045] S45. Collect the changes in the user's gaze position in the scene, and record the gaze start time, end time, and number of consecutive gazes for each product node for subsequent preference analysis and modeling;

[0046] S46, organizing the collected user behavior information into a behavior event sequence according to the time dimension, the interaction object dimension, and the interaction mode dimension;

[0047] S47. Encode the behavior event sequence into a data structure in the form of a tensor, and package it into a behavior log sequence input tensor as input data for the subsequent behavior intention recognition network.

[0048] Optionally, the S5 specifically includes:

[0049] S51. Input the behavior log sequence input tensor output by the user behavior analysis module into the user behavior intention recognition network. The network is built based on a multi-layer Transformer architecture and includes an input embedding layer, a position encoding layer, a Transformer encoder stacking module, and a behavior intention classification output layer.

[0050] S52. In the input embedding layer, the semantic attributes, timestamp, spatial location information, interaction type, and product association ID in each behavioral event are discretized and embedded into a high-dimensional behavioral vector space to construct a structured input sequence.

[0051] S53. In the position encoding layer, a fixed position encoding vector is introduced into the input sequence to supplement the temporal sequence features of the behavior events and merge them into the behavior embedding representation;

[0052] S54, feeding the position-encoded behavior sequence into a multi-layer Transformer encoder module, wherein each layer of the multi-layer Transformer encoder includes a self-attention sublayer and a feedforward neural network sublayer, and layer normalization and residual connection are applied within each layer to extract contextual dependencies across events in the behavior sequence;

[0053] S55, perform feature accumulation and cross-layer fusion on the output of each layer to form a context-sensitive semantic embedding representation of the behavior sequence;

[0054] S56, in the behavior intention classification output layer, a multi-class intention label node is set, including commodity interest activation, commodity detailed information exploration, try-on or try-out intention, pre-purchase comparison intention, transaction initiation intention, and exit scene intention, and a Softmax structure is used to complete intention probability distribution calculation;

[0055] S57, according to the intention probability distribution, a main behavior intention classification result of the current period of the user is output, and the classification result is provided as input to a multi-channel attention recommendation algorithm module, used to guide recommendation weight adjustment and commodity candidate set screening.

[0056] Optionally, the S6 specifically comprises:

[0057] S61, three types of input vectors for recommendation processing are constructed, including a user behavior vector, a scene context vector and a commodity embedding vector, wherein the user behavior vector contains a current behavior intention label, a historical preference code and a recent interaction event sequence, the scene context vector contains a virtual scene position, a user gaze hot area, an operation node state and an interaction frame number, and the commodity embedding vector contains a label feature, an image code, an attribute index and a score label;

[0058] S62, the three types of input vectors are respectively input to a user interest modeling channel, a scene context channel and a behavior intention channel in the multi-channel attention recommendation algorithm module, each channel is composed of a feature alignment network, a self-attention mapping layer and a residual connection submodule, used to perform weighted focus calculation within the channel, and output a channel score vector;

[0059] S63, in the user interest modeling channel, a multi-head attention mechanism is performed based on the relevance between the user behavior vector and the commodity embedding vector, a matching mode between high-frequency behavior features and commodity attributes is captured, and an interest weight matrix is generated;

[0060] S64, in the scene context channel, the current user visual focus, the operation node and the scene area state are fused, the spatial relevance of the commodity in the current virtual environment is scored, and a scene adaptation score vector is generated;

[0061] S65, in the behavior intention channel, based on the similarity features between the behavior intention label and the commodity use label, a gating attention mechanism is used to perform intention adaptability analysis on the candidate commodity, and an intention fit score is output;

[0062] S66, the weight matrix and the score vector output by each channel are sent to a channel fusion layer, a comprehensive recommendation score is generated through a weighted fusion strategy, the fusion process introduces a channel importance factor and a confidence threshold regulation mechanism, and a ranking result is generated;

[0063] S67, select the commodity with the highest comprehensive recommendation score from the sorted results to form a candidate recommendation set, and perform consistency elimination on the commodity that is inconsistent with the current behavior intention classification result to generate a final recommended commodity set;

[0064] S68, bind the final recommended commodity set, channel dominant score label and user behavior label, and output as the input basis of the recommendation result rendering module.

[0065] Optionally, the S7 specifically includes:

[0066] S71, receive the final recommended commodity set and its sorting priority, channel dominant score label and user behavior intention label generated by the multi-channel attention recommendation algorithm;

[0067] S72, call the commodity recommendation result rendering module, initialize the scene rendering queue and display layout manager, read the current virtual shopping scene structure information and the user's location area;

[0068] S73, according to the priority order of the recommended commodities, bind the recommended commodities to the interactive display position in the area where the current user's field of view is located, and construct a recommended display element list, the display element including a three-dimensional commodity model, a label identification node, a light effect highlight layer and a transparent background plate;

[0069] S74, attach a recommendation identification component to each recommended commodity node, the component including a recommendation source label, a priority layer number and an interactive animation effect, and express the recommendation prompt through dynamically changing frame, color, transparency and flashing frequency;

[0070] S75, place the recommended commodity with higher matching degree with the current behavior intention label within the main channel range of the user's gaze path, and preferentially allocate the display position in the central visual area, and use visual attraction strategy to control its orientation, zoom and entrance animation effect;

[0071] S76, for the recommended commodity node selected by the user, trigger the information loading mechanism and the recommendation logic backtracking interface, and display the recommendation reason, channel contribution structure and semantic fit degree with the current behavior state of the commodity;

[0072] S77, after the recommendation result rendering is completed, continuously monitor the subsequent interactive operation behavior of the user, and statistically collect the click rate, residence time and recommendation item response rate of the commodity, and feed back to the behavior analysis module for model dynamic update and recommendation strategy adjustment.

[0073] Optionally, the S8 specifically includes:

[0074] S81, after the user selects the target commodity and initiates a transaction request, call the transaction processing module, initialize the order data structure and lock the current commodity inventory state;

[0075] S82, automatically generating a structured order item according to the user interaction behavior record, the item including a commodity identifier, a specification parameter, quantity information, a price snapshot, a user identity identifier, and an interaction timestamp;

[0076] S83, starting an account verification process, calling a user authentication service interface, obtaining user identity credentials through a virtual keyboard input, voice recognition, or a user account binding mechanism, and completing user validity confirmation;

[0077] S84, after verification succeeds, loading a virtual payment interface module, rendering a payment operation interface and allocating a virtual button area and an instruction input channel, supporting a user to complete a payment action through gesture clicking, gaze confirmation, or voice instructions;

[0078] S85, after payment is completed, generating a unique transaction voucher, and encapsulating order status, payment results, and response time as transaction receipt data, returning to a user interaction interface and completing commodity status updating;

[0079] S86, encapsulating user interaction records, commodity decision paths, payment method preferences, and scene response data related to the current transaction behavior as transaction behavior logs, and sending the transaction behavior logs to a user behavior analysis module for archival processing;

[0080] S87, periodically monitoring a transaction completion state, ensuring that an order circulation link is normal, and reporting an abnormal state to an exception handling module for fault determination and compensation strategy triggering.

[0081] Optionally, the S9 specifically includes:

[0082] S91, after a user completes a transaction process or exits a virtual shopping scene, calling a user behavior analysis module to execute a behavior data archival process of the current session;

[0083] S92, extracting behavior record data of a user in a current interaction process, including a browsing path, an interaction action sequence, a voice command trajectory, a gaze hotspot distribution, and a commodity node response state;

[0084] S93, collecting recommendation response logs from a recommendation result rendering module, including a display duration, an interaction frequency, a click confirmation state, and a final purchase result of each recommended commodity;

[0085] S94, structurally encapsulating the behavior data and the recommendation response logs, adding user identification, session number, virtual scene number, and timestamp information, constructing a behavior archive, and writing the behavior archive into a user portrait database;

[0086] S95, the user behavior analysis module analyzes the archived data, updates the user's preference model parameters, the scene path dependence mode and the recommendation response weight, which are used to adjust the initial state of the next round of shopping recommendation task;

[0087] S96, the updated behavior preference information is transmitted to the commodity information management module and the commodity recommendation result rendering module, respectively used for screening the candidate commodity set and configuring the recommendation display strategy;

[0088] S97, the influence of the above adjustment on the recommendation click rate, the residence time and the commodity conversion rate is continuously recorded in the subsequent user session, the long-term evaluation index is established, and the phased iteration and optimization management of the user portrait are provided.

[0089] The beneficial effects of the present application are:

[0090] The present application realizes the technical reconstruction of the traditional e-commerce interactive experience and commodity display mode by constructing a virtual reality shopping experience system integrating virtual scene modeling, multi-modal human-computer interaction, intelligent recommendation calculation and whole-process transaction processing, which brings significant beneficial effects.

[0091] Firstly, the present application relies on the virtual reality shopping environment construction module, which can generate a three-dimensional virtual store scene including a commodity display area, a path navigation area and an interactive interface, and restore the real shopping scene by means of three-dimensional modeling and space rendering technology, enhance the user's spatial immersion and participation, break the information limitation of traditional e-commerce platform two-dimensional text page, and enable the user to freely move, select and operate in the scene, greatly improving the visual perception and interaction authenticity.

[0092] Secondly, the system integrates gesture recognition, speech recognition and eye tracking technology to build a human-computer interaction control module, and the user can complete commodity selection, detail viewing, try-on operation and even payment initiation through natural language, action or gaze, greatly improving the interaction efficiency and intuitiveness. The multi-modal input mechanism effectively solves the problem of single interaction method and discontinuous response in traditional VR shopping system, and realizes a high-degree-of-freedom, low-learning-cost control method.

[0093] In addition, the present application can realize real-time synchronization of commodity SKU, price, inventory and attribute information by connecting the e-commerce platform database through the commodity information management module, realize high binding of three-dimensional model and actual commodity data, guarantee the accuracy, real-time and response of the information received by the user in the virtual environment, and improve the business executability and system integrity in the virtual scene.

[0094] Further, the application introduces a user behavior intention recognition network based on the Transformer structure, which combines user historical behavior, current context and interaction intention, accurately depicts user shopping motivation, and works with a multi-channel attention recommendation algorithm to complete product matching degree calculation and personalized recommendation output, significantly improving the relevance and conversion rate of recommended content. Compared with traditional static or rule-driven recommendation mechanisms, the system has stronger dynamic adaptation ability and user understanding ability.

[0095] Finally, after completing the transaction, the system structures the user behavior log, click feedback and transaction path through the linkage mechanism of the user behavior analysis module and the recommendation result rendering module, and dynamically adjusts the user portrait and recommendation parameters, realizing the whole-process closed loop from interaction collection to recommendation strategy optimization, providing sustainable data support and algorithm foundation for personalized content pushing, recommendation accuracy improvement and user long-term stickiness enhancement.

[0096] In summary, the application not only breaks through the limitations of traditional e-commerce and VR systems in user immersion, interaction ability, recommendation intelligence and transaction process, but also realizes the multiple improvements of technical integrity, business closed loop and user experience through architecture interconnection, data driving and model optimization, and has significant technical value and application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0097] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, together with the embodiments of the application, to explain the application, and do not constitute a limitation on the application. In the drawings:

[0098] Figure 1 A structural schematic diagram of an interactive shopping experience system in virtual reality is proposed for the application;

[0099] Figure 2 A whole process diagram of an interactive shopping experience method in virtual reality is proposed for the application. DETAILED DESCRIPTION

[0100] The application will now be described in further detail in conjunction with the accompanying drawings. These drawings are all simplified schematic diagrams, and only illustrate the basic structure of the application in a schematic manner, and therefore only show the components related to the application.

[0101] REFERENCE Figure 1 An interactive shopping experience system in virtual reality includes the following modules:

[0102] A virtual reality shopping environment construction module is configured to generate a three-dimensional virtual store scene, which includes a commodity display area, a navigation area and an interaction interface, and the module is configured to construct a shelf, a commodity model and a scene light and shadow effect in a virtual space based on a three-dimensional modeling technology, so as to present a shopping environment with immersion and spatial sense, and support a user to walk, observe and select and purchase in the virtual space.

[0103] A human-computer interaction control module is configured to realize natural interaction operations of a user in a virtual space, including commodity selection, information calling, trial starting and payment confirmation, and the human-computer interaction control module includes a gesture recognition unit, a voice recognition unit and an eye movement tracking unit, which are respectively configured to recognize hand actions, voice instructions and visual focus behaviors of the user, and map an interaction intention into a system control instruction.

[0104] A commodity information management module is configured to be connected and synchronized with a backend data of an e-commerce platform in real time, the module is configured to access a commodity SKU information library, and to call commodity image, price, inventory status and attribute parameter information, bind the above information to a corresponding three-dimensional commodity model, and dynamically load and display when a user interacts.

[0105] A transaction processing module is configured to execute order generation, account verification, payment operation and payment data encryption transmission after a user completes commodity selection, and the transaction processing module includes a virtual payment interface, an order generation unit and a payment authentication unit, the virtual payment interface supports digital wallet operation, virtual keyboard input and voice payment control, the order generation unit generates structured order information according to interaction records of the user in a virtual scene, and the payment authentication unit performs user identity verification and payment result confirmation.

[0106] A user behavior analysis module is configured to collect and analyze behavior trajectories of a user in a virtual scene, including a browsing path, a visual dwell time, an interaction click frequency and a commodity trial record, and to construct a user preference model based on behavior data, so as to drive subsequent recommendation logic.

[0107] A commodity recommendation result rendering module is configured to dynamically generate a recommended commodity list according to a user behavior analysis result, and to visually mark in a virtual environment in a highlighted display, an interface top or a significant label manner, so that a user obtains personalized guidance and intelligent push in a shopping process.

[0108] The virtual reality shopping experience system constructed in the application realizes full-process module coupling from virtual scene modeling, multi-modal human-computer interaction, commodity data synchronization, behavior-driven intelligent recommendation to transaction link closed loop at a system architecture level, effectively solves the problems of function fragmentation, interaction interruption and recommendation disconnection in a traditional VR shopping system, and supports usability and landing of an immersive business scene.

[0109] Reference Figure 2 A method for interactive shopping experience in virtual reality, comprising the following steps:

[0110] S1, initialize the virtual reality shopping environment construction module, load the virtual scene template and three-dimensional commodity model resources, and construct a three-dimensional virtual shopping scene containing space layout, interactive hot spot area and path navigation;

[0111] S2, start the human-computer interaction control module, bind the multi-modal input signal channel, respectively access the gesture recognition unit, voice recognition unit and eye movement tracking unit, which are used to capture the natural action of the user and generate interaction control instructions;

[0112] S3, call the commodity information management module, synchronize the commodity meta-information, inventory status and real-time price from the commodity database, and map and bind the commodity information to the corresponding three-dimensional model node in the virtual scene;

[0113] S4, in the user browsing process, collect continuous user behavior sequence, including browsing path, operation action, visual focus and voice feedback, and construct behavior log sequence input tensor;

[0114] S5, input the behavior log sequence into the user behavior intention recognition network based on the multi-layer Transformer structure, which is composed of position encoding layer, self-attention layer and feedforward network layer, and output the user current behavior intention classification result;

[0115] S6, input the behavior intention classification result and user historical preference model into the multi-channel attention recommendation algorithm, which integrates commodity label vector, scene interaction context and user behavior vector, calculates the matching score of each candidate commodity and generates the recommended commodity set;

[0116] S7, pass the recommended commodity set to the commodity recommendation result rendering module, combine the matching score sorting result, adjust the display priority of the commodity, and highlight the display in the virtual reality scene in the form of label marking, interface highlighting and visual focusing;

[0117] S8, when the user selects the target commodity to initiate a transaction request, the transaction processing module generates structured order data, performs account verification process and calls virtual payment interface to complete transaction process submission, and the payment process supports virtual keyboard input and voice confirmation interaction;

[0118] S9, after the user completes the transaction or exits the scene, archive the round of behavior and recommendation data, update the user preference model, and use it to optimize the subsequent recommendation strategy and display configuration.

[0119] The application divides the shopping process into nine ordered steps, clearly defines the execution logic, data interface and behavior dependency relationship between modules, so that the user's browsing, decision-making, payment and recommendation process in the virtual environment can be responded and smoothly connected in real time, and the interactive stability, response time delay control ability and multi-module concurrent cooperation efficiency of the system are improved.

[0120] In the embodiment, the S2 specifically comprises:

[0121] S21, start the human-computer interaction control module, access the multi-modal input signal channel connected with the virtual reality terminal device, and complete the device state detection and initialization configuration;

[0122] S22, load the gesture recognition unit, collect the dynamic trajectory of the user's upper limbs and hands by using the spatial positioning sensor or somatosensory capture device, recognize the click, grab, slide and rotate standard gesture instructions, and map the instructions to the interaction event of the corresponding commodity node in the scene;

[0123] S23, load the speech recognition unit, activate the microphone array, collect user voice data in real time, perform speech signal coding processing and command word extraction operation, extract key word instructions for commodity search, information callout, filtering switching and transaction confirmation control;

[0124] S24, load the eye movement tracking unit, complete the initial calibration of the user eye movement tracking system, collect the user's binocular gaze position and gaze time data, determine the user's current attention commodity area, and execute the commodity focusing, information loading and automatic confirmation operation;

[0125] S25, establish a multi-channel fusion strategy between the gesture recognition unit, the speech recognition unit and the eye movement tracking unit, when two or more input signals are triggered at the same time, generate a unified control instruction based on the priority judgment or fusion intention mechanism;

[0126] S26, input the unified control instruction into the commodity information management module, drive the corresponding interaction response in the three-dimensional scene, including commodity highlight display, detail window expansion, try-on interface loading and payment button activation.

[0127] The application unifies gestures, speech and eye movement channels into structured control instructions through a multi-modal input fusion strategy, ensures that the user's interaction intention in the virtual scene can be accurately captured and mapped under the intervention of any channel, effectively improves the redundancy fault tolerance and behavior fluency of the interaction instruction recognition, and significantly improves the interaction adaptation ability of the weak learning user group.

[0128] In the embodiment, the S3 specifically comprises:

[0129] S31, start the commodity information management module, establish a network connection with the backend database of the e-commerce platform, and initialize the commodity information synchronization process;

[0130] S32, pull the commodity meta information associated with the current virtual scene from the database, the commodity meta information including commodity name, SKU code, main picture resource, price data, inventory status, attribute parameter and sales label;

[0131] S33, the commodity meta information is mapped according to the unique SKU identifier and the three-dimensional commodity model, and a binding configuration item is generated, which is used to drive the information loading and display synchronization in the subsequent interactive response;

[0132] S34, load the commodity main picture resource to the visual map area of the virtual scene corresponding to the commodity node, and the visual map area is a replaceable display area specified for the surface of the three-dimensional model;

[0133] S35, configure the interactive feature information of the commodity according to the commodity attribute parameters, including rotatable viewing, try-on simulation and multi-style switching operation options, and register in the response event table of the man-machine interaction control module;

[0134] S36, real-time synchronization monitoring of inventory status and price data, periodic access to e-commerce platform database or cache server, getting the latest value and refreshing the commodity state information displayed in the current scene;

[0135] S37, when the user performs a commodity interaction operation, according to the user request to call the commodity detailed information, load and display the commodity introduction text, high-definition atlas, parameter explanation, user evaluation and related recommendation link, and configure different display layout formats according to the commodity type.

[0136] The commodity information management module of the application binds the SKU mapping mechanism and the three-dimensional model binding strategy, ensures that the commodity attributes, image resources and inventory status can be presented and updated in real time in the virtual space, the system supports automatic layout adjustment and state refresh strategy of commodity display component, significantly enhances the accuracy of commodity interaction response and the immediacy of business logic.

[0137] In the embodiment, the S4 specifically includes:

[0138] S41, after the user enters the virtual shopping scene, behavior data collection is performed, gesture track recording channel, voice instruction listening channel, gaze tracking channel and location path tracking channel are opened;

[0139] S42, real-time record the moving path, stay area, commodity proximity distance and stay duration of the user in the scene, generate the space-time behavior track of the user in the virtual space;

[0140] S43, collect gesture interaction action sequences of the user on the commodity, including grabbing, rotating, zooming and clicking operation actions, and mark them as structured interaction event data;

[0141] S44, collect user voice input information, extract instruction type sentences, interrogative type sentences and descriptive type sentences, and record corresponding time stamps and commodity interaction contexts;

[0142] S45, collect the gaze position change of the user's visual line in the scene, record the gaze start time, end time and continuous gaze number of each commodity node, which is used for subsequent preference analysis modeling;

[0143] S46, arrange the collected user behavior information into behavior event sequences according to time dimension, interaction object dimension and interaction mode dimension;

[0144] S47, encode the behavior event sequence into a tensor form data structure, and pack it into a behavior log sequence input tensor as input data of a subsequent behavior intention recognition network.

[0145] In the user behavior collection process, the application constructs a behavior tensor structure covering space, time, semantics and action dimensions, which can convert the original signal sequence into a behavior event sequence input with consistent semantics and unified format, support high-level semantic intention modeling and behavior path analysis tasks, and improve the perception resolution of the system to complex interaction behaviors and the modeling efficiency of input data.

[0146] In the embodiment, the S5 specifically includes:

[0147] S51, input the behavior log sequence input tensor output by the user behavior analysis module into the user behavior intention recognition network, which is constructed based on a multi-layer Transformer architecture, and includes an input embedding layer, a position encoding layer, a Transformer encoder stacking module and a behavior intention classification output layer;

[0148] S52, in the input embedding layer, the semantic attributes, time stamps, spatial position information, interaction types and commodity association IDs in each behavior event are discretely coded and embedded into a high-dimensional behavior vector space to construct a structured input sequence;

[0149] S53, in the position encoding layer, a fixed position encoding vector is introduced to the input sequence to supplement the time sequence features of the behavior event and fused into the behavior embedding representation;

[0150] S54, send the position coded behavior sequence into a multi-layer Transformer encoder module, each layer of the multi-layer Transformer encoder comprising a self-attention sublayer and a feedforward neural network sublayer, and layer normalization and residual connection are applied inside each layer, for extracting context dependency between events in the behavior sequence;

[0151] S55, accumulate features and fuse across layers for each layer output to form a context-sensitive behavior sequence semantic embedding representation;

[0152] S56, in the behavior intention classification output layer, a multi-class intention label node is set, including product interest activation, product detailed information exploration, try-on or try-use intention, pre-purchase comparison intention, transaction initiation intention, and exit scene intention, and a Softmax structure is used to complete intention probability distribution calculation;

[0153] S57, according to the intention probability distribution, output the main behavior intention classification result of the current period of the user, and provide the classification result as input to the multi-channel attention recommendation algorithm module for guiding recommendation weight adjustment and product candidate set screening.

[0154] The application can realize long-term dependence and semantic shift modeling in the user behavior sequence by introducing a behavior intention recognition network based on a multi-layer Transformer architecture, and significantly enhance the model's perception ability of cross-time period behavior purpose by using position encoding, self-attention mechanism and classification label fusion structure, and maintain high robustness and recognition accuracy in multiple rounds of conversation.

[0155] In the embodiment, the S6 specifically comprises:

[0156] S61, three types of input vectors for recommendation processing are constructed, including a user behavior vector, a scene context vector and a product embedding vector, wherein the user behavior vector comprises a current behavior intention label, a historical preference code and a recent interaction event sequence, the scene context vector comprises a virtual scene position, a user gaze hot area, an operation node state and an interaction frame number, and the product embedding vector comprises a label feature, an image code, an attribute index and a score label;

[0157] S62, input the above three types of input vectors into a user interest modeling channel, a scene context channel and a behavior intention channel in the multi-channel attention recommendation algorithm module respectively, each channel is composed of a feature alignment network, a self-attention mapping layer and a residual connection sub-module, for performing weighted focus calculation in the channel, and outputting a channel score vector;

[0158] S63, in the user interest modeling channel, a multi-head attention mechanism is performed based on the relevance between the user behavior vector and the commodity embedding vector, high-frequency behavior features and commodity attributes are captured, and an interest weight matrix is generated;

[0159] S64, in the scene context channel, the current user visual focus, operation node and scene region state are fused, the spatial correlation of the commodity in the current virtual environment is scored, and a scene adaptation score vector is generated;

[0160] S65, in the behavior intention channel, based on the similarity features between the behavior intention label and the commodity use label, the candidate commodity is analyzed by the gating attention mechanism, and the intention matching score is output;

[0161] S66, the weight matrix and the score vector output by each channel are sent to the channel fusion layer, and the comprehensive recommendation score is generated through the weighted fusion strategy, the channel importance factor and the confidence threshold regulation mechanism are introduced in the fusion process, and the ranking result is generated;

[0162] S67, the candidate recommendation set is composed of the commodities whose comprehensive recommendation scores are in the front of the pre-set ranking in the ranking result, and the consistency of the commodities which are contrary to the current behavior intention classification result is removed, and the final recommended commodity set is generated;

[0163] S68, the final recommended commodity set, the channel dominant score label and the user behavior label are bound and output as the input basis of the recommendation result rendering module.

[0164] The multi-channel attention recommendation algorithm designed in the application supports parallel attention calculation and fusion of three kinds of heterogeneous features of user behavior, scene context and intention label, three independent attention channels of user interest, situation adaptation and intention matching are constructed, and a channel fusion mechanism is introduced, multi-dimensional matching score generation of the commodity candidate set is realized, and the relevance, explainability and context response ability of the recommendation output are effectively improved.

[0165] In the embodiment, the S7 specifically comprises:

[0166] S71, receiving the final recommended commodity set, the ranking priority thereof, the channel dominant score label and the user behavior intention label generated by the multi-channel attention recommendation algorithm;

[0167] S72, calling the commodity recommendation result rendering module, initializing the scene rendering queue and the display layout manager, reading the current virtual shopping scene structure information and the position area where the user is located;

[0168] S73, according to the recommended commodity priority order, the recommended commodity is bound to the interactive display position of the area where the current user's field of view is located, and a recommended display element list is constructed, the display element includes a three-dimensional commodity model, a label identification node, a light effect highlight layer and a transparent background plate;

[0169] S74, a recommendation identification component is attached to each recommended commodity node, the component includes a recommendation source label, a priority layer number and an interactive animation effect, and the recommendation prompt is expressed through dynamically changing border, color, transparency and flicker frequency;

[0170] S75, the recommended commodity with higher matching degree with the current behavior intention label is placed in the main channel range of the user's gaze path, and the display position of the central visual angle area is preferentially assigned, and the visual attraction strategy is used to control the heading, zooming and entry animation effect;

[0171] S76, for the recommended commodity node selected by the user, a information loading mechanism and a recommendation logic backtracking interface are triggered, and the recommendation reason, the channel contribution structure and the semantic fit degree with the current behavior state of the commodity are displayed;

[0172] S77, after the recommended result rendering is completed, the subsequent interactive operation behavior of the user is continuously monitored, the commodity click rate, the residence time and the recommended item response rate are statistically collected, and feedback to the behavior analysis module is used for model dynamic updating and recommendation strategy adjustment.

[0173] The recommended result rendering module in the application binds the gaze position and the recommendation weight, combines dynamic visual markers, highlight levels and animation attraction mechanisms, embeds the recommended commodity into the user's gaze path in a priority driven manner, significantly improves the recommendation exposure rate and click conversion rate, and supports the semantic understanding of the user to the recommended result source through the recommendation identification and explanation window.

[0174] In the embodiment, the S8 specifically includes:

[0175] S81, after the user selects the target commodity and initiates a transaction request, a transaction processing module is called, an order data structure is initialized, and the current commodity inventory state is locked;

[0176] S82, a structured order item is automatically generated according to the user interaction behavior record, and the item includes a commodity identification, a specification parameter, a quantity information, a price snapshot, a user identity and an interaction timestamp;

[0177] S83, start the account verification process, call the user authentication service interface, obtain the user identity credentials through the virtual keyboard input, voice recognition or user account binding mechanism, and complete the user validity confirmation;

[0178] S84, after verification succeeds, load the virtual payment interface module, render the payment operation interface and distribute the virtual button area and instruction input channel, support the user to complete the payment action through gesture click, gaze confirmation or voice instruction;

[0179] S85, after payment is completed, generate a unique transaction voucher, and encapsulate the order state, payment result and response time as transaction receipt data, return to the user interaction interface and complete the commodity state update;

[0180] S86, the user interaction record related to the current transaction behavior, the commodity decision path, the payment method preference and the scene response data are encapsulated as transaction behavior log and sent to the user behavior analysis module for archiving processing;

[0181] S87, the transaction completion state is periodically monitored to ensure that the order circulation link is normal and the abnormal state is reported to the exception handling module for fault judgment and compensation strategy triggering.

[0182] The application builds a complete transaction chain of order generation, account verification, payment interaction and receipt confirmation through the transaction processing module, and the virtual payment process supports synchronous access of three modes of voice instruction, gaze confirmation and gesture input, which can optimize the virtual interaction operation path under the premise of ensuring transaction security, realize the real-time and immersive continuity of transaction execution.

[0183] In the embodiment, the S9 specifically includes:

[0184] S91, after the user completes the transaction process or exits the virtual shopping scene, the user behavior analysis module is called to execute the behavior data archiving process of the current session;

[0185] S92, extract the behavior record data of the user in the current interaction process, including the browsing path, interaction action sequence, voice command track, gaze hotspot distribution and commodity node response state;

[0186] S93, collect the recommendation response log from the recommendation result rendering module, including the display time length, interaction times, click confirmation state and final purchase result of each recommended commodity;

[0187] S94, structure and encapsulate the behavior data and the recommendation response log, add user identification, session number, virtual scene number and timestamp information, build a behavior file and write it into the user portrait database;

[0188] S95, the user behavior analysis module analyzes the archived data, updates the preference model parameters, scene path dependence mode and recommendation response weight of the user, which is used to adjust the initial state of the next round of shopping recommendation task;

[0189] S96, deliver the updated behavior preference information to the commodity information management module and the commodity recommendation result rendering module, respectively used for screening the candidate commodity set and configuring the recommendation display strategy;

[0190] S97, continuously record the influence of the above adjustment on the recommendation click rate, the dwell time and the commodity conversion rate in the subsequent user session, establish long-term evaluation indexes and provide phased iteration and optimization management of the user portrait.

[0191] The application introduces a unified behavior archiving mechanism after the transaction is completed, binds the recommendation response log and the user operation track into a structured behavior archive, performs model fine-tuning and strategy updating by the user behavior analysis module, constructs a self-evolution closed-loop logic of the recommendation strategy, realizes continuous precision optimization of the recommendation system and dynamic evolution of the user portrait, and improves the content adaptation efficiency and user stickiness in the long-term interaction scene.

[0192] Embodiment 1:

[0193] In order to verify the feasibility of the application in implementation, the application is applied to a large domestic e-commerce enterprise, a set of virtual reality shopping experience system constructed by the enterprise during the "520 Valentine's Day theme promotion" as a background scene, facing its newly launched digital shopping guide service, deploying the virtual reality interactive shopping system proposed in the application in the VR digital store, and continuously tracking and quantitatively evaluating the interaction efficiency, commodity recommendation accuracy and transaction conversion rate of consumers in the use process, verifying the technical effect of the system.

[0194] The system is applied to the "Heart-moving Gift · Virtual Lover Street" theme scene built by the enterprise. The user wears a PICO 4 head-mounted device and enters a three-dimensionally reconstructed virtual shopping street environment. The scene is configured with a virtual jewelry counter, a digital fragrance laboratory, a couple's clothing showcase and an AI virtual model interaction point. The system loads three-dimensional commodity models and interaction layouts through the virtual reality shopping environment construction module, and sets hot path and information triggering area in the panoramic scene to ensure that the user can naturally browse, stop and try and quickly purchase.

[0195] In the interaction process after the user enters the system, the human-computer interaction control module accesses the user's PICO motion controller, the microphone embedded in the head-mounted device and the eye tracking module. The system can recognize the user's gesture operations in real time, such as clicking, grabbing, rotating the perfume bottle and other actions; synchronously analyze voice commands such as "view ingredients", "try this one", "add to shopping cart", and identify the goods that the user gazes at for more than 3 seconds through eye movement analysis and pop up the detail page. The entire interaction process does not require a keyboard, mouse or additional UI controls, significantly improving naturalness and smoothness.

[0196] In the recommendation aspect, the system records the user's stay time, interaction frequency and voice query content through the user behavior analysis module during browsing, and inputs the behavior sequence data into the intention recognition network based on the multi-layer Transformer structure for processing. In the actual deployment scenario, the model contains 8 layers of encoder, each layer with 768-dimensional embedding, supporting 300 types of intention labels. Taking a user as an example, when browsing the perfume area, he stared at two luxury fragrances for more than 8 seconds and asked "Is it suitable for daily use?" and "Is there a lighter one?" The system identifies the intention as "personalized light fragrance recommendation" and links the multi-channel attention recommendation algorithm to filter out 12 fragrance products with an adaptation degree of more than 0.88 from 260 candidate goods, and presents the top 5 goods in the user's main visual area with visual highlights and virtual model fragrance testing animations.

[0197] In the transaction processing aspect, when the user issues an "immediate purchase" instruction through voice, the transaction processing module automatically pops up a payment interface, the user confirms the virtual payment button through eye movement, and the system calls the bound WeChat applet payment interface to complete order generation and payment completion return within 3.2 seconds. The whole process does not need to exit the VR environment or use external mobile devices to complete verification, ensuring the immersion and continuity of the payment path.

[0198] The system served about 21357 virtual store visits in a week, of which 14392 were effective interaction behaviors (stay time exceeding 90 seconds and performing more than 5 operations), and the system recorded more than 1180 million cumulative behavior logs. According to the back-end statistics, the conversion rate of transactions completed through the system is 16.3%, which is significantly higher than the 7.9% of the traditional text and image product page; the average browsing-recommendation-click response time is 1.92 seconds, which is 55% lower than the original system of 4.3 seconds; the user's click matching degree score of recommended goods (user subjective questionnaire survey) is 8.7 / 10 on average, which is significantly better than the 6.4 / 10 of the traditional collaborative recommendation algorithm system.

[0199] More importantly, the system solves three core technical problems in the original VR shopping system: first, the scene is static and the interaction is rigid, making it difficult to form an immersive and continuous experience; second, the recommended content does not match the user's real-time behavior, and cannot dynamically respond to interest shifts; third, the transaction chain is broken, requiring a web page or mobile phone operation, resulting in a high exit rate. The invention makes systematic improvements from bottom modeling to application interaction through the construction of a modular system architecture and a behavior-driven data closed-loop mechanism, making virtual shopping a full-link system with operational feasibility, intelligent push capability and business conversion basis.

[0200] In summary, the embodiment demonstrates the application effect of the application deployed in a real business environment, and the quantitative verification of real user behavior data and performance indicators confirms that the application has significant technical advantages and practical application value in immersive interaction, intent modeling, recommendation optimization, and closed-loop transactions.

[0201] The above merely describes a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, within the technical scope disclosed by the present application, according to the technical solution and inventive concept of the present application, makes equivalent replacements or changes, which should be covered within the protection scope of the present application.

Claims

1. An interactive shopping experience system in virtual reality, characterized in that: Includes the following modules: A virtual reality shopping environment building module is used to generate a three-dimensional virtual store scene, presenting a shopping environment with immersive and spatial sense; The human-computer interaction control module is used to enable natural user interaction in the virtual space, recognize user hand movements, voice commands, and eye focus, and map interaction intentions into system control commands; The product information management module is used to connect and synchronize with the back-end data of the e-commerce platform in real time, and dynamically load and display it during user interaction; The transaction processing module is used to execute order generation, account verification, payment operations and encrypted transmission of payment data after the user completes product selection; User behavior analysis module, which is used to collect and analyze user behavior trajectories in virtual scenes, and build user preference models based on behavior data to drive recommendation logic; The product recommendation result rendering module is used to dynamically generate a list of recommended products based on the results of user behavior analysis, and visually mark them in the virtual environment by highlighting, pinning them to the top of the interface, or by prominently labeling them.

2. The interactive shopping experience system in virtual reality according to claim 1, characterized in that: The modules are implemented as follows: S1. Initialize the virtual reality shopping environment construction module, load the virtual scene template and 3D product model resources, and build a 3D virtual shopping scene including space layout, interactive hot spots and path navigation; S2. Start the human-computer interaction control module, bind the multimodal input signal channel, and connect it to the gesture recognition unit, speech recognition unit, and eye tracking unit respectively; S3. Call the product information management module to synchronize product metadata, inventory status, and real-time prices from the product database, and map the product information to the corresponding 3D model node in the virtual scene; S4. During the user browsing process, collect continuous user behavior sequences and construct behavior log sequence input tensors; S5. Inputting the behavior log sequence into a user behavior intention recognition network based on a multi-layer Transformer structure; S6. Input the behavioral intention classification results into the multi-channel attention recommendation algorithm, calculate the matching score of each candidate product and generate a recommended product set; S7. The recommended product set is passed to the product recommendation result rendering module. The display priority of the products is adjusted based on the matching score ranking results. The products are then highlighted in the virtual reality scene through labeling, interface highlighting, and visual focus. S8. When the user selects a target product and initiates a transaction request, the transaction processing module generates structured order data, performs the account verification process, and calls the virtual payment interface to complete the transaction process submission; S9. After the user completes the transaction or exits the scenario, archive the current round of behavior and recommendation data and update the user preference model.

3. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S2 specifically includes: S21. Start the human-computer interaction control module, access the multimodal input signal channel connected to the virtual reality terminal device, and complete device status detection and initialization configuration; S22. Load the gesture recognition unit, use a spatial positioning sensor or somatosensory capture device to capture the dynamic trajectory of the user's upper limbs and hands, identify standard gesture commands such as click, grab, slide, and rotate, and map the commands to interaction events corresponding to product nodes in the scene; S23, loading the speech recognition unit, activating the microphone array, collecting user speech data in real time, performing speech signal encoding processing and command word extraction operations, and extracting keyword instructions; S24, loading the eye tracking unit, completing the initial calibration of the user's eye tracking system, collecting the user's binocular gaze position and gaze duration data, and determining the product area the user is currently focusing on; S25. Establish a multi-channel fusion strategy among the gesture recognition unit, the speech recognition unit, and the eye tracking unit, and generate a unified control instruction based on priority judgment or fusion intention mechanism when two or more input signals are triggered simultaneously; S26. Input the unified control instruction into the commodity information management module.

4. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S3 specifically includes: S31. Start the product information management module, establish a network connection with the e-commerce platform backend database, and initialize the product information synchronization process; S32: Pulling product meta information associated with the current virtual scene from the database, the product meta information including product name, SKU code, main image resource, price data, inventory status, attribute parameters and sales tag; S33: Establish a mapping relationship between the product meta information and the three-dimensional product model according to the unique SKU identifier, and generate a binding configuration item; S34, loading the product main image resource into the visual mapping area of ​​the corresponding product node in the virtual scene, wherein the visual mapping area is a designated replaceable display area on the surface of the three-dimensional model; S35. Configure the interactive feature information of the product according to the product attribute parameters, including rotation viewing, try-on simulation, and multi-style switching operation options, and register it in the response event table of the human-computer interaction control module; S36. Real-time synchronous monitoring of inventory status and price data, periodic access to the e-commerce platform database or cache server, obtaining the latest values ​​and refreshing the product status information displayed in the current scene; S37. When a user interacts with a product, detailed product information is retrieved according to the user's request, and product introduction text, high-definition pictures, parameter descriptions, user reviews and related recommended links are loaded and displayed, and different display layout formats are configured according to the product type.

5. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S4 specifically includes: S41. After the user enters the virtual shopping scene, behavioral data is collected, and a gesture trajectory recording channel, a voice command monitoring channel, a gaze tracking channel, and a position path tracking channel are enabled; S42: Record the user's movement path, stay area, proximity to products, and stay duration in real time in the scene to generate a spatiotemporal behavior trajectory of the user in the virtual space; S43: Collect user gesture interaction action sequences on the product, including grabbing, rotating, scaling, and clicking actions, and mark them as structured interaction event data; S44: Collect user voice input information, extract command sentences, question sentences, and descriptive sentences, and record the corresponding timestamps and product interaction context; S45, collecting changes in the user's gaze position in the scene, and recording the gaze start time, end time, and number of consecutive gazes for each product node; S46, organizing the collected user behavior information into a behavior event sequence according to the time dimension, the interaction object dimension, and the interaction mode dimension; S47. Encode the behavior event sequence into a data structure in the form of a tensor, and package it into a behavior log sequence input tensor.

6. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S5 specifically includes: S51. Input the behavior log sequence input tensor output by the user behavior analysis module into the user behavior intention recognition network. The network is built based on a multi-layer Transformer architecture and includes an input embedding layer, a position encoding layer, a Transformer encoder stacking module, and a behavior intention classification output layer. S52. In the input embedding layer, the semantic attributes, timestamp, spatial location information, interaction type, and product association ID in each behavioral event are discretized and embedded into a high-dimensional behavioral vector space to construct a structured input sequence. S53. In the position encoding layer, a fixed position encoding vector is introduced into the input sequence and fused into the behavior embedding representation; S54, sending the position-encoded behavior sequence to a multi-layer Transformer encoder module, wherein each layer of the multi-layer Transformer encoder includes a self-attention sublayer and a feedforward neural network sublayer, and layer normalization and residual connection are applied within each layer; S55, perform feature accumulation and cross-layer fusion on the output of each layer to form a context-sensitive semantic embedding representation of the behavior sequence; S56. In the behavioral intention classification output layer, multiple intent label nodes are set, including product interest activation, product detailed information exploration, try-on or trial intention, pre-purchase comparison intention, transaction initiation intention, and exit scenario intention. The Softmax structure is used to complete the intention probability distribution calculation; S57. Output the classification result of the user's main behavior intention in the current time period according to the intention probability distribution, and provide the classification result as input to the multi-channel attention recommendation algorithm module.

7. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S6 specifically includes: S61. Construct three types of input vectors for recommendation processing, including user behavior vectors, scene context vectors, and product embedding vectors. The user behavior vector includes the current behavior intention label, historical preference encoding, and recent interaction event sequence. The scene context vector includes the virtual scene location, user gaze hotspot, operation node status, and interaction frame number. The product embedding vector includes label features, image encoding, attribute index, and rating label. S62, inputting these three types of input vectors into the user interest modeling channel, scene context channel, and behavior intention channel in the multi-channel attention recommendation algorithm module respectively. Each channel is composed of a feature alignment network, a self-attention mapping layer, and a residual connection submodule; S63. In the user interest modeling channel, a multi-head attention mechanism is implemented based on the correlation between the user behavior vector and the product embedding vector to capture the matching pattern between high-frequency behavior features and product attributes and generate an interest weight matrix. S64. In the scene context channel, the current user's visual focus, the operation node, and the scene area state are integrated to score the spatial relevance of the product in the current virtual environment and generate a scene adaptation score vector. S65. In the behavioral intention channel, based on the similarity between the behavioral intention label and the product usage label, the gated attention mechanism is used to analyze the candidate products' intent adaptability and output an intent fit score. S66: Send the weight matrix and score vector output by each channel to the channel fusion layer, and generate a comprehensive recommendation score through a weighted fusion strategy. The fusion process introduces a channel importance factor and a confidence threshold control mechanism to generate a ranking result; S67: Select products with comprehensive recommendation scores ranked at the top of the preset ranking from the sorting results to form a candidate recommendation set, and perform consistency elimination on products that conflict with the current behavioral intention classification result to generate a final recommended product set; S68: Bind the final recommended product set, channel dominance score label, and user behavior label and output them.

8. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S7 specifically includes: S71, receiving the final recommended product set generated by the multi-channel attention recommendation algorithm and its sorting priority, channel dominance score label and user behavior intention label; S72: Calling the product recommendation result rendering module, initializing the scene rendering queue and display layout manager, and reading the current virtual shopping scene structure information and the user's location area; S73: Bind the recommended products to interactive display locations in the current user's field of view based on their priority, and construct a list of recommended display elements, including a 3D product model, a label identification node, a light effect highlight layer, and a transparent background panel. S74. Add a recommendation identification component to each recommended product node. The component includes a recommendation source label, a priority layer number, and an interactive animation effect. The recommendation prompt is expressed through dynamically changing borders, colors, transparency, and flashing frequency. S75. Place recommended products that have a high degree of match with the current behavioral intention label within the main channel of the user's gaze path, prioritize their display in the central viewing area, and use visual attraction strategies to control their orientation, scaling, and entry animation effects. S76. For the recommended product node selected by the user, trigger the information loading mechanism and the recommendation logic backtracking interface to display the product's recommendation reason, channel contribution structure, and semantic fit with the current behavior state; S77. After the recommendation results are rendered, continue to monitor the user's subsequent interactive operations, collect statistics on the product click-through rate, dwell time, and recommendation response rate, and feed them back to the behavior analysis module for dynamic model updates and recommendation strategy adjustments.

9. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S8 specifically includes: S81. After the user selects the target product and initiates a transaction request, the transaction processing module is called to initialize the order data structure and lock the current product inventory status; S82. Automatically generate a structured order entry based on the user interaction behavior record, wherein the entry includes a product identifier, specification parameters, quantity information, a price snapshot, a user identifier, and an interaction timestamp; S83. Start the account verification process, call the user authentication service interface, obtain the user identity credentials through virtual keyboard input, voice recognition or user account binding mechanism, and complete the user validity confirmation; S84: After successful verification, load the virtual payment interface module, render the payment operation interface, and allocate virtual button areas and command input channels; S85. After payment is completed, a unique transaction voucher is generated, and the order status, payment result, and response time are encapsulated as transaction receipt data, which is returned to the user interface and the product status is updated; S86: Encapsulate the user interaction records, product decision paths, payment method preferences, and scenario response data related to the current transaction behavior into a transaction behavior log, and send it to the user behavior analysis module for archiving; S87. Periodically monitor the transaction completion status.

10. The interactive shopping experience system in virtual reality according to claim 2, characterized in that: The S9 specifically includes: S91. After the user completes the transaction process or exits the virtual shopping scene, the user behavior analysis module is called to execute the behavior data archiving process of this round of conversation; S92: Extracting user behavior record data during this round of interaction, including browsing path, interaction action sequence, voice command trajectory, gaze hotspot distribution, and product node response status; S93. Collect recommendation response logs from the recommendation result rendering module, including the display duration, number of interactions, click confirmation status, and final purchase result of each recommended product; S94. Structurally encapsulate the behavior data and the recommendation response log, attach the user ID, session number, virtual scene number, and timestamp information, construct a behavior profile, and write it into the user portrait database; S95. The user behavior analysis module analyzes the archived data and updates the user's preference model parameters, the path dependency pattern within the scenario, and the recommendation response weight; S96: Transmit the updated behavior preference information to the product information management module and the product recommendation result rendering module; S97. Continue to record the impact of the above adjustments on recommendation click-through rate, dwell time, and product conversion rate in subsequent user sessions, establish long-term evaluation indicators, and provide phased iteration and optimization management of user portraits.

Images