Method and system for switching content of landing page based on human-computer interaction

Abstract

The present application relates to the field of computer software and information processing, and discloses a landing page content switching method and system based on human-computer interaction; including content block splitting of a target landing page, building a content block topology graph containing dependency relationship, mutual exclusion relationship and rollback pointer; collecting session interaction events and performing stabilization processing to obtain a stable interaction state feature vector; generating candidate switching content blocks based on the intention recognition result, and performing rendering layer screening in combination with terminal rendering load, expected network delay and page memory occupation; performing preloading and shadow rendering on the target switching content block, and performing resource integrity checking and structure consistency checking; performing atomic commit switching after the checking, and keeping the form state, scroll anchor point and input focus continuous; triggering rollback when switching timeout, checking failure, frame rate threshold value or user return, and restoring the previous stable content block version; recording the behavior results before and after switching and updating the switching strategy parameters.

Description

Technical Field

[0001] This invention relates to the field of computer software and information processing, specifically to a method and system for switching landing page content based on human-computer interaction. Background Technology

[0002] As the online customer acquisition and conversion process continues to move forward, landing pages have evolved from simple static display pages into conversational entry points that integrate content presentation, behavior collection, and interactive guidance. A user's actions during a single visit, such as clicking, swiping, pausing, inputting, and returning, exhibit distinct stages. Different stages have varying requirements for information density, component combinations, interaction paths, and loading strategies. Therefore, landing page content typically needs to be dynamically switched based on the conversation state.

[0003] Existing solutions primarily employ fixed-rule switching, A / B filtering, behavior-based recommendation sorting, or templated component replacement to update page content. Some solutions incorporate lazy loading, resource caching, asynchronous requests, and front-end tracking to improve response speed and statistical capabilities. While these technologies have a foundation in content matching and delivery optimization, most focus on click-through rate or conversion rate targets, with relatively insufficient consideration given to execution constraints during the switching process, such as rendering load, network latency, memory usage, resource consistency, state continuity, and abnormal rollback.

[0004] Under complex network and multi-terminal conditions, existing technologies are prone to problems such as frequent switching caused by trigger jitter, discontinuity of form and focus state after switching, and unclear recovery path in case of anomalies. The root cause is the lack of a collaborative control mechanism that unifies the judgment of interaction intent with the executability of rendering, the verification of submission, and the rollback of anomalies, resulting in a disconnect between the content decision results and the stable execution of the terminal. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a landing page content switching method and system based on human-computer interaction, thereby solving the technical problem of the lack of a collaborative switching control mechanism that takes into account session intent recognition, terminal executability constraints, and abnormal rollback recovery in existing technologies.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0007] Methods for switching landing page content based on human-computer interaction include:

[0008] S1: Break down the target landing page into content blocks, construct a content block topology diagram containing dependencies, mutual exclusions and rollback pointers, and create a resource list and version identifier for each content block;

[0009] S2: Collect human-computer interaction events of the current session. Human-computer interaction events include at least click events, swipe events, pause events, input events, and return events, and sort them according to a unified timestamp to form an interaction event stream;

[0010] S3: Perform stabilization processing on the interactive event stream to obtain a stable interactive state feature vector for intent recognition;

[0011] S4: Input the stable interaction state feature vector into the intent recognition model, output the session intent category and intent confidence, and generate a set of candidate switching content blocks based on the session intent category;

[0012] S5: Perform rendering layer filtering on the candidate switching content block set based on terminal rendering load, expected network latency and page memory usage to obtain the target switching content block; when the intent confidence is lower than the first threshold or the terminal rendering load is higher than the second threshold, execute the degradation strategy. The degradation strategy is to keep the currently displayed content block from switching and pause the submission of candidate switching content blocks until the next stable interaction state feature vector is generated and then the rendering layer filtering is re-executed.

[0013] S6: Perform preloading and shadow rendering on the target switching content block, and perform resource integrity verification and structural consistency verification;

[0014] S7: When both resource integrity and structural consistency checks pass, perform an atomic commit switch, replacing the currently displayed content block with the target switch content block in one go, while maintaining the continuity of form state, scroll anchors, and input focus;

[0015] S8: When a switching timeout, rendering failure, or user return operation triggers a rollback condition, restore to the previous stable content block version according to the rollback pointer;

[0016] S9: Record the interaction state feature vector before switching, the switching action, and the user behavior results after switching, generate causal attribution records, and update the switching strategy parameters for subsequent sessions.

[0017] Preferably, the stabilization process includes:

[0018] Aggregate the interactive event stream within a preset time window threshold;

[0019] When the same event type is repeatedly triggered within a preset minimum trigger interval threshold, subsequent triggering events in the interactive event stream are suppressed.

[0020] Sliding events with displacement less than a preset displacement threshold and dwell events with dwell time less than a preset dwell time threshold in the interactive event stream are identified as noise events and filtered out.

[0021] Preferably, inputting the stable interaction state feature vector into the intent recognition model includes:

[0022] The event type, event time interval, sliding speed, input length, and number of returns are used to construct a temporal feature vector;

[0023] The intent recognition model outputs the probability value corresponding to each session intent category;

[0024] The session intent category corresponding to the highest probability is determined as the current session intent category, and the highest probability is determined as the intent confidence level.

[0025] Preferably, generating a set of candidate switching content blocks based on the session intent category includes:

[0026] Extract content blocks that satisfy the dependency relationship from the set of successor content blocks in the content block topology graph of the currently displayed content block;

[0027] Remove content blocks that are mutually exclusive with the currently displayed content block;

[0028] The candidate switching content block set is obtained by filtering according to the session intent category and the content block mapping table.

[0029] Preferably, the step of performing rendering layer filtering based on terminal rendering load, expected network latency, and page memory usage includes:

[0030] Calculate the estimated rendering time, estimated memory increment, and estimated network latency for each candidate content block to be switched.

[0031] Candidate switching content blocks with expected rendering time not exceeding the rendering time threshold, expected memory increment not exceeding the memory threshold, and expected network latency not exceeding the latency threshold are identified as executable content blocks;

[0032] In the executable content block, a comprehensive score is calculated based on intent confidence, estimated rendering time, estimated memory increment, and estimated network latency, and the executable content block with the highest comprehensive score is selected as the target switching content block.

[0033] Preferably, the preloading and shadow rendering include:

[0034] Write the static resources corresponding to the target switching content block to the local cache.

[0035] Write the dynamic data request result corresponding to the target switching content block to the session cache.

[0036] After the target switching content block structure tree is constructed in the shadow rendering area, resource check code comparison is performed, and structural consistency verification is performed based on the structure tree version identifier and data field pattern identifier.

[0037] Preferably, the execution of the atomic commit switch includes:

[0038] Establish the current display buffer and the target display buffer;

[0039] After the target display buffer has completed full rendering, the current display buffer is replaced by a single commit operation within the same rendering frame;

[0040] After the replacement is complete, restore the form state snapshot, scroll anchor marker, and input focus marker saved before the switch.

[0041] Preferably, the rollback condition includes any of the following conditions:

[0042] The execution duration of the switch exceeds the timeout threshold;

[0043] Resource integrity check failed, or structural consistency check failed;

[0044] The frame drop rate obtained within a consecutive preset number of frames after the switch is greater than the frame drop threshold.

[0045] A user return operation was detected within the preset rollback observation window;

[0046] When the rollback condition is triggered, press the rollback pointer to restore the previous stable content block version and restore the corresponding state snapshot.

[0047] Preferably, the step of generating causal attribution records and updating the switching strategy parameters for subsequent sessions includes:

[0048] Assign a unique action identifier to each switching action;

[0049] Record the interaction state feature vector before switching, the dwell time after switching, the preset target behavior completion mark and return mark. The preset target behavior completion mark is either the form submission completion mark or the button click completion mark.

[0050] Incremental parameter updates are performed based on the sample set associated with action identifiers to obtain updated switching strategy parameters, which are then used for selecting target switching content blocks in subsequent sessions.

[0051] The updated switching strategy parameters include content block priority parameters, resource preloading level parameters, switching timeout threshold parameters, and rollback trigger threshold parameters.

[0052] A landing page content switching system based on human-computer interaction includes a content block topology graph construction module, which is used to split the target landing page into content blocks and construct a content block topology graph containing dependencies, mutual exclusions and rollback pointers.

[0053] The interactive event acquisition module is used to collect human-computer interaction events in the current session and form an interactive event stream;

[0054] An interaction stabilization processing module is used to perform time window aggregation, repeated trigger suppression, and noise filtering on the interaction event stream.

[0055] The intent recognition module is used to output the session intent category and intent confidence level;

[0056] The candidate switching generation module is used to generate a set of candidate switching content blocks based on the session intent category and the content block topology relationship;

[0057] The rendering layer filtering module is used to determine the target switching content block based on terminal rendering load, expected network latency and page memory usage, and to execute the degradation strategy when the degradation conditions are met.

[0058] The preloading and shadow rendering module is used to complete resource preloading, shadow rendering, resource integrity verification, and structural consistency verification of the target switching content block;

[0059] The atomic switching execution module is used to perform atomic replacement of content blocks when the verification passes.

[0060] The rollback control module is used to restore the previous stable content block version based on the rollback pointer when the rollback condition is triggered;

[0061] The attribution update module is used to record the switching action and the behavior results after the switching, and update the switching strategy parameters;

[0062] Each module transmits interactive events, filtering results, verification results, and rollback control commands through a communication interface.

[0063] In summary, the present invention has the following main beneficial effects:

[0064] This application achieves the goal of completing structural legality verification and semantic relevance verification before switching by splitting the landing page into independently schedulable content blocks and constructing a content block topology graph containing dependencies, mutual exclusions, and rollback pointers. Combined with session intent recognition and candidate content block filtering, it also achieves the goal of ensuring execution feasibility while ensuring content relevance by introducing terminal rendering load, estimated network latency, and page memory usage during the decision-making stage. This reduces invalid switching, conflicting switching, and stuttering issues under high load conditions.

[0065] This application achieves the goal of intercepting errors before submission through a collaborative mechanism of preloading, shadow rendering, resource integrity verification, and structural consistency verification; and achieves the goal of continuous interface and interaction before and after switching by using double buffering single-frame atomic submission and synchronous restoration of form state, scroll anchors, and input focus, thereby reducing the probability of white screen, flickering, field misalignment, and user operation interruption, and improving page usability and stability.

[0066] This application achieves the goal of quickly restoring the previous stable version in abnormal scenarios by setting multiple conditions for rollback triggering rules, such as switching timeout, verification failure, frame drop exceeding the threshold, and return operation. By recording the state before switching, switching actions, and switching results and performing incremental updates under boundary constraints, it achieves the goal of continuously optimizing the content block priority, preloading level, and rollback-related thresholds, thereby forming a verifiable, recoverable, and iterative closed-loop control link, improving the long-term operating effect of the system and the reliability of engineering implementation. Attached Figure Description

[0067] Figure 1 This is a flowchart of the method of the present invention;

[0068] Figure 2 This is a schematic diagram of the content block topology of the present invention;

[0069] Figure 3 This is a flowchart of the candidate generation and rendering layer selection process of the present invention. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1

[0072] refer to Figure 1 Methods for switching landing page content based on human-computer interaction include:

[0073] S1: Break down the target landing page into content blocks, construct a content block topology diagram containing dependencies, mutual exclusions and rollback pointers, and create a resource list and version identifier for each content block;

[0074] S2: Collect human-computer interaction events of the current session. Human-computer interaction events include at least click events, swipe events, pause events, input events, and return events, and sort them according to a unified timestamp to form an interaction event stream;

[0075] S3: Perform stabilization processing on the interactive event stream to obtain a stable interactive state feature vector for intent recognition;

[0076] S4: Input the stable interaction state feature vector into the intent recognition model, output the session intent category and intent confidence, and generate a set of candidate switching content blocks based on the session intent category;

[0077] S5: Perform rendering layer filtering on the candidate switching content block set based on terminal rendering load, expected network latency and page memory usage to obtain the target switching content block; when the intent confidence is lower than the first threshold or the terminal rendering load is higher than the second threshold, execute the degradation strategy. The degradation strategy is to keep the currently displayed content block from switching and pause the submission of candidate switching content blocks until the next stable interaction state feature vector is generated and then the rendering layer filtering is re-executed.

[0078] S6: Perform preloading and shadow rendering on the target switching content block, and perform resource integrity verification and structural consistency verification;

[0079] S7: When both resource integrity and structural consistency checks pass, perform an atomic commit switch, replacing the currently displayed content block with the target switch content block in one go, while maintaining the continuity of form state, scroll anchors, and input focus;

[0080] S8: When a switching timeout, rendering failure, or user return operation triggers a rollback condition, restore to the previous stable content block version according to the rollback pointer;

[0081] S9: Record the interaction state feature vector before switching, the switching action, and the user behavior results after switching, generate causal attribution records, and update the switching strategy parameters for subsequent sessions.

[0082] S1: Content Block Splitting and Topology Construction: The server performs content block splitting on the target landing page, dividing the page into content block units that can be loaded, rendered, and replaced independently. Each content block is configured with a content block identifier, version identifier, dependency set, mutual exclusion set, resource list, field pattern identifier, and rollback pointer.

[0083] The server constructs a content block topology graph based on the aforementioned metadata. This graph describes the sequential dependencies, mutual exclusions, and version rollback relationships between content blocks. The resource manifest at least includes static resource addresses, resource checksums, dynamic interface identifiers, template version identifiers, and field pattern identifiers. Rollback pointers are used to point to the previous stable version.

[0084] In implementation, the content block topology graph can be denoted as: ,in, A collection of content block nodes. For a set of dependent edges, For a set of mutually exclusive edges, This is a set of rollback pointers.

[0085] S1 first models the switching object, transforming the original page-level control into content block-level control, and explicitly graphs the dependencies, mutual exclusions, and rollback relationships. Subsequent steps then filter target blocks under graph constraints and execute the switch, ensuring the switch path has prior validity and recoverability.

[0086] By using content block-level modeling, target blocks with unmet dependencies or mutual conflicts can be intercepted before switching, reducing the probability of erroneous switching. By binding rollback pointers to version identifiers, the previous stable content block version can be quickly restored in case of anomalies, thereby improving page switching stability and system fault tolerance.

[0087] S2: Interactive Event Acquisition and Event Stream Construction: The terminal collects human-computer interaction events of the current session. Interactive events include at least click events, swipe events, pause events, input events, and return events. Each event record includes at least the session identifier, event type, target element identifier, displacement, pause duration, input length, and timestamp.

[0088] The terminal performs unified timestamp sorting on event records to form an interactive event stream. If there is a time offset between the terminal clock and the server clock, the clock offset issued during the session initialization phase is used for time alignment to ensure cross-terminal time semantic consistency.

[0089] S2 unifies heterogeneous interactive behaviors into standard event records, and then organizes them into an interactive event stream according to time sequence. This event stream is the sole input source for subsequent stabilization processing and intent recognition, ensuring consistent data standards throughout the entire process.

[0090] By unifying the event model and timeline, misjudgment caused by inconsistent event formats and time offsets can be reduced, improving the accuracy of subsequent feature extraction and enhancing the system's consistent performance under different terminals and network conditions.

[0091] S3: Interactive event stabilization processing: The system aggregates interactive events within a preset time window threshold to obtain statistical characteristics such as event frequency, average scrolling speed, and average dwell time.

[0092] When the same event type is repeatedly triggered within a preset minimum trigger interval threshold, subsequent triggering events are suppressed. Events with displacement amounts less than a preset displacement threshold in sliding events, and events with dwell times less than a preset dwell time threshold in dwell events, are identified as noise events and filtered out.

[0093] After stabilization, output stable interaction state feature vectors. As input to the intent recognition model.

[0094] S3 uses a three-layer process of window aggregation, repetition suppression, and noise filtering to transform the high-frequency jittering raw event stream into a smooth, discriminative state feature vector, thus avoiding momentary misinterpretations that dominate decision-making results.

[0095] This step can significantly improve the feature signal-to-noise ratio, reduce interference from non-target behaviors such as tapping, accidental touches, and minor swipes, make subsequent intent recognition more stable, and reduce invalid and accidental switching.

[0096] S4: Intent recognition and candidate switching content block generation server will stabilize the interaction state feature vector Input the intent recognition model and output the session intent category and intent confidence score. Session intent categories include at least browsing intent, comparison intent, decision intent, and return-risk intent. The intent probability is calculated using the following formula:

[0097]

[0098] in, Total number of session intent categories This is the index of the target intent category. The traversal index is the intent category. This is the current session time or the current time window number. The stable interaction state feature vector is used to characterize the state at time t. Or the interaction statistics within the corresponding time window, The dimension is . For the first The weight parameter vector corresponding to the class intent, Dimensions and Consistency, for dimensional vector; For the first The bias parameter corresponding to the class intent is a scalar. The denominator contains... and The first The meanings of the weight parameter vector and bias parameter corresponding to the class intent are as follows: , They are the same, but belong to different categories. It is an exponential function. This is obtained from the above formula. For the first Class intent at time The probability value satisfies and . The confidence level of intent is defined as the maximum value among the probability values ​​of various intents.

[0099] S4 first performs semantic judgment, then performs structural filtering. The model maps behavioral sequences to session intents, and then constrains the legality of the candidate set by topological relationships, ultimately forming a set of candidate content blocks that both conform to the user's current state and satisfy the structural rules.

[0100] Compared to simple rule switching, S4 introduces intent confidence as a decision-making criterion, making target selection more aligned with the current session stage. Furthermore, the addition of dependency and mutual exclusion constraints effectively reduces irrelevant and conflicting switches.

[0101] The system calculates the estimated rendering time, estimated memory increment, and estimated network latency for each candidate content block to be switched. The estimated network latency can be estimated using the following formula:

[0102]

[0103] in, For candidate content blocks The estimated network latency, for ease of implementation, is... The terminal can be used to check the most recent The round-trip time of each network request is statistically analyzed. The round-trip time is the time interval from initiating the request to receiving the first packet response. The moving average method can be used for updating. The length of the terminal's current queue of pending requests, which includes requests that are in the queue, being sent, or waiting for a response. For candidate content blocks The target resource volume is determined by the sum of the number of static resource bytes in the resource list and the estimated value of the dynamic data response volume. To estimate bandwidth, the terminal can use the nearest... The throughput of each resource download was statistically analyzed, and the throughput is the ratio of the number of bytes downloaded to the download time. , , These are non-negative weighting coefficients used to adjust the contribution of each factor to the expected network latency. The initial value can be determined by historical operation log statistics or terminal baseline testing, and adjusted within preset upper and lower limits. Threshold constraints must be met. , , The content block is entered into the executable set. For candidate content blocks The estimated rendering time is used to characterize the estimated time required to complete the rendering of this content block under the current rendering load conditions of the terminal. It can be obtained from the rendering time statistics of the shadow rendering stage, or estimated from the moving average of historical rendering time. For candidate content blocks The projected memory increment is used to characterize the increase in memory usage relative to the currently displayed content block after loading and rendering the content block resource. This can be obtained from the incremental statistics of preloaded resource volume and runtime memory sampling, or estimated from the moving average of historical memory increments. Threshold , , These are the rendering duration threshold, memory threshold, and network latency threshold, respectively, which can be determined by terminal baseline testing and historical operation log statistics, and updated within preset upper and lower limits. The executable set is defined as simultaneously satisfying... , , The set of candidate content blocks; if the executable set is empty, the currently displayed content block is not switched, and the calculation is recalculated after the feature vector of the next stable interaction state is obtained.

[0104] The executable set can be used to determine the target switching content block based on a comprehensive score:

[0105]

[0106] in, To normalize the intent matching degree, To normalize the estimated rendering time, To normalize the expected memory increment, To normalize the estimated network latency, For normalized structure fit;

[0107] in, to These are non-negative weight parameters used to characterize the relative importance of intent matching, estimated rendering time, estimated memory increment, estimated network latency, and structural adaptability in the overall score. The initial values ​​of the weight parameters can be determined by historical running log statistics or offline evaluation, and are updated within preset upper and lower limits. This indicates that the candidate content block with the highest overall score within the executable set is selected as the target switching content block. Normalization can be implemented using min-max normalization, for example, normalizing the expected rendering time to... ,in and These are the minimum and maximum estimated rendering times within the current candidate set, respectively. The normalization of the estimated memory increment and estimated network latency uses the same method, ensuring that each normalization result falls within... In the case of a tie in the overall score, the tie will be eliminated in the following order: shorter expected rendering time, shorter expected network latency, and lower rollback risk level.

[0108] When scores are tied, they are eliminated in the following order: shorter expected rendering time, shorter expected network latency, and lower rollback risk.

[0109] When the intent confidence level is lower than the first threshold or the terminal rendering load is higher than the second threshold, a degradation strategy is implemented: the currently displayed content block is not switched and candidate submissions are paused until the feature vector of the next stable interaction state is generated and then re-selected.

[0110] S5 jointly models what users want to see and whether the device can display it stably. It first filters out unexecutable candidates with three types of hard constraints, and then performs multi-objective scoring optimization on the executable set to ensure that the target block is both relevant and feasible.

[0111] This step incorporates rendering load, memory pressure, and network latency into the same decision-making chain, reducing the probability of stuttering, frame drops, and submission failures after switching. The degradation strategy provides proactive steady-state protection in low-confidence or high-load scenarios, preventing frequent page flickering.

[0112] S6: Preloading, Shadow Rendering and Dual Verification: The system preloads the target switching content block, writes static resources to the local cache, and writes dynamic data request results to the session cache.

[0113] After constructing the target switching content block structure tree in the shadow rendering area, resource integrity checks and structural consistency checks are performed. Structural consistency checks include at least the template version identifier, data field schema identifier, and required field constraints.

[0114] Only when both resource integrity verification and structural consistency verification pass can the target switching content block be allowed to enter the commit stage.

[0115] S6 employs an execution mechanism that first builds the invisible area and then verifies it before submission, intercepting resource and structural errors in advance to prevent problems from being exposed in the user's visible area.

[0116] By using preloading and shadow rendering for dual verification, issues such as white screen, misalignment, and missing fields after switching can be significantly reduced, improving display consistency and submission success rate.

[0117] S7: Atomic Commit Switching and State Continuity: The terminal establishes a current display buffer and a target display buffer. After the target display buffer completes full rendering and passes double verification, it replaces the current display buffer with a single commit operation within the same rendering frame, achieving atomic commit switching.

[0118] Before submission, a state snapshot is saved, which includes at least the form field values, scroll anchor positions, and input focus indicators; after submission, the corresponding state is restored according to the snapshot to maintain operation continuity.

[0119] S7 employs a double-buffered single-frame replacement mechanism to avoid intermediate state presentation; and achieves continuity of interactive context before and after switching through state snapshot migration.

[0120] This step reduces the risk of screen flickering, half-rendered exposure, and focus loss, minimizes user interruptions, and improves session continuity and usability.

[0121] S8: Rollback Trigger Judgment and Version Recovery: Rollback is triggered when any of the following conditions occur: the switching execution duration is greater than the timeout threshold, the resource integrity check fails, the structural consistency check fails, the frame drop rate obtained within a consecutive preset number of frames after the switch is greater than the frame drop threshold, or a user return operation is detected within the preset rollback observation window.

[0122] Rollback triggering can be represented as:

[0123]

[0124] in, This is the rollback trigger flag. To switch execution duration, To switch timeout thresholds, For the resource integrity verification results, For the structural consistency verification results, For frame rate drops, The frame drop threshold, This is a return operation flag.

[0125] After a rollback is triggered, the system restores to the previous stable content block version based on the rollback pointer, restores the corresponding state snapshot, and records the rollback reason code.

[0126] S8 integrates execution exceptions, rendering exceptions, and explicit user denials into the rollback determination framework, directly locating the stable version and completing state restoration through the rollback pointer.

[0127] This step makes the exception handling path clear and the recovery action verifiable, which can limit the scope of error impact to a single switching cycle, thereby improving the system's fault tolerance and session stability.

[0128] S9: Causal Attribution Recording and Strategy Parameter Update: The system assigns a unique action identifier to each switching action, records the interaction state feature vector before the switch, the switching action, the dwell time after the switch, the preset target behavior completion marker, and the return marker. The preset target behavior completion marker is defined as a form submission completion marker or a button click completion marker.

[0129] Incremental updates are performed based on samples associated with action identifiers. The updated switching strategy parameters include at least the content block priority parameter, resource preloading level parameter, switching timeout threshold parameter, and rollback trigger threshold parameter.

[0130] Parameter updates can be performed using:

[0131]

[0132] in, For content blocks At any moment Priority parameters, For the current sample results, To update the step size, For parameter boundaries.

[0133] S9 establishes a closed loop of state, action, and result, feeds back the session results to the parameter layer, and prevents parameter mutations through boundary constraints to ensure that updates are controllable.

[0134] This step enables the system to continuously and adaptively optimize under different traffic and terminal conditions, improving long-term hit rate and stability, while retaining traceable records for easy review, interpretation, and troubleshooting.

[0135] This embodiment provides a landing page content switching system based on human-computer interaction, used to execute the aforementioned method embodiments. The system includes a content block topology graph construction module, an interaction event acquisition module, an interaction stabilization processing module, an intent recognition module, a candidate switching generation module, a rendering layer filtering module, a preloading and shadow rendering module, an atomic switching execution module, a rollback control module, and an attribution update module.

[0136] The content block topology graph construction module is used to split the target landing page into content blocks and construct a content block topology graph that includes dependencies, mutual exclusions and rollback pointers.

[0137] The interaction event acquisition module is used to collect human-computer interaction events of the current session and form an interaction event stream; the interaction stabilization processing module is used to perform time window aggregation, repeated trigger suppression and noise filtering on the interaction event stream, and output a stable interaction state feature vector.

[0138] The intent recognition module outputs the session intent category and intent confidence; the candidate switching generation module generates a set of candidate switching content blocks based on the session intent category and the content block topology relationship; the rendering layer filtering module determines the target switching content block by combining terminal rendering load, expected network latency and page memory usage, and triggers a degradation strategy when the conditions are met.

[0139] The preloading and shadow rendering module is used to complete the resource preloading, shadow rendering, resource integrity verification, and structural consistency verification of the target switching content block; the atomic switching execution module is used to perform atomic replacement of the content block when the verification passes and maintain the continuity of form state, scroll anchor point, and input focus; the rollback control module is used to restore the previous stable content block version according to the rollback pointer when the rollback condition is triggered; the attribution update module is used to record the switching action and the behavior results after the switching and update the switching strategy parameters.

[0140] Each module interacts with other modules via a communication interface to exchange data and control commands. The communication data includes at least a session identifier, a content block identifier, a version identifier, a scoring result, a verification result, and a rollback control command.

[0141] The working principle of this application can be described as the following process:

[0142] After the system starts, the server first splits the target landing page into content blocks and builds a content block topology graph that includes dependencies, mutual exclusions and rollback pointers; at the same time, it creates a resource list, version identifier and field pattern identifier for each content block.

[0143] This topological constraint allows the system to determine whether candidate content blocks meet the entry conditions before switching, thus avoiding invalid switching caused by structural conflicts or missing prerequisites.

[0144] During the session, the terminal continuously collects click events, swipe events, pause events, input events, and return events, forming an interactive event stream.

[0145] The interactive event stream first undergoes stabilization processing, including time window aggregation, repeated trigger suppression, and noise filtering, outputting a stable interactive state feature vector to reflect the true interactive state of the current session and reduce the interference of accidental touches and jitter on decision-making.

[0146] The server inputs the stable interaction state feature vector into the intent recognition model to obtain the session intent category and intent confidence; then, combined with the successor relationship, dependency relationship and mutual exclusion relationship of the currently displayed content block in the topology graph, it generates a set of candidate switching content blocks.

[0147] Based on this, the system introduces three types of execution constraints: terminal rendering load, expected network latency, and page memory usage. These constraints are used to filter the candidate set at the rendering layer to determine the target content block to be switched. When the confidence level is insufficient or the terminal load is too high, a degradation strategy is implemented to keep the currently displayed content block from switching until the next round of stable features arrives and the selection is re-selected.

[0148] After the target content block is determined, the terminal first performs resource preloading, then completes the structure tree construction in the shadow rendering area, and performs resource integrity verification and structure consistency verification in sequence.

[0149] Only when both checks pass will the system perform an atomic submission switch within the same rendering frame, replacing the currently displayed content block with the target switching content block in one go, and simultaneously restoring the form state, scroll anchors, and input focus to ensure continuous interaction before and after the switch.

[0150] If a timeout, verification failure, frame drop rate exceeding the threshold, or user return operation occurs during the switching process, the system will trigger a rollback mechanism to restore the previous stable content block version according to the rollback pointer and restore the corresponding state snapshot, thereby limiting the impact of the anomaly to a single switching cycle.

[0151] After a switch is completed, the system records the interaction state feature vector before the switch, the switch action, and the user behavior results after the switch, generates a causal attribution record, and performs incremental updates on the content block priority parameter, resource preloading level parameter, switch timeout threshold parameter, and rollback trigger threshold parameter under boundary constraints.

[0152] This forms a closed-loop control chain encompassing interactive data collection, intent recognition, executable filtering, secure submission, exception rollback, and attribution updates, enabling real-time, stable execution and continuous self-optimization of landing page content switching.

[0153] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A landing page content switching method based on human-computer interaction, characterized in that, Executed collaboratively by terminal devices and servers, including: S1: Break down the target landing page into content blocks, construct a content block topology diagram containing dependencies, mutual exclusions and rollback pointers, and create a resource list and version identifier for each content block; S2: Collect human-computer interaction events of the current session, including click events, swipe events, pause events, input events and return events, and sort them according to a unified timestamp to form an interaction event stream; S3: Perform stabilization processing on the interactive event stream to obtain a stable interactive state feature vector for intent recognition; S4: Input the stable interaction state feature vector into the intent recognition model, output the session intent category and intent confidence, and generate a set of candidate switching content blocks based on the session intent category; S5: Perform rendering layer filtering on the candidate switching content block set based on terminal rendering load, expected network latency and page memory usage to obtain the target switching content block; S6: Perform preloading and shadow rendering on the target switching content block, and perform resource integrity verification and structural consistency verification; S7: When both the resource integrity check and the structural consistency check pass, perform an atomic commit switch to replace the currently displayed content block with the target switch content block in one go; S8: When a switching timeout, rendering failure, or rollback condition is detected, restore to the previous stable content block version according to the rollback pointer; S9: Record the stable interaction state feature vector before switching, the switching action, and the user behavior results after switching, generate causal attribution records, and update the switching strategy parameters for subsequent sessions.

2. The landing page content switching method based on human-computer interaction according to claim 1, characterized in that, The stabilization process includes: Aggregate the interactive event stream within a preset time window threshold; When the same event type is repeatedly triggered within a preset minimum trigger interval threshold, subsequent triggering events in the interactive event stream are suppressed. Sliding events with displacement less than a preset displacement threshold and dwell events with dwell time less than a preset dwell time threshold in the interactive event stream are identified as noise events and filtered out.

3. The landing page content switching method based on human-computer interaction according to claim 2, characterized in that, Inputting stable interaction state feature vectors into the intent recognition model includes: The event type, event time interval, sliding speed, input length, and number of returns are used to construct a temporal feature vector; The intent recognition model outputs the probability value corresponding to each session intent category; The session intent category corresponding to the highest probability is determined as the current session intent category, and the highest probability is determined as the intent confidence level.

4. The landing page content switching method based on human-computer interaction according to claim 3, characterized in that, The step of generating a set of candidate switching content blocks based on the session intent category includes: Extract content blocks that satisfy the dependency relationship from the set of successor content blocks in the content block topology graph of the currently displayed content block; Remove content blocks that are mutually exclusive with the currently displayed content block; The candidate switching content block set is obtained by filtering according to the session intent category and the content block mapping table.

5. The landing page content switching method based on human-computer interaction according to claim 4, characterized in that, The process of performing rendering layer filtering based on terminal rendering load, estimated network latency, and page memory usage includes: Calculate the estimated rendering time, estimated memory increment, and estimated network latency for each candidate content block to be switched. Candidate switching content blocks with expected rendering time not exceeding the rendering time threshold, expected memory increment not exceeding the memory threshold, and expected network latency not exceeding the latency threshold are identified as executable content blocks; In the executable content block, a comprehensive score is calculated based on intent confidence, estimated rendering time, estimated memory increment, and estimated network latency, and the executable content block with the highest comprehensive score is selected as the target switching content block; When the confidence level of the intent is lower than the first threshold or the terminal rendering load is higher than the second threshold, a degradation strategy is executed. The degradation strategy is to keep the currently displayed content block from switching and to pause the execution of the atomic submission switch.

6. The landing page content switching method based on human-computer interaction according to claim 5, characterized in that, The preloading and shadow rendering include: Write the static resources corresponding to the target switching content block to the local cache. Write the dynamic data request result corresponding to the target switching content block to the session cache. After the target switching content block structure tree is constructed in the shadow rendering area, resource check code comparison is performed, and structural consistency verification is performed based on the structure tree version identifier and data field pattern identifier.

7. The landing page content switching method based on human-computer interaction according to claim 6, characterized in that, The execution of the atomic commit switch includes: Establish the current display buffer and the target display buffer; After the target display buffer has completed full rendering, the current display buffer is replaced by a single commit operation within the same rendering frame; After the replacement is complete, restore the form state snapshot, scroll anchor marker, and input focus marker saved before the switch.

8. The landing page content switching method based on human-computer interaction according to claim 7, characterized in that, The rollback condition includes any of the following conditions: The execution duration of the switch exceeds the timeout threshold; Resource integrity check failed, or structural consistency check failed; The frame drop rate obtained within a consecutive preset number of frames after the switch is greater than the frame drop threshold. A user return operation was detected within the preset rollback observation window; When the rollback condition is triggered, press the rollback pointer to restore the previous stable content block version and restore the corresponding state snapshot.

9. The landing page content switching method based on human-computer interaction according to claim 8, characterized in that, The process of generating causal attribution records and updating the switching strategy parameters for subsequent sessions includes: Assign a unique action identifier to each switching action; Record the stable interaction state feature vector before switching, the dwell time after switching, the preset target behavior completion mark and the return mark, wherein the preset target behavior completion mark is a form submission completion mark or a button click completion mark; Incremental parameter updates are performed based on the sample set associated with action identifiers to obtain updated switching strategy parameters, which are then used for selecting target switching content blocks in subsequent sessions. The updated switching strategy parameters include content block priority parameters, resource preloading level parameters, switching timeout threshold parameters, and rollback trigger threshold parameters.

10. A landing page content switching system based on human-computer interaction, applicable to the landing page content switching method based on human-computer interaction as described in any one of claims 1-9, characterized in that, include: The content block topology graph construction module is used to split the target landing page into content blocks and build a content block topology graph that includes dependencies, mutual exclusions and rollback pointers. The interactive event acquisition module is used to collect human-computer interaction events in the current session and form an interactive event stream; An interaction stabilization processing module is used to perform time window aggregation, repeated trigger suppression, and noise filtering on the interaction event stream. The intent recognition module is used to output the session intent category and intent confidence level; The candidate switching generation module is used to generate a set of candidate switching content blocks based on the session intent category and the content block topology relationship; The rendering layer filtering module is used to determine the target switching content block based on terminal rendering load, expected network latency and page memory usage, and to execute the degradation strategy when the degradation conditions are met. The preloading and shadow rendering module is used to complete resource preloading, shadow rendering, resource integrity verification, and structural consistency verification of the target switching content block; The atomic switching execution module is used to perform atomic replacement of content blocks when the verification passes. The rollback control module is used to restore the previous stable content block version based on the rollback pointer when the rollback condition is triggered; The attribution update module is used to record the switching action and the behavior results after the switching, and update the switching strategy parameters; Each module transmits interactive events, filtering results, verification results, and rollback control commands through a communication interface.

Images