A Dynamic Interactive Response Method and System for E-commerce Applications Based on Sensor Data

Abstract

This invention discloses a dynamic interactive response method and system for e-commerce applications based on sensor data, comprising the following steps: collecting and preprocessing data from the e-commerce application's operation to generate a standardized interactive input data set; extracting continuous state features and discrete event features to construct a dual-path coupling control sequence; performing stage correlation calculations to generate interactive stage indicator vectors; inputting the dual-path coupling control sequence and interactive stage indicator vectors into an improved NCDE model to generate a dynamic interactive hidden state sequence; performing response sensitivity modulation to construct a dynamic interactive response strategy set; dynamically adjusting the interface based on the dynamic interactive response strategy set to generate dynamic interactive response results; collecting subsequent feedback and writing back updates to optimize the dynamic interactive hidden state sequence and the dynamic interactive response strategy set. This invention improves the real-time performance, accuracy, and adaptive optimization capabilities of e-commerce interactive responses.

Description

Technical Field

[0001] This invention relates to the field of data sensing and interaction technology, and in particular to a dynamic interactive response method and system for e-commerce applications based on sensor data. Background Technology

[0002] As e-commerce applications continue to evolve towards more refined operations, platforms are placing higher demands on the real-time perception, dynamic analysis, and response control of user interactions. In existing e-commerce applications, interface content display, information prompt triggering, service entry configuration, and interactive feedback process adjustments largely rely on preset rules, static tag profiles, or historical behavior statistics. Common processing methods focus on offline analysis of click, browsing, purchase, favorites, and dwell time records, then use this data to perform recommendation sorting, page arrangement, or marketing outreach. While this approach can handle basic interaction adjustments, it lacks fine-grained characterization of continuous state changes, sudden event changes, and stage transitions within the current session.

[0003] In existing technologies, some solutions introduce terminal-side sensing information or temporal modeling methods to dynamically analyze user interaction behavior. These solutions often directly concatenate sensor data and behavioral data into a unified model, focusing on single-state recognition or single-type response generation. They lack hierarchical modeling methods to address the differences between continuous state change trends and the impact of discrete event triggers; they lack clear stage-related calculation processes for the interaction stage changes between browsing, comparison, conversion, and exit stages; and their model outputs often remain at the level of recommendation result generation or prompt action triggering, lacking a linkage adjustment mechanism around content display order, information prompting methods, service entry status, and interaction feedback flow. Existing technologies also lack a closed-loop update structure to write back the dynamic interaction response results to the input construction process and the hidden state evolution process. This results in a lack of feedback constraints for subsequent state evolution and response strategy adjustments, insufficient dynamic adaptability, and difficulty in meeting the technical requirements of real-time interactive responses in e-commerce application scenarios.

[0004] Therefore, how to provide a dynamic interactive response method and system for e-commerce applications based on sensor data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a dynamic interactive response method and system for e-commerce applications based on sensor data. This invention utilizes sensor data analysis, dual-path coupling modeling, stage correlation calculation, and continuous evolution and updating of hidden states to achieve dynamic generation and feedback write-back updates of e-commerce interface response strategies. It has the advantages of timely response, accurate stage identification, fine-grained interaction adjustment, and strong adaptability.

[0006] The dynamic interactive response method for e-commerce applications based on sensor data according to an embodiment of the present invention includes the following steps:

[0007] Data from the e-commerce application's operation is collected and preprocessed to generate a standardized interactive input data set. Continuous state features and discrete event features are extracted from this standardized interactive input data set to construct a continuous sensing control path and an interactive event triggering path. Coupled encoding is then performed to generate a dual-path coupled control sequence. Stage correlation calculations are performed on the standardized interactive input data set to generate interactive stage indicator vectors. The dual-path coupled control sequence and interactive stage indicator vectors are input into an improved NCDE model to perform continuous evolution updates of the hidden states. Gating modulation is applied to the state update weights during the hidden state evolution process to generate a dynamic interactive hidden state sequence. Response sensitivity modulation is then applied to the dynamic interactive hidden state sequence. The system calculates and generates response priority parameters, intervention sensitivity parameters, interaction form preference parameters, and feedback adaptation parameters to construct a dynamic interaction response strategy set. It then dynamically adjusts the content display order, information prompting method, service entry status, and interaction feedback process of the current e-commerce application interface to generate dynamic interaction response results. Finally, it collects subsequent click data, dwell time data, closing action data, page exit data, and conversion behavior data corresponding to the dynamic interaction response results, constructs a feedback correction vector, writes it back to the dual-path coupling control sequence construction process and the hidden state initialization process of the improved NCDE model, updates the dynamic interaction hidden state sequence and the dynamic interaction response strategy set, and generates dynamic interaction response results.

[0008] Optionally, the e-commerce application operation process data includes terminal sensor data, user operation behavior data, and application context data; the preprocessing includes timestamp alignment, missing value completion, outlier removal, duplicate record removal, numerical normalization, and sequence sorting.

[0009] Optionally, the process of generating the dual-path coupling control sequence includes:

[0010] The touch pressure data, sliding displacement data, terminal posture data and page dwell time data in the standardized interactive input data set are sorted according to timestamps. The sorted data is then segmented and sliced. The numerical change, direction of change, rate of change and duration between adjacent moments in each segment are calculated to generate a continuous state feature sequence.

[0011] The click data, swipe data, search data, page jump data, target object access data, add to cart data and close action data in the standardized interactive input data set are sorted according to the event occurrence time. The event type label, event occurrence time, time interval before and after the event, event corresponding page identifier and event corresponding target object identifier are extracted from the sorted data to generate a discrete event feature sequence.

[0012] Perform continuous-time interpolation on the continuous state feature sequence to reconstruct the continuous state change trajectory on a unified time axis, and construct a sensing continuous control path based on the reconstructed continuous state change trajectory.

[0013] Perform event trigger localization on discrete event feature sequences, map each discrete event to the corresponding trigger time in a unified time axis, generate event trigger increments at the corresponding trigger times, and construct interactive event trigger paths in chronological order based on each event trigger increment;

[0014] Align the continuous control path of the sensor and the triggering path of the interactive event according to a unified time axis. Extract the continuous state component in the continuous control path of the sensor and the event triggering component in the triggering path of the interactive event at each time position. Execute the corresponding positions and splice and associate them to generate the initial coupling sequence.

[0015] Joint mapping encoding is performed on the continuous state components and event-triggered components in the initial coupling sequence. The continuous state change trend and the discrete event triggering effect are written into the encoding result at the same time position to generate a dual-path coupling control sequence.

[0016] Optionally, the process of generating the interaction phase indication vector includes:

[0017] The standardized interactive input data set is divided into segments according to a unified timeline. Within each time segment, the page dwell time, number of swipes, swipe interval, number of page jumps, number of visits to the target object, and number of feedback actions are counted. The stage association features that represent the level of interactive activity, page switching, target object attention, and feedback response in the current time segment are extracted. The magnitude and direction of change of the stage association features of adjacent time segments are compared in chronological order, and the continuous change relationship and transition relationship between adjacent time segments are calculated.

[0018] Aggregation calculations are performed on the stage-related features over several consecutive time periods to determine the concentrated distribution of the stage-related features within the continuous time range. Based on the continuous change relationship, transition change relationship and concentrated distribution, stage-related calculations are performed to determine the stage marker value corresponding to each time period. The stage marker values ​​corresponding to each time period are arranged in chronological order to generate an interactive stage indicator vector.

[0019] Optionally, the process of generating the dynamic interaction hidden state sequence includes:

[0020] Read the dual-path coupling control sequence, interaction phase indicator vector and hidden state of the previous time position in sequence according to the unified time axis. At the current time position, input the continuous state component and event trigger component in the dual-path coupling control sequence into the improved NCDE model, and input the phase marker value in the interaction phase indicator vector into the phase perception gating structure.

[0021] Based on the stage marker value, the write intensity of the continuous state component at the current time position, the write intensity of the event-triggered component at the current time position, and the transmission intensity of the hidden state from the previous time position to the current time position are gated and modulated.

[0022] The gated modulation continuous state component, the event trigger component, and the hidden state of the previous time position are input together into the controlled differential state evolution structure. Continuous integration update is performed at the current time position. The continuous state component has a continuous driving effect on the hidden state, and the event trigger component has a trigger driving effect on the hidden state. The hidden state of the previous time position is retained under gate modulation and superimposed on the current time position to generate the hidden state of the current time position.

[0023] The hidden state at the current time position is passed to the next time position. At the next time position, gated modulation and continuous integral update are repeated. After the continuous integral update of all time positions is completed along a unified time axis, a dynamic interactive hidden state sequence is generated.

[0024] Optionally, the process of constructing the dynamic interactive response strategy set includes:

[0025] Read the dynamic interactive hidden state sequence corresponding to each time position according to a unified timeline;

[0026] Response sensitivity modulation calculation is performed on the dynamic interaction latent state at each time position to generate response priority parameters, intervention sensitivity parameters, interaction form preference parameters and feedback adaptation parameters corresponding to each time position;

[0027] At the current time position, the content display action, information prompt action, service entry trigger action and interactive feedback action are sorted according to the response priority parameter. The action at the top of the sorting results is determined as the target response action at the current time position. If there are two or more target response actions at the same time position, the execution order of the target response actions is determined according to the size of the response priority parameter.

[0028] At the current time and location, based on the intervention sensitivity parameter, a target response intervention level corresponding to the current time and location is selected from the preset response intervention level set. When the target response intervention level is the first intervention level, a single content display adjustment is performed on the target response action. When the target response intervention level is the second intervention level, the content display adjustment and information prompt triggering are performed on the target response action. When the target response intervention level is the third intervention level, the content display adjustment, information prompt triggering, and service entry triggering are performed on the target response action. The execution result corresponding to the target response intervention level is written into the target response action at the current time and location, generating the response intervention result corresponding to the current time and location.

[0029] At the current time and location, the response intervention result is matched with the interaction form preference parameters. The interaction presentation method corresponding to the current time and location is determined from the content display order adjustment method, information prompt method, service entry presentation method and interaction feedback process. Based on the feedback adaptation parameters, the feedback reception adaptation is performed on the interaction presentation method corresponding to the current time and location. The feedback receiving position, feedback recording method and feedback connection order corresponding to the interaction presentation method are determined.

[0030] The target response actions, response intervention results, interactive presentation methods, and feedback acceptance results corresponding to each time location are combined in chronological order to generate a dynamic interactive response strategy set.

[0031] Optionally, the process of generating dynamic interactive response results includes:

[0032] Read the set of dynamic interactive response strategies corresponding to the current time position according to the unified timeline, extract the content identifiers corresponding to each content item from the current e-commerce application interface, rearrange the content identifiers according to the sorting position corresponding to the target response action, and write the rearranged content identifiers back to the current e-commerce application interface to complete the content display order update.

[0033] Read the information prompt trigger command corresponding to the current time position, turn off the information prompt in the current display state, write the target information prompt into the target display area, write the trigger time of the target information prompt into the current time position, and complete the information prompt method update;

[0034] Read the service entry trigger command corresponding to the current time and location, switch the current service entry from hidden to displayed, switch the current service entry from untriggerable to triggerable, write the receiving page identifier corresponding to the current service entry into the service entry jump relationship, and complete the service entry status update;

[0035] Read the feedback acceptance instruction corresponding to the current time position, write the feedback input control into the target feedback area, write the feedback record fields into the feedback record link according to the preset record order, and write the jump order between the feedback page and the current e-commerce application interface into the feedback connection link to complete the interactive feedback process update;

[0036] The updated content display order, information prompting method, service entry status, and interactive feedback process are combined to generate a dynamic interactive response result corresponding to the current time and location.

[0037] Optionally, the process of constructing the feedback correction vector includes: aligning the subsequent click data, dwell time data, close action data, page exit data, and conversion behavior data corresponding to the dynamic interaction response results according to a unified timeline; extracting click markers, dwell time, close markers, exit markers, and conversion markers at each time position; combining the click markers, dwell time, close markers, exit markers, and conversion markers corresponding to each time position according to a preset field order to generate feedback correction records; and arranging the feedback correction records corresponding to each time position in chronological order to construct the feedback correction vector.

[0038] Optionally, the process of generating dynamic interactive response results includes:

[0039] The feedback correction vector is written back to each time position in the construction process of the dual-path coupled control sequence according to a unified time axis. At each time position, the feedback correction record corresponding to the feedback correction vector is written into the continuous state component and the event trigger component to correct the input content at each time position in the dual-path coupled control sequence.

[0040] The feedback correction vector is written back to each time position in the hidden state initialization process of the improved NCDE model according to the unified time axis. At each time position, the feedback correction record is written into the hidden state initialization content of the corresponding time position, so that the hidden state initialization process simultaneously includes the input content corresponding to the dual-path coupling control sequence and the feedback content corresponding to the feedback correction record, thereby changing the starting state of the continuous evolution and update of the hidden state.

[0041] Based on the corrected dual-path coupling control sequence and the corrected hidden state initialization content, the hidden state continuous evolution update is re-executed along the unified time axis. The continuous state components, event triggering components and feedback content at each time position jointly participate in the hidden state update at the current time position. The hidden state updated at the previous time position continues to be passed to the next time position, generating the updated dynamic interactive hidden state sequence.

[0042] Based on the updated dynamic interaction hidden state sequence, the response sensitivity modulation calculation and dynamic interaction response strategy construction are re-executed. The target response action, response intervention result, interaction presentation method and feedback acceptance result corresponding to each time position are updated synchronously with the updated dynamic interaction hidden state sequence, generating an updated dynamic interaction response strategy set. Based on the updated dynamic interaction response strategy set, dynamic adjustment is re-executed to generate an updated dynamic interaction response result.

[0043] The dynamic interactive response system for e-commerce applications based on sensor data according to an embodiment of the present invention includes the following modules:

[0044] The data acquisition and preprocessing module is used to collect data during the operation of e-commerce applications and preprocess it to generate a standardized set of interactive input data.

[0045] The dual-path coupling construction module is used to extract continuous state features and discrete event features from a standardized interactive input data set, construct a continuous sensing control path and an interactive event triggering path, perform coupling encoding, and generate a dual-path coupling control sequence.

[0046] The phase indicator generation module is used to extract phase association features from the standardized interactive input data set, perform phase association calculations, and generate interactive phase indicator vectors.

[0047] The hidden state evolution module is used to input the dual-path coupling control sequence and the interaction stage indicator vector into the improved NCDE model, perform continuous evolution and update of the hidden state, and perform gating modulation on the state update weights during the hidden state evolution process to generate a dynamic interaction hidden state sequence.

[0048] The response strategy construction module is used to perform response sensitivity modulation calculation on the dynamic interaction hidden state sequence and construct a set of dynamic interaction response strategies.

[0049] The dynamic adjustment module is used to dynamically adjust the content display order, information prompting method, service entry status and interactive feedback process of the current e-commerce application interface, and generate dynamic interactive response results.

[0050] The feedback write-back update module is used to collect data corresponding to the dynamic interaction response results, construct feedback correction vectors, write back to the dual-path coupling control sequence construction process and the hidden state initialization process of the improved NCDE model, update the dynamic interaction hidden state sequence and the dynamic interaction response strategy set, and generate dynamic interaction response results.

[0051] The beneficial effects of this invention are:

[0052] (1) This invention extracts continuous state features, discrete event features, performs stage correlation calculation and constructs dual-path coupling control sequences from the data of e-commerce application operation process, so that the continuous state change trend and the discrete event triggering influence form a collaborative representation under a unified time axis, which improves the problem of the single characterization of the user's real-time interaction process and insufficient state recognition granularity in the existing technology, and improves the completeness and accuracy of dynamic interaction state modeling.

[0053] (2) The present invention inputs the dual-path coupling control sequence and the interaction stage indicator vector into the improved NCDE model (improved neural controlled differential equation model), performs continuous evolution update of the hidden state, and performs gating modulation on the state update weights in the process of hidden state evolution, so that the model can simultaneously consider the continuous state driving, event triggering driving and stage constraint effects in the process of state evolution, improve the problem of lack of stage perception and continuous evolution linkage mechanism in the prior art, and improve the ability to identify user interaction stage migration and response timing changes;

[0054] (3) This invention constructs a set of dynamic interactive response strategies by performing response sensitivity modulation calculation on the dynamic interactive hidden state sequence, and combines the feedback correction vector back to the construction process of the dual-path coupling control sequence and the hidden state initialization process of the improved NCDE model to realize the dynamic adjustment and closed-loop update of the content display order, information prompting method, service entry state and interactive feedback process. This improves the problem of the disconnect between the response result and the subsequent state evolution in the prior art, and enhances the adaptive capability, continuous optimization capability and overall response effect of the dynamic interactive response of e-commerce applications. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is an overall flowchart of the dynamic interactive response method for e-commerce applications based on sensor data proposed in this invention;

[0057] Figure 2 This is a module connection diagram of the dynamic interactive response system for e-commerce applications based on sensor data proposed in this invention.

[0058] Figure 3 This is a schematic diagram of the improved NCDE model structure of the dynamic interactive response method for e-commerce applications based on sensor data proposed in this invention. Detailed Implementation

[0059] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0060] refer to Figures 1-3 A dynamic interactive response method for e-commerce applications based on sensor data includes the following steps:

[0061] Data from the e-commerce application's operation is collected and preprocessed to generate a standardized interactive input data set. Continuous state features and discrete event features are extracted from this standardized interactive input data set. A continuous sensing control path and an interactive event triggering path are constructed and coupled, generating a dual-path coupled control sequence to characterize the continuous state change trend and the impact of discrete event triggering. Stage correlation calculations are performed on the standardized interactive input data set to generate interactive stage indicator vectors. The dual-path coupled control sequence and interactive stage indicator vectors are input into an improved NCDE model to perform continuous evolution updates of hidden states. The state update weights during the hidden state evolution process are gated and modulated to generate a dynamic interactive hidden state sequence. The dynamic interactive hidden state sequence... The algorithm performs response sensitivity modulation calculations to generate response priority parameters, intervention sensitivity parameters, interaction form preference parameters, and feedback adaptation parameters, and constructs a dynamic interaction response strategy set. It dynamically adjusts the content display order, information prompting method, service entry status, and interaction feedback process of the current e-commerce application interface to generate dynamic interaction response results. It collects subsequent click data, dwell time data, closing action data, page exit data, and conversion behavior data corresponding to the dynamic interaction response results, constructs a feedback correction vector, and writes it back to the dual-path coupling control sequence construction process and the hidden state initialization process of the improved NCDE model. This updates the dynamic interaction hidden state sequence and the dynamic interaction response strategy set, generating dynamic interaction response results.

[0062] In this embodiment, the e-commerce application operation process data includes terminal sensor data, user operation behavior data, and application context data. The terminal sensor data includes touch pressure data, swipe displacement data, terminal posture data, and page dwell time data. The user operation behavior data includes click data, swipe data, search data, page jump data, target object access data, add to cart data, and close action data. The application context data includes current page identifier data, target object identifier data, activity status data, time segment data, and network status data. The preprocessing includes timestamp alignment, missing value completion, outlier removal, duplicate record removal, numerical normalization, and sequence sorting.

[0063] In this embodiment, the process of generating the dual-path coupling control sequence includes:

[0064] The touch pressure data, swipe displacement data, terminal posture data, and page dwell time data in the standardized interactive input data set are sorted according to timestamps. The sorted data is then segmented and sliced. The numerical changes, directions, rates of change, and durations between adjacent moments within each segment are calculated to generate a continuous state feature sequence. Segmentation refers to continuously dividing the sorted touch pressure data, swipe displacement data, terminal posture data, and page dwell time data according to a preset time window on a unified time axis, so that each data segment corresponds to a unique time interval.

[0065] The click data, swipe data, search data, page jump data, target object access data, add to cart data and close action data in the standardized interactive input data set are sorted according to the event occurrence time. The event type label, event occurrence time, time interval before and after the event, event corresponding page identifier and event corresponding target object identifier are extracted from the sorted data to generate a discrete event feature sequence.

[0066] Perform continuous-time interpolation on the continuous state feature sequence to reconstruct the continuous state change trajectory on a unified time axis, and construct a sensing continuous control path based on the reconstructed continuous state change trajectory.

[0067] Event triggering is performed on the feature sequence of discrete events, mapping each discrete event to a corresponding triggering time on a unified time axis. An event triggering increment is generated at the corresponding triggering time, and an interactive event triggering path is constructed along the time sequence based on each event triggering increment. The event triggering increment refers to writing an event tag value, event type value, page identifier value, and target object identifier value at a unique time position to represent the current occurrence state of the discrete event, so that the discrete event forms an event triggering representation on the unified time axis that can participate in subsequent coupled coding.

[0068] The sensing continuous control path and the interactive event triggering path are aligned step-by-step according to a unified timeline. At each time point, the continuous state component in the sensing continuous control path and the event triggering component in the interactive event triggering path are extracted. Position correspondence splicing and association annotation are performed to generate an initial coupling sequence. Specifically, position correspondence splicing involves combining the continuous state component and the event triggering component into the same time position according to a preset field order at the same time position on the unified timeline. Specifically, association annotation involves writing the page identifier correspondence, target object identifier correspondence, and time position correspondence into the time position record, so that the continuous state component and the event triggering component form a clear association at the same time position.

[0069] Joint mapping encoding is performed on the continuous state components and event-triggered components in the initial coupling sequence. The continuous state change trend and the discrete event triggering effect are written into the encoding result at the same time position to generate a dual-path coupling control sequence. The joint mapping encoding refers to the unified dimensional transformation of the continuous state components and event-triggered components in each time position record of the initial coupling sequence.

[0070] In this embodiment, the process of generating the interaction phase indicator vector includes:

[0071] The standardized interactive input data set is divided into segments according to a unified timeline. Within each time segment, the page dwell time, number of swipes, swipe interval, number of page jumps, number of visits to the target object, and number of feedback actions are counted. Stage correlation features representing the level of interactive activity, page switching, target object attention, and feedback response in the current time segment are extracted. The magnitude and direction of change of stage correlation features in adjacent time segments are compared in chronological order. The continuous change relationship and the transitional change relationship between adjacent time segments are calculated. The continuous change relationship refers to the correspondence that the stage correlation features between adjacent time segments maintain the same direction or change slightly in value. The transitional change relationship refers to the correspondence that the stage correlation features between adjacent time segments change direction or change in magnitude beyond a preset range in value.

[0072] Aggregation calculations are performed on stage-related features over several consecutive time periods. Continuous cumulative comparisons of stage-related features over several consecutive time periods are conducted to determine whether features of the same type appear continuously and maintain a high-frequency distribution over several adjacent time periods. The concentrated distribution state of stage-related features over a continuous time range is determined. Stage-related calculations are performed based on continuous change relationships, transition change relationships, and concentrated distribution state to determine the stage marker value corresponding to each time period. The stage marker values ​​corresponding to each time period are arranged in chronological order to generate an interactive stage indicator vector.

[0073] In this embodiment, the improved NCDE model includes the following structure:

[0074] The dual-path coupled input structure is used to receive the dual-path coupled control sequence, extract the continuous state component and event trigger component from the dual-path coupled control sequence according to each time position in the unified time axis, write the continuous state component into the sensing continuous state input branch, and write the event trigger component into the interactive event trigger input branch.

[0075] The phase-aware gating structure is used to receive the interaction phase indication vector, extract the phase marker value from the interaction phase indication vector according to each time position in the unified time axis, and generate the corresponding state update gating coefficient based on the phase marker value at each time position.

[0076] The controlled differential state evolution structure is used to receive the continuous state components in the continuous state input branch of the sensing, the event trigger components in the interactive event trigger input branch, and the state update gating coefficients output by the stage perception gating structure at each time position. The controlled differential evolution calculation is performed on the continuous state components and the event trigger components to generate a dynamic interactive hidden state sequence.

[0077] The response sensitivity modulation structure is used to receive the dynamic interaction hidden state sequence, perform response sensitivity modulation calculation on the dynamic interaction hidden state sequence according to each time position in the unified time axis, and output response priority parameters, intervention sensitivity parameters, interaction form preference parameters and feedback adaptation parameters.

[0078] The feedback write-back correction structure is used to receive the feedback correction vector corresponding to the dynamic interaction response result, and write the feedback correction vector back to the dual-path coupled input structure and the controlled differential state evolution structure according to each time position in the unified time axis, so as to correct the continuous state components, event-triggered components and dynamic interaction hidden state sequence.

[0079] In this embodiment, the process of generating a dynamic interactive hidden state sequence includes:

[0080] Read the dual-path coupling control sequence, interaction phase indicator vector and hidden state of the previous time position in sequence according to the unified time axis. At the current time position, input the continuous state component and event trigger component in the dual-path coupling control sequence into the improved NCDE model, and input the phase marker value in the interaction phase indicator vector into the phase perception gating structure.

[0081] Based on the stage marker value, the write intensity of the continuous state component at the current time position, the write intensity of the event-triggered component at the current time position, and the transmission intensity of the hidden state from the previous time position to the current time position are gated and modulated.

[0082] Specifically, at the current time position on a unified time axis, based on the corresponding stage marker value, the influence strength of the continuous state component written into the current hidden state, the influence strength of the event-triggered component written into the current hidden state, and the retention strength of the hidden state from the previous time position transmitted to the current time position are weighted and controlled respectively. When the stage marker value indicates that the current interaction stage is in a continuously changing state, the influence strength of the continuous state component is increased while the transmission strength of the hidden state from the previous time position is maintained. When the stage marker value indicates that the current interaction stage is in a sudden response state, the influence strength of the event-triggered component is increased while the retention strength of the hidden state from the previous time position is decreased. Then, the modulated continuous state component, the event-triggered component, and the hidden state from the previous time position are input together into the controlled differential state evolution structure to complete the hidden state update at the current time position.

[0083] The gated modulation continuous state component, the event trigger component, and the hidden state of the previous time position are input together into the controlled differential state evolution structure. Continuous integration update is performed at the current time position. The continuous state component has a continuous driving effect on the hidden state, and the event trigger component has a trigger driving effect on the hidden state. The hidden state of the previous time position is retained under gate modulation and superimposed on the current time position to generate the hidden state of the current time position.

[0084] The hidden state at the current time position is passed to the next time position. At the next time position, gated modulation and continuous integral update are repeated. After the continuous integral update of all time positions is completed along a unified time axis, a dynamic interactive hidden state sequence is generated.

[0085] In this embodiment, the process of constructing a dynamic interactive response strategy set includes:

[0086] Read the dynamic interactive hidden state sequence corresponding to each time position according to a unified timeline;

[0087] Response sensitivity modulation calculation is performed on the dynamic interaction latent state at each time position to generate response priority parameters, intervention sensitivity parameters, interaction form preference parameters and feedback adaptation parameters corresponding to each time position;

[0088] In this embodiment, the response sensitivity modulation calculation specifically involves: sequentially reading the dynamic interaction latent states corresponding to each time position according to a unified time axis; performing state decomposition on the dynamic interaction latent states at the current time position to extract content attention components, interaction activity components, event-driven components, and feedback acceptance components; combining and comparing the content attention components and interaction activity components to determine the urgency of responding to each candidate response action at the current time position; jointly judging the interaction activity components and event-driven components to determine the response intervention level corresponding to the current time position; matching and judging the content attention components and event-driven components to determine the interaction presentation mode corresponding to the current time position; adapting and judging the feedback acceptance component and the interaction presentation mode corresponding to the current time position to determine the feedback connection mode corresponding to the current time position; generating response priority parameters based on the determined response urgency, generating intervention sensitivity parameters based on the determined response intervention level, generating interaction form preference parameters based on the determined interaction presentation mode, and generating feedback adaptation parameters based on the determined feedback connection mode.

[0089] At the current time position, the content display action, information prompt action, service entry trigger action and interactive feedback action are sorted according to the response priority parameter. The action at the top of the sorting results is determined as the target response action at the current time position. If there are two or more target response actions at the same time position, the execution order of the target response actions is determined according to the size of the response priority parameter.

[0090] At the current time and location, based on the intervention sensitivity parameter, a target response intervention level corresponding to the current time and location is selected from the preset response intervention level set. When the target response intervention level is the first intervention level, a single content display adjustment is performed on the target response action. When the target response intervention level is the second intervention level, the content display adjustment and information prompt triggering are performed on the target response action. When the target response intervention level is the third intervention level, the content display adjustment, information prompt triggering, and service entry triggering are performed on the target response action. The execution result corresponding to the target response intervention level is written into the target response action at the current time and location, generating the response intervention result corresponding to the current time and location.

[0091] At the current time and location, the response intervention result is matched with the interaction form preference parameters. The interaction presentation method corresponding to the current time and location is determined from the content display order adjustment method, information prompt method, service entry presentation method and interaction feedback process. Based on the feedback adaptation parameters, the feedback reception adaptation is performed on the interaction presentation method corresponding to the current time and location. The feedback receiving position, feedback recording method and feedback connection order corresponding to the interaction presentation method are determined.

[0092] The target response actions, response intervention results, interactive presentation methods, and feedback acceptance results corresponding to each time location are combined in chronological order to generate a dynamic interactive response strategy set.

[0093] In this embodiment, the process of generating dynamic interactive response results includes:

[0094] Read the set of dynamic interactive response strategies corresponding to the current time position according to the unified timeline, extract the content identifiers corresponding to each content item from the current e-commerce application interface, rearrange the content identifiers according to the sorting position corresponding to the target response action, and write the rearranged content identifiers back to the current e-commerce application interface to complete the content display order update.

[0095] Read the information prompt trigger command corresponding to the current time position, turn off the information prompt in the current display state, write the target information prompt into the target display area, write the trigger time of the target information prompt into the current time position, and complete the information prompt method update;

[0096] Read the service entry trigger command corresponding to the current time and location, switch the current service entry from hidden to displayed, switch the current service entry from untriggerable to triggerable, write the receiving page identifier corresponding to the current service entry into the service entry jump relationship, and complete the service entry status update;

[0097] Read the feedback acceptance instruction corresponding to the current time position, write the feedback input control into the target feedback area, write the feedback record fields into the feedback record link according to the preset record order, and write the jump order between the feedback page and the current e-commerce application interface into the feedback connection link to complete the interactive feedback process update;

[0098] The updated content display order, information prompting method, service entry status, and interactive feedback process are combined to generate a dynamic interactive response result corresponding to the current time and location.

[0099] In this embodiment, the process of constructing the feedback correction vector includes: aligning the subsequent click data, dwell time data, close action data, page exit data, and conversion behavior data corresponding to the dynamic interaction response results according to a unified timeline; extracting click markers, dwell time, close markers, exit markers, and conversion markers at each time position; combining the click markers, dwell time, close markers, exit markers, and conversion markers corresponding to each time position according to a preset field order to generate feedback correction records; and arranging the feedback correction records corresponding to each time position in chronological order to construct the feedback correction vector.

[0100] In this embodiment, the process of generating dynamic interactive response results includes:

[0101] The feedback correction vector is written back to each time position in the construction process of the dual-path coupled control sequence according to a unified time axis. At each time position, the feedback correction record corresponding to the feedback correction vector is written into the continuous state component and the event trigger component to correct the input content at each time position in the dual-path coupled control sequence.

[0102] The feedback correction vector is written back to each time position in the hidden state initialization process of the improved NCDE model according to the unified time axis. At each time position, the feedback correction record is written into the hidden state initialization content of the corresponding time position, so that the hidden state initialization process simultaneously includes the input content corresponding to the dual-path coupling control sequence and the feedback content corresponding to the feedback correction record, thereby changing the starting state of the continuous evolution and update of the hidden state.

[0103] Based on the corrected dual-path coupling control sequence and the corrected hidden state initialization content, the hidden state continuous evolution update is re-executed along the unified time axis. The continuous state components, event triggering components and feedback content at each time position jointly participate in the hidden state update at the current time position. The hidden state updated at the previous time position continues to be passed to the next time position, generating the updated dynamic interactive hidden state sequence.

[0104] Based on the updated dynamic interaction hidden state sequence, the response sensitivity modulation calculation and dynamic interaction response strategy construction are re-executed. The target response action, response intervention result, interaction presentation method and feedback acceptance result corresponding to each time position are updated synchronously with the updated dynamic interaction hidden state sequence, generating an updated dynamic interaction response strategy set. Based on the updated dynamic interaction response strategy set, dynamic adjustment is re-executed to generate an updated dynamic interaction response result.

[0105] In this embodiment, the new dynamic adjustment is specifically performed by sequentially reading the updated dynamic interactive response strategy set corresponding to each time position according to a unified timeline. At the current time position, the target content item in the current e-commerce application interface is redefined based on the updated target response action, and the content identifier corresponding to the target content item is written into the preset display position to complete the rearrangement of the content display order. Based on the updated response intervention result, the corresponding information prompt triggering command, service entry triggering command, and interactive feedback triggering command are read. At the current time position, the original mismatched prompt content, service entry status, and feedback process nodes are closed, and the updated prompt content, service entry display status, service entry jump relationship, feedback input position, and feedback connection order are written. Based on the updated interactive presentation method and feedback acceptance result, the rearranged content display result, updated information prompt result, updated service entry status result, and updated interactive feedback process result are synchronously written into the current e-commerce application interface to complete the dynamic adjustment at the current time position. This process is repeated along the unified timeline to generate the updated dynamic interactive response result.

[0106] The dynamic interactive response system for e-commerce applications based on sensor data according to an embodiment of the present invention includes the following modules:

[0107] The data acquisition and preprocessing module is used to collect data during the operation of e-commerce applications and preprocess it to generate a standardized set of interactive input data.

[0108] The dual-path coupling construction module is used to extract continuous state features and discrete event features from a standardized interactive input data set, construct a continuous sensing control path and an interactive event triggering path, perform coupling encoding, and generate a dual-path coupling control sequence.

[0109] The phase indicator generation module is used to extract phase association features from the standardized interactive input data set, perform phase association calculations, and generate interactive phase indicator vectors.

[0110] The hidden state evolution module is used to input the dual-path coupling control sequence and the interaction stage indicator vector into the improved NCDE model, perform continuous evolution and update of the hidden state, and perform gating modulation on the state update weights during the hidden state evolution process to generate a dynamic interaction hidden state sequence.

[0111] The response strategy construction module is used to perform response sensitivity modulation calculation on the dynamic interaction hidden state sequence and construct a set of dynamic interaction response strategies.

[0112] The dynamic adjustment module is used to dynamically adjust the content display order, information prompting method, service entry status and interactive feedback process of the current e-commerce application interface, and generate dynamic interactive response results.

[0113] The feedback write-back update module is used to collect data corresponding to the dynamic interaction response results, construct feedback correction vectors, write back to the dual-path coupling control sequence construction process and the hidden state initialization process of the improved NCDE model, update the dynamic interaction hidden state sequence and the dynamic interaction response strategy set, and generate dynamic interaction response results.

[0114] Example 1: To verify the feasibility of this invention in practice, it was applied to a real-time interactive response scenario in a mobile e-commerce application. This scenario involves continuous interactive processes such as product browsing, details viewing, information prompts, service entry triggering, and feedback reception. Under existing processing methods, the e-commerce interface's perception of user state changes mainly relies on click records, dwell time records, and historical browsing records. Interface responses often follow pre-set fixed rules. When a user first enters the interface, the system has already triggered a prompt; when the user is in a rapid swiping state, the system continues to pop up service entry points; when the user returns to the same target page multiple times, the content display order and prompt method remain unchanged. This processing method makes it difficult to distinguish between the differences brought about by continuous state changes and discrete event triggers, easily leading to problems such as premature prompting, content arrangement not matching the current focus, and unsmooth service entry reception, resulting in decreased dwell time, increased closing actions, increased page exit rate, and insufficient conversion behavior.

[0115] In this embodiment, the terminal continuously collects data on the operation of the e-commerce application, forming a standardized set of interactive input data. The collected content covers changes in touch pressure, changes in swipe displacement, changes in terminal posture, page dwell time, click actions, page jump actions, target object access actions, adding to cart actions, closing actions, as well as the current page identifier and target object identifier. After entering the processing chain, the system does not directly respond to the page based on the raw records. Instead, it first extracts continuous state features and discrete event features from the standardized interactive input data set. Touch pressure, swipe displacement, terminal posture, and dwell changes are constructed into a continuous sensing control path, and click, jump, access, and close actions are constructed into interactive event trigger paths. Then, a dual-path coupled control sequence is formed through coupling encoding. After this processing, the system can simultaneously see two types of information: "the user is continuously browsing" and "the user has just triggered a key action," avoiding one-sided judgments based on a single click or dwell time.

[0116] As the interface interaction continues, the system further performs stage-based correlation calculations on the standardized set of interactive input data. Page dwell time, swipe frequency, page jump frequency, target object revisit behavior, and generated feedback data are uniformly mapped onto the timeline, thereby generating an interaction stage indicator vector. After this processing, the quick scan stage, comparison and filtering stage, attention and focus stage, and response and acceptance stage can be continuously distinguished. For the same detail page revisit action, in the quick scan stage, the system judges it as a continuation of ordinary browsing; in the attention and focus stage, the system judges it as continuous attention to a specific target object. This stage differentiation capability solves the problem of existing technologies lacking hierarchical characterization of user states.

[0117] After constructing the dual-path coupling control sequence and the interaction phase indicator vector, the system inputs both into the improved NCDE model. The improved NCDE model does not simply perform ordinary time-series calculations; instead, it simultaneously incorporates continuous state components, event trigger components, and phase marker values ​​during the continuous evolution of the hidden state. Continuous state components reflect whether the browsing rhythm is stable, whether the dwell time is concentrated, and whether the operations are frequent. Event trigger components reflect whether the user has just performed a page jump, accessed a target object, closed an action, or added an item to the shopping cart. The phase marker value controls the state update weight at the current time position, ensuring that the hidden state update is consistent with the changes in the interaction phase. After continuous evolution, the dynamic interaction hidden state sequence is output to the response sensitivity modulation calculation process. The response sensitivity modulation calculation generates response priority parameters, intervention sensitivity parameters, interaction form preference parameters, and feedback adaptation parameters, based on which a dynamic interaction response strategy set is constructed. In practical applications, the system will rearrange the order of content on the interface based on the current results, prioritizing the display of the target objects, related attributes, and connecting information that the user is continuously focusing on; switch the information prompting method, turning off mismatched prompts and writing prompts that are more suitable for the current state into the target area; switch the service entry status, opening the service entry at the appropriate stage and updating the entry position synchronously with the subsequent feedback chain; and adjust the interaction feedback process so that after the user clicks, closes, stays, or converts, the feedback can directly flow back to the status update process at the corresponding time position.

[0118] To more realistically demonstrate the problem-solving capabilities of this invention, this embodiment selects high-frequency browsing, detail page revisit, service entry point access, and prompt information triggering as key interaction scenarios for continuous verification. Two comparison groups were set up: one group used a fixed-rule response method, adjusting the page solely based on historical behavior statistics; the other group used the method of this invention, employing dynamic interactive responses based on a dual-path coupling control sequence, interaction stage indicator vectors, and an improved NCDE model. Continuous observation results show that with the fixed-rule response method, users are frequently interrupted during high-speed scrolling, resulting in numerous prompts to close the page, a low revisit rate of the target object's detail page, and a low rate of continued browsing after service entry point triggering. Using the method of this invention, the interface more accurately distinguishes between rapid scanning and focused states, the content display order is closer to the current focus, information prompts are more concentrated in the dwell enhancement and revisit enhancement stages, the service entry point is more natural, and the feedback connection chain is shorter. The data results are shown in Table 1 below.

[0119] Table 1: Comparison of Dynamic Interactive Response Effects

[0120] Comparison indicators Fixed rule response method The method of the present invention is dynamically adjusted for the first time. The method of this invention provides feedback and write-back. Improvement status Average page dwell time 18.6 26.1 27.4 Continuous improvement Target object revisit rate 21.8% 31.9% 34.6% Significant improvement Tip close rate 31.2% 17.5% 14.9% Significant decline Page exit rate 22.7% 15.1% 13.5% Significant decline Service entry click-through rate 16.4% 26.7% 29.8% Significant improvement Details page depth browsing ratio 26.1% 38.4% 41.7% Significant improvement Add to cart conversion rate 8.9% 13.6% 14.7% Continuous improvement Submit conversion rate 3.8% 5.9% 6.5% Continuous improvement Target content click-through rate 12.7% 19.8% 21.4% Significant improvement Feedback Link Completion Rate 54.3% 68.9% 72.6% Significant improvement

[0121] As shown in Table 1, the interactive effect of the method of this invention is superior to the fixed rule response method in multiple dimensions, including dwell time, connection, conversion, and feedback loop. The average page dwell time increased from 18.6 to 26.1, and further increased to 27.4 after feedback writing, indicating that the invention can more accurately match the user's current interaction state, making the interface content and prompt triggers more closely match the user's attention process, reducing invalid interruptions, and enhancing the willingness to continue browsing.

[0122] In terms of user engagement retention, the secondary visit rate of the target object increased from 21.8% to 31.9%, reaching 34.6% after feedback; the proportion of in-depth browsing of the details page increased from 26.1% to 38.4%, further increasing to 41.7%; and the initial click-through rate of the target content increased from 12.7% to 19.8%, reaching 21.4% subsequently. These data indicate that by jointly modeling the continuous changes in user state and discrete event triggers through a dual-path coupling control sequence, interaction stage indicator vector, and improved NCDE model, this invention can more effectively identify the user's focus process on the target object and enhance the sustained attractiveness of the target object by adjusting the content display order to present highly relevant content first.

[0123] From the perspective of interface intervention quality, the prompt closure rate decreased from 31.2% to 17.5%, and further decreased to 14.9% after feedback was written back; the page exit rate decreased from 22.7% to 15.1%, and further decreased to 13.5%. This change indicates that, under the combined effect of response priority parameters, intervention sensitivity parameters, interaction form preference parameters, and feedback adaptation parameters, this invention can adjust the information prompting method, service entry status, and interaction feedback process to a more appropriate time position, reducing the interruption caused to users by mismatched prompts and reducing interface interruptions and direct exits.

[0124] From the perspective of service acceptance and conversion effects, the service entry click acceptance rate increased from 16.4% to 26.7%, reaching 29.8% after feedback; the add-to-cart conversion rate increased from 8.9% to 13.6%, reaching 14.7% subsequently; and the submission conversion rate increased from 3.8% to 5.9%, reaching 6.5% subsequently. These results demonstrate that this invention not only improves the dynamic responsiveness of the front-end interface but also enhances the overall efficiency of the link from interface touchpoint to service acceptance and conversion completion. The service entry point is triggered at a more appropriate stage, and the target action is presented in a more suitable interactive format, making it easier for users to transition from browsing to acceptance and conversion states.

[0125] From the perspective of feedback loop capability, the feedback link completion rate increased from 54.3% to 68.9%, and further increased to 72.6% after feedback write-back. This indicator shows that the present invention does not complete the interface adjustment in one go, but can construct a feedback correction vector based on subsequent click data, dwell time data, close action data, page exit data, and conversion behavior data. The feedback results continue to affect the construction process of the dual-path coupled control sequence and the hidden state initialization process of the improved NCDE model, thereby continuously updating the subsequent dynamic interaction hidden state sequence and dynamic interaction response strategy set. The comparison of the two sets of data in the table, one for the initial dynamic adjustment and one for feedback write-back, further shows that the present invention has continuous optimization capabilities, and can continue to reduce the close rate and exit rate, and continue to improve the acceptance rate and conversion rate on the basis of the existing response, demonstrating a clear closed-loop adaptive advantage.

[0126] Overall, the data shown in Table 1 fully demonstrates that the present invention has achieved significant results in solving problems such as inaccurate response timing, insufficient stage identification, mismatched prompts and interventions, unsmooth service entry connection, and insufficient utilization of feedback in the prior art. It can improve the precision of dynamic interactive response and the actual use effect of e-commerce applications.

[0127] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A dynamic interactive response method for e-commerce applications based on sensor data, characterized in that, Includes the following steps: Collect data from the e-commerce application's operation process, preprocess it to generate a standardized interactive input data set; extract continuous state features and discrete event features from the standardized interactive input data set, construct a continuous sensing control path and an interactive event triggering path, perform coupled encoding, and generate a dual-path coupled control sequence; perform stage association calculation on the standardized interactive input data set to generate an interactive stage indicator vector; The dual-path coupling control sequence and the interaction stage indicator vector are input into the improved NCDE model to perform continuous evolution and update of the hidden state. The state update weights in the process of hidden state evolution are gated and modulated to generate a dynamic interaction hidden state sequence. The system performs response sensitivity modulation calculations on the dynamic interaction latent state sequence to generate response priority parameters, intervention sensitivity parameters, interaction form preference parameters, and feedback adaptation parameters, thus constructing a dynamic interaction response strategy set. It dynamically adjusts the content display order, information prompting method, service entry status, and interaction feedback process of the current e-commerce application interface to generate dynamic interaction response results. It collects subsequent click data, dwell time data, closing action data, page exit data, and conversion behavior data corresponding to the dynamic interaction response results, constructs a feedback correction vector, and writes it back to the dual-path coupling control sequence construction process and the latent state initialization process of the improved NCDE model. This updates the dynamic interaction latent state sequence and the dynamic interaction response strategy set, generating the dynamic interaction response results.

2. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The e-commerce application operation process data includes terminal sensor data, user operation behavior data, and application context data. The preprocessing includes timestamp alignment, missing value completion, outlier removal, duplicate record removal, numerical normalization, and sequence sorting.

3. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The process of generating the dual-path coupling control sequence includes: The touch pressure data, sliding displacement data, terminal posture data and page dwell time data in the standardized interactive input data set are sorted according to timestamps. The sorted data is then segmented and sliced. The numerical change, direction of change, rate of change and duration between adjacent moments in each segment are calculated to generate a continuous state feature sequence. The click data, swipe data, search data, page jump data, target object access data, add to cart data and close action data in the standardized interactive input data set are sorted according to the event occurrence time. The event type label, event occurrence time, time interval before and after the event, event corresponding page identifier and event corresponding target object identifier are extracted from the sorted data to generate a discrete event feature sequence. Perform continuous-time interpolation on the continuous state feature sequence to reconstruct the continuous state change trajectory on a unified time axis, and construct a sensing continuous control path based on the reconstructed continuous state change trajectory. Perform event trigger localization on discrete event feature sequences, map each discrete event to the corresponding trigger time in a unified time axis, generate event trigger increments at the corresponding trigger times, and construct interactive event trigger paths in chronological order based on each event trigger increment; Align the continuous control path of the sensor and the triggering path of the interactive event according to a unified time axis. Extract the continuous state component in the continuous control path of the sensor and the event triggering component in the triggering path of the interactive event at each time position. Execute the corresponding positions and splice and associate them to generate the initial coupling sequence. Joint mapping encoding is performed on the continuous state components and event-triggered components in the initial coupling sequence. The continuous state change trend and the discrete event triggering effect are written into the encoding result at the same time position to generate a dual-path coupling control sequence.

4. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The process of generating the interaction phase indicator vector includes: The standardized interactive input data set is divided into segments according to a unified timeline. Within each time segment, the page dwell time, number of swipes, swipe interval, number of page jumps, number of visits to the target object, and number of feedback actions are counted. The stage association features that represent the level of interactive activity, page switching, target object attention, and feedback response in the current time segment are extracted. The magnitude and direction of change of the stage association features of adjacent time segments are compared in chronological order, and the continuous change relationship and transition relationship between adjacent time segments are calculated. Aggregation calculations are performed on the stage-related features over several consecutive time periods to determine the concentrated distribution of the stage-related features within the continuous time range. Based on the continuous change relationship, transition change relationship and concentrated distribution, stage-related calculations are performed to determine the stage marker value corresponding to each time period. The stage marker values ​​corresponding to each time period are arranged in chronological order to generate an interactive stage indicator vector.

5. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The process of generating the dynamic interactive hidden state sequence includes: Read the dual-path coupling control sequence, interaction phase indicator vector and hidden state of the previous time position in sequence according to the unified time axis. At the current time position, input the continuous state component and event trigger component in the dual-path coupling control sequence into the improved NCDE model, and input the phase marker value in the interaction phase indicator vector into the phase perception gating structure. Based on the stage marker value, the write intensity of the continuous state component at the current time position, the write intensity of the event-triggered component at the current time position, and the transmission intensity of the hidden state from the previous time position to the current time position are gated and modulated. The gated modulation continuous state component, the event trigger component, and the hidden state of the previous time position are input together into the controlled differential state evolution structure. Continuous integration update is performed at the current time position. The continuous state component has a continuous driving effect on the hidden state, and the event trigger component has a trigger driving effect on the hidden state. The hidden state of the previous time position is retained under gate modulation and superimposed on the current time position to generate the hidden state of the current time position. The hidden state at the current time position is passed to the next time position. At the next time position, gated modulation and continuous integral update are repeated. After the continuous integral update of all time positions is completed along a unified time axis, a dynamic interactive hidden state sequence is generated.

6. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The process of constructing the dynamic interactive response strategy set includes: Read the dynamic interactive hidden state sequence corresponding to each time position according to a unified timeline; Response sensitivity modulation calculation is performed on the dynamic interaction latent state at each time position to generate response priority parameters, intervention sensitivity parameters, interaction form preference parameters and feedback adaptation parameters corresponding to each time position; At the current time position, the content display action, information prompt action, service entry trigger action and interactive feedback action are sorted according to the response priority parameter. The action at the top of the sorting results is determined as the target response action at the current time position. If there are two or more target response actions at the same time position, the execution order of the target response actions is determined according to the size of the response priority parameter. At the current time and location, based on the intervention sensitivity parameter, a target response intervention level corresponding to the current time and location is selected from the preset response intervention level set. When the target response intervention level is the first intervention level, a single content display adjustment is performed on the target response action. When the target response intervention level is the second intervention level, the content display adjustment and information prompt triggering are performed on the target response action. When the target response intervention level is the third intervention level, the content display adjustment, information prompt triggering, and service entry triggering are performed on the target response action. The execution result corresponding to the target response intervention level is written into the target response action at the current time and location, generating the response intervention result corresponding to the current time and location. At the current time and location, the response intervention result is matched with the interaction form preference parameters. The interaction presentation method corresponding to the current time and location is determined from the content display order adjustment method, information prompt method, service entry presentation method and interaction feedback process. Based on the feedback adaptation parameters, the feedback reception adaptation is performed on the interaction presentation method corresponding to the current time and location. The feedback receiving position, feedback recording method and feedback connection order corresponding to the interaction presentation method are determined. The target response actions, response intervention results, interactive presentation methods, and feedback acceptance results corresponding to each time location are combined in chronological order to generate a dynamic interactive response strategy set.

7. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The process of generating dynamic interactive response results includes: Read the set of dynamic interactive response strategies corresponding to the current time position according to the unified timeline, extract the content identifiers corresponding to each content item from the current e-commerce application interface, rearrange the content identifiers according to the sorting position corresponding to the target response action, and write the rearranged content identifiers back to the current e-commerce application interface to complete the content display order update. Read the information prompt trigger command corresponding to the current time position, turn off the information prompt in the current display state, write the target information prompt into the target display area, write the trigger time of the target information prompt into the current time position, and complete the information prompt method update; Read the service entry trigger command corresponding to the current time and location, switch the current service entry from hidden to displayed, switch the current service entry from untriggerable to triggerable, write the receiving page identifier corresponding to the current service entry into the service entry jump relationship, and complete the service entry status update; Read the feedback acceptance instruction corresponding to the current time position, write the feedback input control into the target feedback area, write the feedback record fields into the feedback record link according to the preset record order, and write the jump order between the feedback page and the current e-commerce application interface into the feedback connection link to complete the interactive feedback process update; The updated content display order, information prompting method, service entry status, and interactive feedback process are combined to generate a dynamic interactive response result corresponding to the current time and location.

8. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The process of constructing the feedback correction vector includes: aligning the subsequent click data, dwell time data, close action data, page exit data, and conversion behavior data corresponding to the dynamic interaction response results according to a unified timeline; extracting click markers, dwell time, close markers, exit markers, and conversion markers at each time position; combining the click markers, dwell time, close markers, exit markers, and conversion markers corresponding to each time position according to a preset field order to generate feedback correction records; and arranging the feedback correction records corresponding to each time position in chronological order to construct the feedback correction vector.

9. The dynamic interactive response method for e-commerce applications based on sensor data according to claim 1, characterized in that, The process of generating dynamic interactive response results includes: The feedback correction vector is written back to each time position in the construction process of the dual-path coupled control sequence according to a unified time axis. At each time position, the feedback correction record corresponding to the feedback correction vector is written into the continuous state component and the event trigger component to correct the input content at each time position in the dual-path coupled control sequence. The feedback correction vector is written back to each time position in the hidden state initialization process of the improved NCDE model according to the unified time axis. At each time position, the feedback correction record is written into the hidden state initialization content of the corresponding time position, so that the hidden state initialization process simultaneously includes the input content corresponding to the dual-path coupling control sequence and the feedback content corresponding to the feedback correction record, thereby changing the starting state of the continuous evolution and update of the hidden state. Based on the corrected dual-path coupling control sequence and the corrected hidden state initialization content, the hidden state continuous evolution update is re-executed along the unified time axis. The continuous state components, event triggering components and feedback content at each time position jointly participate in the hidden state update at the current time position. The hidden state updated at the previous time position continues to be passed to the next time position, generating the updated dynamic interactive hidden state sequence. Based on the updated dynamic interaction hidden state sequence, the response sensitivity modulation calculation and dynamic interaction response strategy construction are re-executed. The target response action, response intervention result, interaction presentation method and feedback acceptance result corresponding to each time position are updated synchronously with the updated dynamic interaction hidden state sequence, generating an updated dynamic interaction response strategy set. Based on the updated dynamic interaction response strategy set, dynamic adjustment is re-executed to generate an updated dynamic interaction response result.

10. A dynamic interactive response system for e-commerce applications based on sensor data, applied to the dynamic interactive response method for e-commerce applications based on sensor data as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect data during the operation of e-commerce applications and preprocess it to generate a standardized set of interactive input data. The dual-path coupling construction module is used to extract continuous state features and discrete event features from a standardized interactive input data set, construct a continuous sensing control path and an interactive event triggering path, perform coupling encoding, and generate a dual-path coupling control sequence. The phase indicator generation module is used to extract phase association features from the standardized interactive input data set, perform phase association calculations, and generate interactive phase indicator vectors. The hidden state evolution module is used to input the dual-path coupling control sequence and the interaction stage indicator vector into the improved NCDE model, perform continuous evolution and update of the hidden state, and perform gating modulation on the state update weights during the hidden state evolution process to generate a dynamic interaction hidden state sequence. The response strategy construction module is used to perform response sensitivity modulation calculation on the dynamic interaction hidden state sequence and construct a set of dynamic interaction response strategies. The dynamic adjustment module is used to dynamically adjust the content display order, information prompting method, service entry status and interactive feedback process of the current e-commerce application interface, and generate dynamic interactive response results. The feedback write-back update module is used to collect data corresponding to the dynamic interaction response results, construct feedback correction vectors, write back to the dual-path coupling control sequence construction process and the hidden state initialization process of the improved NCDE model, update the dynamic interaction hidden state sequence and the dynamic interaction response strategy set, and generate dynamic interaction response results.

Images