A Human-Computer Interaction Method and System for a Learning Machine User Interface
By generating recent behavior sequences and multidimensional state vectors, error types are distinguished and feedback content is adjusted, solving the problem of mismatch between traditional learning machine interface feedback and achieving more accurate learning record updates and feedback adjustments.
Patent Information
- Application Number
- CN202611386178.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-08
- Publication Date
- 2026-11-10
Smart Images

Figure CN123067754A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of human-computer interaction technology, and in particular to a human-computer interaction method and system for a learning machine user interface. Background Technology
[0002] The field of human-computer interaction technology mainly involves users inputting operational information into a terminal through touchscreens, buttons, voice input, etc. The terminal collects, recognizes, judges, stores, and invokes the interface based on the click location, input content, operation sequence, and device status, and outputs interactive content to the user through text, images, animation, or audio. In particular, the human-computer interaction method of a traditional learning machine user interface refers to the learning machine receiving user input of selected characters, writing, clicking, or voice responses during literacy practice. It records the response content, accuracy rate, response time, number of clicks, and practice progress. The response results are compared with standard answers and preset rules in a question bank. Based on the correct, incorrect, timeout, or continuous response status, it invokes pre-configured prompts, voice instructions, demonstration animations, or redirects to pages. The question number, response result, and operation time are stored locally or uploaded to a server for subsequent practice content selection and learning record retrieval.
[0003] Traditional learning machines' literacy practice interfaces mostly rely on single-response results and preset rules to call text prompts, voice instructions, or demonstration animations. The feedback content and playback method are mainly determined by the correct, incorrect, timeout, and continuous response status. Although the response time, number of clicks, and practice records are stored or uploaded, there is a lack of correlation analysis between interactive actions and learning performance at different time periods. It is difficult to distinguish the causes of errors based on recent behavior. Uniform feedback is prone to mismatch with the user's mastery status and recall process. Accidental touches or induced clicks during feedback can also be mixed into the learning records, affecting the judgment of subsequent practice and the basis for interface control. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose a human-computer interaction method and system for a learning machine user interface.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: a human-computer interaction method for a learning machine user interface, comprising the following steps: S1: Collect continuous interaction data and learning performance data through the learning machine's literacy practice interface, and aggregate the continuous interaction data and learning performance data according to the recent time window to generate a recent behavior sequence and a multi-dimensional state vector. S2: In response to the error response signal, determine the current error type based on the recent behavior sequence as a short-duration high-mastery error, a delayed recall error, or a repetitive low-accuracy error, and calculate the dynamic cognitive ability index based on the multidimensional state vector; S3: Based on the recent behavior sequence, the current error type, and the dynamic cognitive ability index, filter error records that belong to short-duration high-mastery errors or recall delay errors and correspond to complete animation feedback, count the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks, calculate the feedback mismatch index based on the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks, and determine the target feedback level and error correction intensity based on the comparison result of the feedback mismatch index and the preset mismatch threshold; S4: Invoke the target feedback level and the error correction intensity to adjust the display depth, display duration, forced playback attribute and jump attribute of the feedback content, output the interface feedback content of the learning machine, and generate feedback output records; S5: Based on the feedback output record, collect the interactive actions during and after the feedback output, and match the interactive actions with preset interface inducement behavior features to obtain interface inducement behavior markers; S6: Based on the interface-induced behavior marker and the feedback output record, correct the interactive actions during the feedback period in the continuous interaction data and the answering time and error records in the learning performance data to obtain the learning behavior record, and update the recent behavior sequence and real-time multidimensional state vector based on the learning behavior record.
[0006] As a further aspect of the present invention, the process of S1 is specifically as follows: S101: Collect answer behavior data, feedback interaction data, and follow-up answer data as continuous interaction data through the event listening interface of the learning machine's literacy practice interface. For each answer behavior data, associate word identifier, answer result, answer time, original interaction time, and feedback interaction data. The answer time and the error record corresponding to the answer result being incorrect constitute learning performance data. S102: Based on the continuous interaction data and the learning performance data, sort the data by time using a recent time window, filter the continuous interaction data and learning performance data that fall into the recent time window, associate the answer results and answer time in the continuous interaction data and learning performance data according to word identifiers, and generate a recent behavior sequence; S103: Based on the recent behavior sequence, the answer result and answer time corresponding to each word identifier are statistically analyzed. Combined with the feedback interaction data, the recent accuracy rate corresponding to each word identifier is calculated. The overall recent accuracy rate, average answer time, average number of invalid clicks during the feedback period, and error distribution dispersion are calculated to generate a multi-dimensional state vector.
[0007] As a further aspect of the present invention, the process of S2 is specifically as follows: S201: Based on the answer results corresponding to the current answer word in the recent behavior sequence, the recent accuracy rate corresponding to the current answer word is obtained, the current answer time corresponding to the current answer word is read, the average answer time of the questions in the learning machine question record is read, the time ratio of the current answer time to the average answer time of the questions is calculated, and the error judgment data is obtained. S202: Based on the error determination data, compare the recent accuracy rate with the preset accuracy rate threshold, compare the time consumption ratio with the preset time consumption threshold, and determine the current error type among short-time high-mastery errors, recall delay errors, and repetitive low-accuracy errors; S203: Based on the multidimensional state vector, read the overall recent accuracy rate, average response time, and error distribution dispersion. Perform negative correlation normalization on the average response time to obtain normalized time data. Calculate the weighted sum of the overall recent accuracy rate and the normalized time data. Adjust the weighted sum according to the error distribution dispersion to obtain the dynamic cognitive ability index.
[0008] As a further aspect of the present invention, the process of S3 is specifically as follows: S301: Based on the recent behavior sequence and the current error type, filter error records where the current error type is a short-time high-mastery error or a recall delay error, read the feedback playback type corresponding to each error record, filter error records where the feedback playback type is a complete animation feedback, and obtain candidate error records; S302: Based on the candidate error records, count the number of error records, obtain the average playback duration and invalid click frequency of the complete animation feedback, calculate the total playback duration corresponding to the number of error records and the average playback duration of the complete animation feedback, calculate the cumulative invalid clicks corresponding to the total playback duration and the invalid click frequency, and calculate the feedback mismatch index; S303: Based on the comparison between the feedback mismatch index and the preset mismatch threshold, the target feedback level and error correction intensity are determined according to the comparison result, the dynamic cognitive ability index and the current error type.
[0009] As a further aspect of the present invention, the calculation process of the feedback mismatch index is specifically as follows: The system reads the feedback start time, feedback end time, and click events during the feedback period for each candidate error record. It calculates the single feedback playback duration corresponding to the feedback start time and feedback end time, counts the number of click events during the feedback period that occur between the feedback start time and feedback end time and are outside the preset interface function control area, and obtains the single invalid click count. It calculates the average of the single feedback playback durations corresponding to each candidate error record to obtain the average playback duration of the complete animation feedback. It calculates the ratio of the sum of the single invalid click counts corresponding to each candidate error record to the sum of the single feedback playback durations corresponding to each candidate error record, and obtains the invalid click frequency. It calculates the total playback duration corresponding to the number of error records in the candidate error records and the average playback duration of the complete animation feedback, and calculates the cumulative invalid click count corresponding to the total playback duration and the invalid click frequency. Based on the ratio of the cumulative invalid click count to a preset mismatch normalization benchmark, it obtains the feedback mismatch index.
[0010] As a further aspect of the present invention, the process of S4 is specifically as follows: S401: Based on the target feedback level and the error correction intensity, establish the association between the target feedback level, error correction intensity and display depth, display duration, forced playback attribute, and jump attribute, and generate feedback control parameters; S402: Adjust the display depth, display duration, forced playback attribute and jump attribute of the feedback content according to the feedback control parameters, associate the feedback content with the output position of the learning machine's literacy practice interface, and output the interface feedback content. S403: Collect the start time, end time, and playback status based on the interface feedback content, establish the association between the interface feedback content, start time, end time, and playback status, and generate a feedback output record.
[0011] As a further aspect of the present invention, the process of S5 is specifically as follows: S501: Read the start time, end time and playback status according to the feedback output record, collect the interactive actions during and after the feedback output through the event listening interface of the learning machine's literacy practice interface, associate the interactive actions with the feedback output record, and obtain the interactive action record; S502: Obtain preset interface inducement behavior features, extract the click frequency, click position, sliding trajectory and response interval from the interaction action record, and match the click frequency, click position, sliding trajectory and response interval with the preset interface inducement behavior features to obtain interface inducement behavior markers.
[0012] As a further aspect of the present invention, the process of obtaining the interface-induced behavior marker is specifically as follows: The system reads click frequency, click location, swipe trajectory, and response interval from the interaction action record. It compares the click frequency with a preset click frequency range to obtain click frequency hit data, compares the click location with a preset non-functional area range to obtain click location hit data, compares the swipe trajectory with a preset swipe trajectory feature to obtain swipe trajectory hit data, and compares the response interval with a preset response interval threshold to obtain response interval hit data. It then counts the number of hits in the click frequency hit data, click location hit data, swipe trajectory hit data, and response interval hit data, compares the number of hits with a preset hit count threshold, and obtains the interface-induced behavior marker based on the comparison result.
[0013] As a further aspect of the present invention, the process of S6 is specifically as follows: S601: Based on the interface-induced behavior marker and the feedback output record, filter the interaction actions during the feedback period from the continuous interaction data, associate the interaction actions during the feedback period with the interface-induced behavior marker, and obtain the interference action record; S602: Based on the statistics of the interference action record, the feedback waiting time and invalid click time are obtained. The feedback waiting time and invalid click time are deducted from the original interaction time to obtain the pure cognition time. The pure cognition time is used as the corrected response time. The weight adjustment value corresponding to the error record and the interface inducement behavior mark is calculated. The corrected response data is generated based on the corrected response time and the weight adjustment value. S603: Write the corrected response data into the recent behavior sequence, update the multidimensional state vector according to the written recent behavior sequence, establish the association between the corrected response data and the recent behavior sequence, and obtain the learning behavior record.
[0014] A learning machine user interface human-computer interaction system, the learning machine user interface human-computer interaction system being used to execute the above-described learning machine user interface human-computer interaction method, the system comprising: Behavior state aggregation module: Collects continuous interaction data and learning performance data through the learning machine's literacy practice interface, and aggregates the continuous interaction data and learning performance data according to the recent time window to generate recent behavior sequences and multi-dimensional state vectors; Cognitive diagnosis and assessment module: In response to incorrect answer signals, it determines the current error type based on the recent behavior sequence as a short-time high-mastery error, a delayed recall error, or a repetitive low-accuracy error, and calculates the dynamic cognitive ability index based on the multidimensional state vector; The mismatch feedback decision module: Based on the recent behavior sequence, the current error type, and the dynamic cognitive ability index, it filters error records that belong to short-duration high-mastery errors or recall delay errors and correspond to complete animation feedback. It counts the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks. Based on the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks, it calculates the feedback mismatch index. Based on the comparison result of the feedback mismatch index and the preset mismatch threshold, it determines the target feedback level and error correction intensity. Feedback Interface Adjustment Module: Invokes the target feedback level and the error correction intensity to adjust the display depth, display duration, forced playback attribute and jump attribute of the feedback content, outputs the interface feedback content of the learning machine, and generates feedback output records; Induced Behavior Capture Module: Based on the feedback output record, collect the interactive actions during and after the feedback output, and match the interactive actions with preset interface induced behavior features to obtain interface induced behavior markers; Behavior record update module: Based on the interface-induced behavior markers and the feedback output records, the module corrects the interactive actions during the feedback period in the continuous interaction data and the response time and error records in the learning performance data to obtain learning behavior records, and updates the recent behavior sequence and real-time multidimensional state vector based on the learning behavior records.
[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, continuous interaction data and learning performance data are aggregated through recent time windows to form recent behavior sequences and multi-dimensional state vectors. This distinguishes between short-duration, high-mastery errors, delayed recall errors, and recurring, low-accuracy errors, ensuring that feedback is based on both behavior and cognitive state. A feedback mismatch index is formed based on the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks. The target feedback level and error correction intensity are determined according to thresholds, and the display depth, display duration, forced playback attribute, and jump attribute are adjusted to match the feedback with the error type. Interface-induced behavior markers and feedback output records are used to correct interactive actions, response time, and error records, reducing interference from accidental touches on feedback-induced operations and providing a data foundation for state updates. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the human-computer interaction method for the user interface of the learning machine according to the present invention; Figure 2 This is a schematic diagram of the human-computer interaction workflow of the learning machine user interface of the present invention; Figure 3 This is a schematic diagram illustrating the feedback mismatch index processing principle of the present invention; Figure 4This is a schematic diagram illustrating the learning behavior recording and updating principle of the present invention; Figure 5 This is a schematic diagram of the human-computer interaction system structure of the learning machine user interface of the present invention. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0018] This embodiment applies to the continuous answering and feedback interaction process of a literacy practice interface on a learning machine. The literacy practice interface sequentially displays the words to be learned and receives answering actions. When an answer is incorrect, it outputs error correction feedback, and after the feedback ends, it continues to receive and accept answering actions. The entire process establishes data associations around the same learning user, the same literacy practice task, and continuously occurring interface events, ensuring a clear temporal sequence and object correspondence between answering actions, feedback playback, operations during feedback, and continued answering after feedback.
[0019] In this embodiment, the recent time window is a time range selecting consecutive interaction records backward from the current response event. Its configuration is loaded along with the literacy practice task and is used to limit the behavioral records participating in the current judgment. Multiple preset judgment conditions are read from the interaction rule configuration of the literacy practice interface. Among them, conditions related to accuracy, response time, and hit count are used to distinguish behavioral states; conditions related to interface areas are formed based on the operable boundaries of functional controls; conditions related to sliding trajectories are formed based on the standard sliding directions and trajectory continuity supported by the interface; and conditions related to feedback mismatch are formed based on feedback playback rules and operation rules during feedback. Each preset judgment condition serves only as a classification boundary or matching basis and does not change the collected data itself.
[0020] Please see Figure 1 and Figure 2S1: Collect continuous interaction data and learning performance data through the literacy practice interface of the learning machine, and aggregate the continuous interaction data and learning performance data according to the recent time window to generate a recent behavior sequence and a multi-dimensional state vector. Continuous interaction data is used to record the interface events that occur continuously from the time the question is displayed, the answer is submitted, the feedback is played, to the next answer. Learning performance data is used to record the answer result, answer time, and error situation corresponding to each word. The aggregation process first restores the interaction sequence according to the time of event occurrence, and then connects the answer event, feedback event, and subsequent answer event of the same question according to the word identifier, thereby forming a recent behavior sequence that can be called for subsequent error judgment. The multi-dimensional state vector is a set of state information organized according to a fixed field order, which carries the recent correct situation, answer time, operation situation during feedback, and error distribution, and is associated with the recent behavior sequence that generated the state information.
[0021] S101: Data collection includes answer behavior data, feedback interaction data, and follow-up answer data via the event listener interface of the learning machine's literacy practice interface. These three types of data are treated as continuous interaction data. The event listener interface continuously receives interface events after the question is displayed. Answer behavior data includes word identifiers, answer results, answer submission time, and answer duration. Feedback interaction data includes feedback start time, feedback end time, feedback playback status, and click and swipe events occurring during feedback. Follow-up answer data includes operation events generated when moving to the next question or re-answering after feedback ends. For each piece of answer behavior data, the corresponding answer result, answer duration, original interaction duration, and feedback interaction data are associated according to word identifiers and temporal adjacency. The original interaction duration is counted from the moment the current question enters the answerable state until the answer submission event for that question is included in the statistics. When an incorrect answer was given and feedback was output before the submission of the answer, and the answer is resubmitted within the same question, the aforementioned timing interval includes the time for feedback display, feedback waiting, and invalid operations during the feedback period. When no feedback was output before the submission of the answer, or when the answer is a follow-up answer and its timing starts after the feedback ends, the original interaction time does not include the feedback-related time, and the time correction of S6 does not deduct from this record. The answer result, answer time, and error record corresponding to the incorrect answer constitute learning performance data. The error record also saves the word identifier, the answer event, and the feedback record associated with the error, enabling subsequent processing to trace the complete interaction process before and after the error occurred.
[0022] During the data collection process, the integrity of event identifiers, time sequences, and word identifiers is checked. If an event lacks a word identifier, the associated object is restored based on the adjacent question display event and answer submission event. If the event time format is inconsistent, it is first converted to the time sequence marker uniformly used by the literacy practice interface. If the associated object cannot be restored, the event sequence conflicts, or the listening interface has not completed transmission, the corresponding event is written to the exception log and awaits subsequent data collection. The event is not directly added to the recent behavior sequence, thereby avoiding unsourced events from changing subsequent state judgments.
[0023] S102: Sort the data by time based on continuous interaction data and learning performance data, and filter records falling into the recent time window based on the current answer event. The time sorting is performed in the order of question display, answer operation, feedback start, operation during feedback, feedback end, and follow-up answer, so that continuous events of the same question remain adjacent. Then, according to the word identifier, associate the interface events in the continuous interaction data with the answer results and answer time in the learning performance data. Each answer is organized into a behavioral unit containing word object, answer result, answer time, feedback type, error type determined and written back by S2 when the answer is identified as an error, and the operation during feedback, and written into the recent behavior sequence in order of occurrence time.
[0024] When the same word or phrase appears multiple times within a recent time window, each behavior unit is retained separately, without later records overwriting earlier ones. When the same event is uploaded repeatedly, duplicates are identified based on the event identifier, word identifier, and occurrence time, and only records that can be linked to the complete response process are retained. When learning performance data and continuous interaction data cannot be matched, the format of the word identifier and event time is first checked. Records that still cannot be matched after format correction are added to the abnormal records and do not participate in the current aggregation. The resulting recent behavior sequence retains both the order of behavior occurrence and the correspondence between each response and feedback interaction, and serves as a common input for error classification, feedback mismatch judgment, and subsequent behavior correction.
[0025] S103: Based on the recent behavior sequence, summarize the answer results and answer time corresponding to each word identifier, and combine them with the feedback interaction data corresponding to the same word to form a multi-dimensional state vector. The recent accuracy rate corresponding to each word identifier is formed by comparing the correct answer records of that word with all valid answer records, and is used to describe the mastery status of that word within the recent time window. The overall recent accuracy rate is formed by summarizing all valid answer results within the recent behavior sequence, and is used to describe the overall answer status of the current practice process. The average answer time is formed by summarizing the effective answer time and averaging it according to the number of valid answer records, and is used to describe the time investment level in the answer process. The answer time in the behavior unit is uniformly calculated using the cognitive answer time caliber after deducting feedback display, feedback waiting, and invalid operations during feedback; behavior units that have not yet completed the S6 time correction and whose original interaction time includes feedback-related time should first complete the deduction according to the S602 method before participating in the summary. Behavior units that have not completed the deduction are not included in this average answer time, and the data integrity of this state item is recorded in the multi-dimensional state vector.
[0026] The average number of invalid clicks during feedback is determined by filtering click events during the feedback period based on the start and end times, then distinguishing between valid operations and invalid clicks based on whether the click location is within the interface's functional control area. The sum of invalid clicks for each feedback period is averaged based on the number of behavioral units with feedback interaction data in the recent behavior sequence, used to describe the concentration of off-interface operations during the feedback period. Error distribution dispersion is formed based on the distribution of error records among different word identifiers. When errors are concentrated in a small number of words, the state indicates weak local knowledge; when errors are scattered across multiple words, the state indicates fluctuations in the overall learning state. Each state information is written into a multi-dimensional state vector according to a fixed field order, and the corresponding recent behavior sequence version identifier is saved, ensuring that subsequently read state information is updated at the same time as the behavior record. Behavioral units lacking response results or with time-consuming responses do not participate in the formation of corresponding state items, but their usable feedback interaction information is retained, and the data integrity of that state item is recorded in the state vector.
[0027] Please see Figure 1 and Figure 2S2: Responding to incorrect answer signals, based on recent behavioral sequences, it determines the current error type as a short-duration, high-mastery error, a delayed recall error, or a recurring, low-accuracy error, and calculates a dynamic cognitive ability index based on a multi-dimensional state vector. The incorrect answer signal is generated when the answer to the current question is identified as incorrect, carrying the identifier of the currently answered word and the current answer event. Error classification first determines the recent mastery status of the current word, then further subdivides the error cause by combining the deviation direction of the current answer time from the benchmark average answer time. The dynamic cognitive ability index is current cognitive state information formed by jointly processing the overall recent correctness, answer speed, and error distribution, used to further differentiate the depth and intensity of feedback content when error types are the same.
[0028] S201: Read the behavior units with the same identifier as the current answer word based on the recent behavior sequence, calculate the recent accuracy rate corresponding to the current answer word, and read the current answer time from the current answer event. The current answer time adopts the cognitive answer time caliber after deducting feedback display, feedback waiting, and invalid operations during the feedback period, which is consistent with the time caliber used in S203 and S602; when there is feedback output before the current answer event and the corresponding record has not yet completed the time correction of S6, that is, when the current answer event is also the termination event of the interaction interval after the last feedback output, the correction of the record by S6 and the response to the current incorrect answer signal occur at the same interaction node. First, deduct the feedback waiting time and invalid click time from the original interaction time in the manner of S602, and then use the deduction result as the current answer time to participate in the formation of the time ratio. When S6 is executed subsequently, the record is not deducted again; when there is no feedback output before the current answer event, the current answer time is consistent with the original interaction time of this answer. The baseline average response time is read from the learning machine's question record, which is associated with the current word identifier and is loaded along with the question information when the question enters the literacy practice interface. The baseline average response time is used to provide a benchmark for comparing the normal response pace of similar questions, and does not directly replace the current user's actual response time.
[0029] The current response time is compared with the average response time of the benchmark question, using the same time frame, to form a time ratio. The time ratio describes the pace of the current response relative to the benchmark. When the current response time is shorter than the benchmark pace, the time ratio indicates a fast response; when the current response time is longer than the benchmark pace, the time ratio indicates a delayed response. The current word accuracy rate, current response time, average response time of the benchmark question, and time ratio together constitute error judgment data and are passed to the error type determination process. When the average response time of the benchmark question is missing or the format is unparseable, the current error record is marked as missing benchmark data and awaits re-evaluation after the question record is recovered; the error type is not determined based on incomplete comparison results.
[0030] S202: Based on the error judgment data, the recent accuracy rate is first compared with a preset accuracy rate threshold to distinguish whether the current word is in a state of recent high mastery or recent low accuracy rate. If the recent accuracy rate does not reach the preset accuracy rate condition, the current error is identified as a repeated low accuracy rate error, indicating that the same word has been repeatedly misrepresented in the recent behavior sequence, and the current error is not a single operation deviation. If the recent accuracy rate reaches the preset accuracy rate condition, the time consumption ratio is further compared with a preset time consumption threshold.
[0031] When the recent high mastery state is indicated by a time consumption ratio pointing to a rapid response state, the current error is identified as a short-time high mastery error. This indicates that the current word / phrase has a relatively stable recent correct record, but the error occurred during a short interaction, making the error closer to a rapid selection or operational deviation. When the time consumption ratio points to a delayed response state, the current error is identified as a recall-delay error. This indicates that the current word / phrase has a recent correct record, but the error still occurred after a longer recall process. This judgment order first distinguishes the mastery state by recent accuracy rate, and then further subdivides the error type under the high mastery state by time consumption ratio. Therefore, the same erroneous response signal corresponds to only one current error type. The determined current error type is written into the corresponding behavior unit in the recent behavior sequence according to the word / phrase identifier and the current response event, and is retained as the error type of the error record for subsequent feedback mismatch statistics filtering. When comparison data is missing, the classification does not continue across the missing conditions, but instead retains the erroneous response signal and returns to the corresponding data collection process for completion, leaving the error type field of the corresponding behavior unit empty.
[0032] S203: The overall recent accuracy, average response time, and error distribution dispersion are read from the multi-dimensional state vector. When performing negative correlation normalization on the average response time, the time status of different questions is first converted into time status information that can be jointly evaluated, based on the unified comparison criteria for response time within the literacy practice task. Then, shorter and more effective cognitive response times are assigned to higher normalized time statuses, and longer and more effective cognitive response times are assigned to lower normalized time statuses. This processing only changes the comparison direction and scale of time information, without altering the original response time records.
[0033] The overall recent accuracy rate and normalized response time data are combined into a joint state based on a preset weighted relationship. This weighted relationship is retrieved from the cognitive assessment configuration and used to determine the relative roles of correct states and response pace within the joint state. The joint state is then corrected based on the dispersion of the error distribution. When errors are concentrated in a small number of words, the correction process preserves the overall cognitive state while highlighting local error characteristics. When errors are scattered across multiple words, the correction process makes the joint state more reflective of overall fluctuations. The corrected result serves as a dynamic cognitive ability index and is transmitted to the feedback decision-making process along with the current error type. When the data integrity of the multidimensional state vector fails the check, only state items with a clear source and consistent update time are used; state items formed from different sequence versions are not used for joint processing.
[0034] Please see Figure 1 , Figure 2 and Figure 3 S3: Based on recent behavior sequences, current error types, and dynamic cognitive ability indices, filter error records that belong to short-duration, high-mastery errors or delayed recall errors and correspond to complete animated feedback. Statistically analyze the number of error records, the average playback duration of complete animated feedback, and the frequency of invalid clicks. Calculate the feedback mismatch index and determine the target feedback level and correction intensity based on the comparison between the feedback mismatch index and a preset mismatch threshold. The feedback mismatch index describes the correspondence between the cumulative playback investment of complete animated feedback and invalid clicks during the feedback period. When complete animated feedback continuously occupies the interaction process and is accompanied by a large number of invalid clicks, the mismatch between the feedback content and the current learning state increases. The target feedback level represents the display depth of the interface feedback content, from brief prompts to complete error correction instructions. The correction intensity represents the degree of control the feedback process exerts over playback completeness, user navigation, and subsequent response constraints.
[0035] S301: First, determine whether the current error type belongs to either a short-duration, high-mastery error or a delayed recall error. If the current error type belongs to either of these two categories, initiate the feedback mismatch statistics. Read error records based on recent behavior sequences. The error type of each error record is determined by S2 when the error occurred and written into the recent behavior sequence along with the corresponding behavior unit, and retained. Based on this, filter error records with the error type of short-duration, high-mastery error or delayed recall error. The statistical scope includes all error records in the recent behavior sequence whose error type belongs to either of these two categories, not limited to records with the same error type as the current error. If the current error type is a recurring low-accuracy error, do not initiate the feedback mismatch statistics. The current error directly enters the feedback decision based on the current error type and the dynamic cognitive ability index. Similarly, error records with the error type of recurring low-accuracy errors in the recent behavior sequence are not included in the feedback mismatch statistics because this type of error corresponds to insufficient continuous mastery, and its feedback requirement is directly involved in the feedback decision based on the error type and the dynamic cognitive ability index. For each error record entering the filtering scope, read the feedback playback type. The feedback playback type is jointly determined by the feedback content format and playback status saved at the time of feedback output.
[0036] A complete animated feedback indicates that the feedback content enters the playback process in a preset animation order and has a identifiable feedback start time, end time, and playback status. Error records with a complete animated feedback playback type are written to the candidate error record, retaining the word identifier, error type, feedback start time, feedback end time, and click events during the feedback. If a feedback record lacks a start time, end time, or playback status, it is not considered a complete animated feedback record but is written to the feedback record exception item awaiting completion. This filtering allows each record in the candidate error record to support retrospective calculation of playback duration and invalid clicks.
[0037] S302: Count the number of error records based on the candidate error records, and read the feedback start time, feedback end time, and click events during the feedback from each candidate error record. Calculate the single feedback playback duration based on the continuous time interval from the feedback start time to the feedback end time. When the playback status indicates that the feedback is interrupted, the feedback end time is the time corresponding to the actual stop playback event, ensuring that the single feedback playback duration matches the actual interface occupancy.
[0038] The operable area boundaries are read based on the functional control configuration of the literacy practice interface. Click events occurring between the feedback start time and feedback end time, and where the click location is outside the functional control area, are identified as single invalid clicks. The functional control area includes the interface area that can produce effective control results during the feedback process; clicks within this area are retained as valid feedback operations and are not counted as invalid clicks. The average playback duration of the single feedback corresponding to each candidate error record is calculated by averaging the number of candidate error records to obtain the average playback duration of the complete animation feedback. The ratio of the sum of the number of single invalid clicks corresponding to each candidate error record to the sum of the single feedback playback duration corresponding to each candidate error record is used as the invalid click frequency, allowing invalid click behaviors under different playback durations to be compared on the same time scale.
[0039] The total playback duration, comprised of the number of error records and the average playback duration of the complete animation feedback, describes the cumulative feedback playback time occupied by candidate error records. Under the above statistical criteria, this total playback duration equals the sum of the single feedback playback durations of each candidate error record. The total playback duration, comprised of the total playback duration and the invalid click frequency, forms the cumulative invalid click count, describing the cumulative off-screen operations that occur during the continuous playback of the complete animation feedback. Under the above statistical criteria, this cumulative invalid click count is numerically equal to the sum of the number of invalid clicks during the feedback period of each candidate error record. The average playback duration of the complete animation feedback characterizes the level of feedback playback investment for candidate error records, while the invalid click frequency characterizes the density of invalid clicks per unit of feedback playback time. These two, combined with the number of error records in the above order, ensure that feedback playback investment and invalid click density jointly determine the cumulative invalid click count. The cumulative invalid click count is converted into a feedback mismatch index according to a preset mismatch normalization benchmark. The preset mismatch normalization benchmark is read from the feedback rule configuration and its function is to convert the cumulative invalid click status formed within different recent time windows into a unified mismatch comparison criterion. When the candidate error record is empty, a mismatch judgment based on the complete historical animation feedback is not generated. The current error enters the basic feedback level according to the current error type and dynamic cognitive ability index, and the new record is included in the subsequent update after the current feedback is output.
[0040] S303: Compare the feedback mismatch index with a preset mismatch threshold to determine whether the playback input of the complete animation feedback meets the preset mismatch condition. If the preset mismatch condition is met, it indicates a strong correlation between the continuous playback of the complete animation feedback and invalid clicks in the candidate error records. The feedback decision then reduces its reliance on the complete animation playback and shifts to a shallower, more compact feedback control that allows users to proceed to subsequent responses earlier. If the preset mismatch condition is not met, the feedback decision maintains a basic feedback depth matching the current error type and determines the explanatory content and error correction constraints based on the dynamic cognitive ability index.
[0041] When the dynamic cognitive ability index points to a relatively stable cognitive state and the current error type is a short-duration, high-mastery error, the target feedback level focuses on indicating the error location and correct recognition cues, while the error correction intensity focuses on preventing repeated rapid mis-touches and does not prolong animation processes that are mismatched with the current mastery state. When the current error type is a recall delay error, the target feedback level preserves the complete connection between the character shape, meaning, or recognition cues, and the error correction intensity maintains the necessary feedback reading process. The feedback mismatch index, dynamic cognitive ability index, and current error type jointly determine the target feedback level and error correction intensity according to the correspondence in the feedback decision configuration, ensuring that the feedback decision is simultaneously constrained by the historical feedback interaction state and the current cognitive state. When comparison conditions or cognitive states are missing, the basic feedback level is retained, and the error correction intensity is not increased based on the incomplete state.
[0042] Please see Figure 1 and Figure 2 S4: Adjusts the display depth, display duration, forced playback attribute, and jump attribute of the feedback content by invoking the target feedback level and error correction intensity. Outputs the interface feedback content to the learning machine and generates a feedback output record. Display depth controls the explanatory level included in the feedback content; display duration controls whether the feedback content remains visible or plays continuously; forced playback attribute controls whether the feedback needs to complete a predetermined playback stage; and jump attribute controls whether actions such as moving to the next question, re-answering, or exiting the feedback are allowed during the feedback period. All control attributes work together on the same feedback content to avoid conflicts between display level and interaction constraints.
[0043] S401: Based on the target feedback level and error correction intensity, read the association relationships corresponding to display depth, display duration, forced playback attribute, and jump attribute from the feedback control configuration. The target feedback level first determines the components that need to be displayed in the feedback content, and the error correction intensity then determines the continuous state of each component and the user-operable boundaries. When the display depth is shallow, the feedback content retains error prompts and direct correction information. When the display depth is deep, the feedback content further retains word recognition clues and their interpretation order. When the error correction intensity is high, the forced playback attribute maintains the necessary feedback stage, and the jump attribute is only enabled after this stage is completed. When the error correction intensity is low, the forced playback attribute reduces the content that must be played continuously, and the jump attribute is enabled according to the feedback state.
[0044] The above-mentioned correlation results are organized into feedback control parameters. Feedback control parameters are ordered control information associated with the current error record, including the feedback content selection result, the display process control result, and the available jump status, and also save the target feedback level and error correction intensity corresponding to their source. When the correlation configuration is missing or there are conflicts between attributes, the basic control relationship that can display error correction information and allow normal continuation of the response is retained first, while the configuration anomaly is recorded, and conflicting parameters are not directly sent to the interface output process.
[0045] S402: Adjust the display depth, display duration, forced playback attribute, and jump attribute of the feedback content according to the feedback control parameters, and associate the adjusted feedback content with the output position of the learning machine's literacy practice interface. The output position is read from the interface layout information of the current question, ensuring that the feedback content connects to the currently incorrect word and its answer area, without covering functional controls that still need to receive operations. During the display process, the feedback content to be displayed is loaded sequentially according to the feedback control parameters, and the forced playback attribute and jump attribute are applied synchronously.
[0046] Before outputting the interface, check whether the feedback content resources, output position, and control attributes can be parsed by the current literacy practice interface. If the feedback content resources are missing, output basic text correction information corresponding to the current error type instead. If the output position is unavailable, reread the valid feedback area for the current question. If the transmission of control attributes fails, retain the current feedback content and restore the basic jump state, while recording the interface exception. After completing the checks, output the interface feedback content so that the user can receive feedback in the interactive context of the current erroneous word.
[0047] S403: Collect the start time, end time, and playback status based on the interface feedback content. The start time is determined by the event when the feedback content becomes visible or playable, and the end time is determined by the event when the feedback content finishes playing, is allowed to jump to another screen, or stops due to interface switching. The playback status is used to distinguish between normal completion, mid-play stop, and output error. Establish a correlation between the interface feedback content, start time, end time, playback status, current word identifier, and corresponding error record to generate a feedback output record.
[0048] Feedback output records serve as time boundaries and object indices for subsequent interactive action collection. If a start event has occurred but an end event is missing, the actual termination position is supplemented based on subsequent question switching events or interface closing events, and the playback status is marked as abnormal termination. If the feedback content fails to enter a visible state, a normal playback record is not generated; instead, an output failure record is retained for subsequent exception handling. This process of outputting feedback content and generating records allows subsequent processes to distinguish between actions taken during feedback and normal response actions.
[0049] Please see Figure 1 and Figure 2S5: Based on feedback output records, collect interactive actions during and after the feedback output, and match these actions with preset interface-induced behavior features to obtain interface-induced behavior markers. Interface-induced behavior is an operational behavior caused by feedback display occupancy, feedback control limitations, or interaction transitions after the feedback ends. This includes continuous clicking in non-functional areas, swiping that does not correspond to the current function, and abnormal response intervals that occur after the feedback ends. Interface-induced behavior markers are used to indicate a traceable correlation between the corresponding interactive action and the interface feedback process. Subsequent corrections use these markers to distinguish between cognitive response time and waiting or interference time caused by feedback.
[0050] S501: Based on the feedback output record, read the start time, end time, and playback status, and collect interactive actions during and after the feedback output through the event listening interface of the learning machine's literacy practice interface. The feedback output period is defined by the feedback start time and feedback end time. The interactive interval after the feedback output begins at the feedback end event and continues until the earliest event occurring among re-answering, moving to the next question, or leaving the current question. This definition ensures that clicks, swipes, and answer waits after feedback correspond to the immediately following feedback output record and are not mixed with operations of subsequent independent questions.
[0051] The system saves the action type, occurrence time, click location, swipe trajectory, and associated word identifiers for each collected interaction action. It then establishes a connection with the feedback output record in chronological order to obtain the interaction action record. When playback is abnormally terminated, the system still collects interaction actions based on the actual start time and actual termination location, but retains the abnormal playback state in the interaction action record. When interaction event transmission fails, the system detects gaps based on the event identifier and requests a resubmission through the event listening interface. Actions that cannot be re-collected are not included in feature matching, and the data missing state is retained in the interaction action record.
[0052] S502: Obtain preset interface inducement behavior features from the inducement behavior configuration of the literacy practice interface. These features consist of a preset click frequency range, a preset non-functional area range, a preset sliding trajectory feature, a preset answer interval threshold, and a preset hit count threshold. The preset non-functional area range is formed based on the complement of the current interface's functional control areas. The preset sliding trajectory feature is formed based on the sliding direction, start and end areas, and trajectory continuity received by the current feedback page. The remaining comparison conditions are read from the same inducement behavior configuration.
[0053] Click frequency, click location, swipe trajectory, and response interval are extracted from the interaction action records. Click frequency is compared with a preset click frequency range to generate click frequency hit data. Click location is compared with a preset non-functional area range to generate click location hit data. Swipe trajectory is compared with preset swipe trajectory features to generate swipe trajectory hit data. Response interval is compared with a preset response interval threshold to generate response interval hit data. Each hit data item retains the corresponding original interaction action and comparison condition source, avoiding the loss of behavioral evidence while only retaining the matching conclusion.
[0054] The system counts the number of hits in each data point and compares it to a preset hit threshold. When the number of hits reaches the preset hit threshold, a flag indicating an interface-induced behavior is generated, and the type of the hit action is written to the flag. When the preset hit threshold is not reached, a flag indicating that the behavior is not recognized as an interface-induced behavior is generated, while the original interaction action is retained for learning behavior recording. If any feature lacks available input, that feature does not produce a hit result. The count and comparison of the number of hits are only performed based on features that have been matched, and the range of features involved in the matching and their data integrity status are recorded in the interface-induced behavior flag. When there are insufficient available features to complete the determination, the flag is recorded as pending verification, and the interaction action is not directly identified as an interface-induced behavior.
[0055] Please see Figure 1 , Figure 2 and Figure 4 S6: Based on the interface-induced behavior markers and feedback output records, correct the interactive actions during feedback in the continuous interaction data and the response time and error records in the learning performance data to obtain learning behavior records. Then, update the recent behavior sequence and real-time multidimensional state vector based on these learning behavior records. The real-time multidimensional state vector is the current state information reconstructed using the corrected response data. Its field structure is consistent with the multidimensional state vector generated in S103, but the associated recent behavior sequence has been updated with corrected response time and error weights. This process ensures that feedback waiting and interface-induced operations are no longer directly included in the state evaluation as cognitive response processes, while preserving the connection between the original data and the basis for correction.
[0056] S601: Based on the interface-induced behavior markers and feedback output records, filter the interactive actions occurring between the feedback start time and feedback end time from the continuous interaction data, and also read the interactive actions collected in S5 within the interaction interval after feedback output that are matched by the interface-induced behavior markers. During filtering, verify the word identifiers, feedback output records, and event times to prevent operations of adjacent questions from entering the current feedback interval. Establish a correlation between the interactive actions during the feedback period and the aforementioned interactive actions after feedback output with the interface-induced behavior markers to obtain interference action records. The interactive actions after feedback output are used to determine the abnormal answer intervals that occur after the feedback ends and the continued click intervals located in non-functional areas, and serve as interactive evidence for answer time correction and error record weight adjustment. The interference action records save the feedback output record identifier corresponding to the feedback waiting interval, the forced playback attribute and jump attribute status in the feedback control parameters associated with the corresponding error record, click events located in non-functional areas, sliding events not corresponding to the feedback control, and the order of occurrence of each event.
[0057] Before using the interface-induced behavior marker, check if the marker has a complete source of interactive actions, matching results, and feedback output records. When the marker is available, determine the interactive actions during the feedback period and the interfering parts of the interactive actions after the feedback output according to the hit type. When the marker is in the pending verification state, retain the original interactive data and record the exception, without directly deducting the time consumption. When the feedback output record cannot match the interactive action, re-establish the connection based on word identifiers and adjacent events. Records that still cannot match are entered into exception handling, without inferring interfering actions across questions.
[0058] S602: Calculate the feedback waiting time and invalid click duration based on the interference action records and their associated feedback output records. The feedback waiting time is formed by taking the continuous interval during which the feedback is in a forced playback or jump state that is not yet open, based on the start time, end time, playback status, and the forced playback and jump attributes in the feedback control parameters in the feedback output records. During this interval, the user cannot enter a valid response. The invalid click duration is determined based on the interference action records. The invalid click duration is formed by the invalid click event triggered by the interface and its continuous operation interval, excluding operations within functional controls that produce valid control results. The two types of durations are calculated independently, based on the feedback output records and event times, and their source actions are saved separately.
[0059] The pure cognitive time is obtained by subtracting the feedback waiting time and invalid click time from the original interaction time of the re-answered record within the same question associated with the feedback output record. Error-triggered answer records are not included in this deduction because their timing interval ends before the feedback output, and subsequent answer records are not included because their timing start point is after the feedback ends. The pure cognitive time only retains the time interval directly related to recognition, recall, and answer submission after the question enters the answerable state. When the feedback waiting interval and invalid click interval overlap, only the overlapping portion is deducted once to prevent repeated correction of the same time interval. If the deduction result is inconsistent with the sequence of events, the process reverts to the cognitive time interval that can be proven by a valid answer event, while retaining the original interaction time for retrospective purposes. The corrected answer time uses the pure cognitive time.
[0060] Weight adjustment values are generated based on error records and interface-induced behavior markers. When there is a clear correlation between error records and interface-induced behaviors, the weight adjustment process weakens the direct impact of the error on knowledge mastery while preserving the error fact and the source of the induced behavior. When an error record does not match the characteristics of interface-induced behaviors, its original error effect is maintained. The corrected response time, error records, weight adjustment values, word tags, and corresponding feedback output records are correlated to generate corrected response data. This results in the deduction of pure cognitive time, ensuring that the correction of response time and error record correction use the same interactive evidence.
[0061] S603: Write the corrected response data into the corresponding behavior unit in the recent behavior sequence. During writing, the word identifier, response event, and feedback output record are used as the location basis, replacing the response time and error weight used for state calculation, while retaining the original interaction time, original error record, and correction basis. If the corresponding behavior unit cannot be found, do not create a new behavior record lacking the original source; instead, return to the association process to re-examine the event identifier.
[0062] Based on the recently written recent behavior sequence, the system reconstructs the recent correct state of each word, the overall recent correct state, the average response time, the invalid click state during feedback, and the error distribution state, and updates the real-time multidimensional state vector in the original field order. Subsequently, it establishes a correlation between the corrected response data, the recent behavior sequence, the real-time multidimensional state vector, and the feedback output record to obtain the learning behavior record. The learning behavior record simultaneously saves the data before correction, the data after correction, and the source of correction, enabling S1 to regenerate the recent behavior sequence and state information based on the updated behavior record when the next erroneous response signal arrives, thus forming a continuous data feedback loop.
[0063] Please see Figure 5A human-computer interaction system for a learning machine user interface includes a behavior state aggregation module, a cognitive diagnosis and evaluation module, a mismatch feedback decision module, a feedback interface adjustment module, an induced behavior capture module, and a behavior record update module. The behavior state aggregation module corresponds to S1, the cognitive diagnosis and evaluation module corresponds to S2, the mismatch feedback decision module corresponds to S3, the feedback interface adjustment module corresponds to S4, the induced behavior capture module corresponds to S5, and the behavior record update module corresponds to S6. Each component sequentially receives the output of the previous process according to the data flow relationship formed by the method steps and sends the learning behavior record back to the behavior state aggregation process.
[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A human-computer interaction method for a learning machine user interface, characterized in that, Includes the following steps: S1: Collect continuous interaction data and learning performance data through the learning machine's literacy practice interface, and aggregate the continuous interaction data and learning performance data according to the recent time window to generate a recent behavior sequence and a multi-dimensional state vector. S2: In response to the error response signal, determine the current error type based on the recent behavior sequence as a short-duration high-mastery error, a delayed recall error, or a repetitive low-accuracy error, and calculate the dynamic cognitive ability index based on the multidimensional state vector; S3: Based on the recent behavior sequence, the current error type, and the dynamic cognitive ability index, filter error records that belong to short-duration high-mastery errors or recall delay errors and correspond to complete animation feedback, count the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks, calculate the feedback mismatch index based on the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks, and determine the target feedback level and error correction intensity based on the comparison result of the feedback mismatch index and the preset mismatch threshold; S4: Invoke the target feedback level and the error correction intensity to adjust the display depth, display duration, forced playback attribute and jump attribute of the feedback content, output the interface feedback content of the learning machine, and generate feedback output records; S5: Based on the feedback output record, collect the interactive actions during and after the feedback output, and match the interactive actions with preset interface inducement behavior features to obtain interface inducement behavior markers; S6: Based on the interface-induced behavior marker and the feedback output record, correct the interactive actions during the feedback period in the continuous interaction data and the answering time and error records in the learning performance data to obtain the learning behavior record, and update the recent behavior sequence and real-time multidimensional state vector based on the learning behavior record.
2. The human-computer interaction method for a learning machine user interface according to claim 1, characterized in that, The process of S1 is as follows: S101: Collect answer behavior data, feedback interaction data, and follow-up answer data as continuous interaction data through the event listening interface of the learning machine's literacy practice interface. For each answer behavior data, associate word identifier, answer result, answer time, original interaction time, and feedback interaction data. The answer time and the error record corresponding to the answer result being incorrect constitute learning performance data. S102: Based on the continuous interaction data and the learning performance data, sort the data by time using a recent time window, filter the continuous interaction data and learning performance data that fall into the recent time window, associate the answer results and answer time in the continuous interaction data and learning performance data according to word identifiers, and generate a recent behavior sequence; S103: Based on the recent behavior sequence, the answer result and answer time corresponding to each word identifier are statistically analyzed. Combined with the feedback interaction data, the recent accuracy rate corresponding to each word identifier is calculated. The overall recent accuracy rate, average answer time, average number of invalid clicks during the feedback period, and error distribution dispersion are calculated to generate a multi-dimensional state vector.
3. The human-computer interaction method for a learning machine user interface according to claim 1, characterized in that, The process of S2 is as follows: S201: Based on the answer results corresponding to the current answer word in the recent behavior sequence, the recent accuracy rate corresponding to the current answer word is obtained, the current answer time corresponding to the current answer word is read, the average answer time of the questions in the learning machine question record is read, the time ratio of the current answer time to the average answer time of the questions is calculated, and the error judgment data is obtained. S202: Based on the error determination data, compare the recent accuracy rate with the preset accuracy rate threshold, compare the time consumption ratio with the preset time consumption threshold, and determine the current error type among short-time high-mastery errors, recall delay errors, and repetitive low-accuracy errors; S203: Based on the multidimensional state vector, read the overall recent accuracy rate, average response time, and error distribution dispersion. Perform negative correlation normalization on the average response time to obtain normalized time data. Calculate the weighted sum of the overall recent accuracy rate and the normalized time data. Adjust the weighted sum according to the error distribution dispersion to obtain the dynamic cognitive ability index.
4. The human-computer interaction method for a learning machine user interface according to claim 1, characterized in that, The process of S3 is as follows: S301: Based on the recent behavior sequence and the current error type, filter error records where the current error type is a short-time high-mastery error or a recall delay error, read the feedback playback type corresponding to each error record, filter error records where the feedback playback type is a complete animation feedback, and obtain candidate error records; S302: Based on the candidate error records, count the number of error records, obtain the average playback duration and invalid click frequency of the complete animation feedback, calculate the total playback duration corresponding to the number of error records and the average playback duration of the complete animation feedback, calculate the cumulative invalid clicks corresponding to the total playback duration and the invalid click frequency, and calculate the feedback mismatch index; S303: Based on the comparison between the feedback mismatch index and the preset mismatch threshold, the target feedback level and error correction intensity are determined according to the comparison result, the dynamic cognitive ability index and the current error type.
5. The human-computer interaction method for a learning machine user interface according to claim 4, characterized in that, The calculation process for the feedback mismatch index is as follows: The system reads the feedback start time, feedback end time, and click events during the feedback period for each candidate error record. It calculates the single feedback playback duration corresponding to the feedback start time and feedback end time, counts the number of click events during the feedback period that occur between the feedback start time and feedback end time and are outside the preset interface function control area, and obtains the single invalid click count. It calculates the average of the single feedback playback durations corresponding to each candidate error record to obtain the average playback duration of the complete animation feedback. It calculates the ratio of the sum of the single invalid click counts corresponding to each candidate error record to the sum of the single feedback playback durations corresponding to each candidate error record, and obtains the invalid click frequency. It calculates the total playback duration corresponding to the number of error records in the candidate error records and the average playback duration of the complete animation feedback, and calculates the cumulative invalid click count corresponding to the total playback duration and the invalid click frequency. Based on the ratio of the cumulative invalid click count to a preset mismatch normalization benchmark, it obtains the feedback mismatch index.
6. The human-computer interaction method for a learning machine user interface according to claim 1, characterized in that, The process of S4 is as follows: S401: Based on the target feedback level and the error correction intensity, establish the association between the target feedback level, error correction intensity and display depth, display duration, forced playback attribute, and jump attribute, and generate feedback control parameters; S402: Adjust the display depth, display duration, forced playback attribute and jump attribute of the feedback content according to the feedback control parameters, associate the feedback content with the output position of the learning machine's literacy practice interface, and output the interface feedback content. S403: Collect the start time, end time, and playback status based on the interface feedback content, establish the association between the interface feedback content, start time, end time, and playback status, and generate a feedback output record.
7. The human-computer interaction method for a learning machine user interface according to claim 1, characterized in that, The process of S5 is as follows: S501: Read the start time, end time and playback status according to the feedback output record, collect the interactive actions during and after the feedback output through the event listening interface of the learning machine's literacy practice interface, associate the interactive actions with the feedback output record, and obtain the interactive action record; S502: Obtain preset interface inducement behavior features, extract the click frequency, click position, sliding trajectory and response interval from the interaction action record, and match the click frequency, click position, sliding trajectory and response interval with the preset interface inducement behavior features to obtain interface inducement behavior markers.
8. The human-computer interaction method for a learning machine user interface according to claim 7, characterized in that, The process of obtaining the interface-induced behavior marker is as follows: The system reads click frequency, click location, swipe trajectory, and response interval from the interaction action record. It compares the click frequency with a preset click frequency range to obtain click frequency hit data, compares the click location with a preset non-functional area range to obtain click location hit data, compares the swipe trajectory with a preset swipe trajectory feature to obtain swipe trajectory hit data, and compares the response interval with a preset response interval threshold to obtain response interval hit data. It then counts the number of hits in the click frequency hit data, click location hit data, swipe trajectory hit data, and response interval hit data, compares the number of hits with a preset hit count threshold, and obtains the interface-induced behavior marker based on the comparison result.
9. The human-computer interaction method for a learning machine user interface according to claim 1, characterized in that, The process of S6 is as follows: S601: Based on the interface-induced behavior marker and the feedback output record, filter the interaction actions during the feedback period from the continuous interaction data, associate the interaction actions during the feedback period with the interface-induced behavior marker, and obtain the interference action record; S602: Based on the statistics of the interference action record, the feedback waiting time and invalid click time are obtained. The feedback waiting time and invalid click time are deducted from the original interaction time to obtain the pure cognition time. The pure cognition time is used as the corrected response time. The weight adjustment value corresponding to the error record and the interface inducement behavior mark is calculated. The corrected response data is generated based on the corrected response time and the weight adjustment value. S603: Write the corrected response data into the recent behavior sequence, update the multidimensional state vector according to the written recent behavior sequence, establish the association between the corrected response data and the recent behavior sequence, and obtain the learning behavior record.
10. A human-computer interaction system for a learning machine user interface, characterized in that, The system is used to implement the human-computer interaction method for the user interface of a learning machine according to any one of claims 1-9, the system comprising: Behavior state aggregation module: Collects continuous interaction data and learning performance data through the learning machine's literacy practice interface, and aggregates the continuous interaction data and learning performance data according to the recent time window to generate recent behavior sequences and multi-dimensional state vectors; Cognitive diagnosis and assessment module: In response to incorrect answer signals, it determines the current error type based on the recent behavior sequence as a short-time high-mastery error, a delayed recall error, or a repetitive low-accuracy error, and calculates the dynamic cognitive ability index based on the multidimensional state vector; The mismatch feedback decision module: Based on the recent behavior sequence, the current error type, and the dynamic cognitive ability index, it filters error records that belong to short-duration high-mastery errors or recall delay errors and correspond to complete animation feedback. It counts the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks. Based on the number of error records, the average playback duration of complete animation feedback, and the frequency of invalid clicks, it calculates the feedback mismatch index. Based on the comparison result of the feedback mismatch index and the preset mismatch threshold, it determines the target feedback level and error correction intensity. Feedback Interface Adjustment Module: Invokes the target feedback level and the error correction intensity to adjust the display depth, display duration, forced playback attribute and jump attribute of the feedback content, outputs the interface feedback content of the learning machine, and generates feedback output records; Induced Behavior Capture Module: Based on the feedback output record, collect the interactive actions during and after the feedback output, and match the interactive actions with preset interface induced behavior features to obtain interface induced behavior markers; Behavior record update module: Based on the interface-induced behavior markers and the feedback output records, the module corrects the interactive actions during the feedback period in the continuous interaction data and the response time and error records in the learning performance data to obtain learning behavior records, and updates the recent behavior sequence and real-time multidimensional state vector based on the learning behavior records.
