An adaptive streaming method, system, and application product

By acquiring user language unit status data and calculating the language unit load of video data, and filtering video data of appropriate difficulty, the problem of overestimating user ability in recommended content in existing technologies is solved, achieving a more efficient language learning effect.

CN123075223APending Publication Date: 2026-11-10BEIJING SHIQU QIMING TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611347622.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-01
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing language learning products fail to accurately reflect a user's specific language proficiency when watching video learning materials, leading to recommended content that overestimates the user's ability and results in poor learning outcomes.

Method used

By acquiring the user's language unit status data, calculating the language unit load of video data, filtering out video data that meets preset recommendation conditions, and forming a playback sequence of appropriate difficulty, including overall and local language unit load calculation, and dynamically adjusting the language level and video data recommendations.

Benefits of technology

This improves the effectiveness and efficiency of video learning, ensures that recommended content matches the user's actual comprehension ability, and avoids learning difficulties caused by the system overestimating the user's abilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123075223A_ABST
    Figure CN123075223A_ABST
Patent Text Reader

Abstract

This disclosure relates to an adaptive streaming method, system, and program product. The method includes acquiring a user's language unit state data, acquiring multiple candidate video data sets and language unit metadata for each candidate video data set, calculating the language unit load of each candidate video data set relative to the user based on the language unit state data and the language unit metadata, determining video data sets that meet preset recommendation criteria from the multiple candidate video data sets based on the language unit load, and forming or updating a sequence of videos to be played, containing one or more video data sets, based on the video data sets that meet the preset recommendation criteria. This method enables users to obtain video data of appropriate difficulty, avoiding situations where the system overestimates the user's language ability, resulting in overly difficult recommended content, and improving the matching degree between the difficulty of recommended video data and the user's actual comprehension ability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of computer-aided language learning technology, and more specifically, to an adaptive streaming method, an adaptive streaming system, and a computer program product. Background Technology

[0002] Existing language learning products or video learning material recommendation systems typically recommend learning content based on age, grade level, course level, topic tags, human assessment, user subjective selection, overall vocabulary estimation, or other coarse-grained language ability indicators. While such recommendation methods can differentiate content difficulty to some extent, they fail to accurately reflect whether a specific user truly possesses the ability to understand the specific language content of a particular video learning material. In fact, even with traditional methods that categorize content by level, users at the same level can have vastly different abilities.

[0003] Taking children's English listening learning as an example, ordinary English animation, original children's videos, or leveled video content often assume that viewers already possess a certain amount of prior listening vocabulary. For users with insufficient specific abilities, even if they are attracted by the visuals, they may only be able to roughly guess the plot based on the images, without truly understanding the dialogue, sentence meaning, or specific language content in the video. Therefore, recommending content based solely on video level, theme, or overall difficulty estimation can easily lead the system to overestimate the user's ability, exposing the user to video learning materials that they actually cannot understand, resulting in ineffective learning or low learning efficiency. Summary of the Invention

[0004] One objective of this disclosure is to provide a novel adaptive streaming solution that pushes videos of appropriate difficulty to users, thereby improving the effectiveness or efficiency of learning.

[0005] According to a first aspect of this disclosure, an adaptive streaming method is provided, comprising: The language unit status data acquisition step involves acquiring the user's language unit status data. The language unit status data is recorded in units of language units, and the language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The video metadata acquisition step involves acquiring multiple candidate video data and language unit metadata for each candidate video data. The language unit metadata includes at least the language unit information appearing in the corresponding video data. The language unit load calculation step involves calculating the language unit load of each candidate video data relative to the user based on the language unit status data and the language unit metadata. The recommended determination step involves determining video data that meets preset recommendation conditions from the plurality of candidate video data based on the language unit load. The push output step involves forming or updating a sequence of videos to be played that includes one or more of the video data, based on the video data that meets the preset recommendation conditions.

[0006] Optionally, in the language unit load calculation step, the language unit load includes the overall language unit load and / or the local language unit load; The overall language unit load includes at least one of the following: the number of unrecognized language units in a single candidate video data, the ratio of recognized to unrecognized language units in a single candidate video data, the number of language units in a single candidate video data with a recognition level below a preset threshold, and the cumulative or weighted average value of the recognition or non-recognition levels of each language unit in a single candidate video data. The local language unit load is calculated in units of continuous playback time windows or continuous language unit windows within a single candidate video data.

[0007] Optionally, in the recommendation determination step, the preset recommendation conditions include: the overall language unit load is not higher than a preset overall load threshold; and / or The load of each local window language unit does not exceed the preset local load threshold.

[0008] Optionally, the language unit load calculation step further includes: determining an allowable language unit load threshold based on the playback duration of the candidate video data, wherein the shorter the playback duration of the candidate video data, the lower the allowable language unit load threshold, and the longer the playback duration of the candidate video data, the higher the allowable language unit load threshold.

[0009] Optionally, the language unit load calculation step further includes: when multiple video data belong to the same continuous story, the same theme unit, or the same preset content set, the multiple video data are used as aggregation candidate content, and the playback duration of the aggregation candidate content, the total language unit load of the aggregation candidate content, the diversity language unit load, and the local window language unit load are calculated. Whether to recommend the aggregated candidate content is determined based on the playback duration of the aggregated candidate content, the total language unit load of the aggregated candidate content, the language unit load of the episodes, and the language unit load of the local window.

[0010] Optionally, it further includes: a language unit set determination step, which determines one or more language unit sets based on the language unit status data and preset filtering conditions, and calculates the language unit load of each candidate video data relative to the user based on the one or more language unit sets and the language unit metadata, or the one or more language unit sets, the language unit status data, and the language unit metadata.

[0011] Optionally, the push output step further includes: before playing subsequent video data in the sequence to be played, recalculating the language unit load of the subsequent video data relative to the user based on the updated language unit state data and / or language unit set; Determine whether the subsequent video data still meets the preset recommendation conditions; if it no longer does, delete, replace, or pause the playback of the subsequent video data from the sequence to be played.

[0012] Optionally, it also includes: a dynamic adjustment step, which dynamically determines the language unit load threshold of the candidate video data based on the user's current language proficiency; The user's current language proficiency is determined based on at least one of the following: the number of language units recognized, the number of language units mastered, the historical viewing time, the historical language unit verification success rate, or the historical comprehension verification success rate.

[0013] Optionally, in the language unit status data acquisition step, the language unit status data includes at least one of the following: whether the user recognizes the language, whether the user is proficient in the language, level of recognition, whether the user needs to be reinforced, number of occurrences, historical occurrence time, number of verifications, historical verification time, historical verification result, number of successful verifications, number of failed verifications, number of consecutive correct answers, number of consecutive incorrect answers, most recent verification result, or most recent verification time.

[0014] Optionally, in the recommendation determination step, if multiple candidate video data all meet the preset recommendation conditions, the recommendation order is adjusted according to at least one of the following: the number of unfamiliar language units, the number of language units with recognition level below a preset threshold, the cumulative degree of unfamiliarity, the weighted average degree of recognition or degree of unfamiliarity, and the viewing history.

[0015] According to a second aspect of this disclosure, an adaptive streaming system is also provided, comprising: The language unit status data acquisition module is used to acquire the user's language unit status data. The language unit status data is recorded in units of language units. The language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The video metadata acquisition module is used to acquire multiple candidate video data and language unit metadata of each candidate video data. The language unit metadata includes at least the language unit information appearing in the corresponding video data. The language unit load calculation module is used to calculate the language unit load of each candidate video data relative to the user based on the language unit status data and the language unit metadata. The recommendation determination module is used to determine, based on the language unit load, video data to be recommended that meets preset recommendation conditions from the plurality of candidate video data; The push output module is used to form or update a sequence of videos to be played, which includes one or more of the video data, based on the video data that meets the preset recommendation conditions.

[0016] Optionally, the language unit load calculation module is further configured to: determine the allowable language unit load threshold based on the playback duration of the candidate video data, wherein the shorter the playback duration of the candidate video data, the lower the allowable language unit load threshold, and the longer the playback duration of the candidate video data, the higher the allowable language unit load threshold.

[0017] Optionally, the language unit load calculation module is further configured to: when multiple video data belong to the same continuous story, the same theme unit, or the same preset content set, use the multiple video data as aggregation candidate content, and calculate the playback duration of the aggregation candidate content, the total language unit load of the aggregation candidate content, the diversity language unit load, and the local window language unit load; Whether to recommend the aggregated candidate content is determined based on the playback duration of the aggregated candidate content, the total language unit load of the aggregated candidate content, the language unit load of the episodes, and the language unit load of the local window.

[0018] According to a third aspect of this disclosure, a computer program product is also provided, including a computer program that, when executed by a processor, implements the method described in accordance with a first aspect of this disclosure.

[0019] According to a fourth aspect of this disclosure, a language unit verification method based on interactive behavior is provided, comprising: The language unit status data acquisition step involves acquiring the user's language unit status data. The language unit status data is recorded in units of language units, and the language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The video playback process involves playing video data that meets preset recommendation criteria to the user. The interactive behavior monitoring step involves monitoring preset interactive operations performed by the user on the video playback interface during the playback of the video data. The help event recognition step involves determining the target statement or target video segment corresponding to the preset interactive operation when the preset interactive operation is detected, based on the time correspondence between the occurrence time of the preset interactive operation and the video playback timeline. The video metadata acquisition step involves acquiring the language unit metadata of the video data, wherein the language unit metadata includes at least the language unit information appearing in the corresponding video data. The candidate language unit determination step involves determining at least one language unit appearing in the target statement or target video segment as a candidate language unit to be verified based on the language unit metadata of the video data. The verification task generation step involves generating a language unit verification task for the candidate language units to be verified and outputting it to the user. The verification result processing steps include obtaining feedback information from the user regarding the language unit verification task, and updating the user's language unit status data based on the feedback information.

[0020] Optionally, the preset interactive operation includes at least one of the following: Trigger the help interaction control set in the video playback interface, and determine the corresponding current statement, previous statement or most recently played statement based on the current playback time point, subtitle timestamp, speech recognition timeline, statement start and end time or statement identifier; Trigger sentence-level playback control operation, determine the corresponding target sentence according to the target sentence identifier, the target sentence identifier includes one or more of the following: previous sentence operation, next sentence operation, replay current sentence operation, jump to the start of current sentence operation, or jump to the start of previous sentence operation; Trigger a playback progress bar drag operation, determine the corresponding target statement or semantic segment based on the playback progress position after the operation, and drag the playback progress back to the beginning of the statement or semantic segment.

[0021] Optionally, in the verification task generation step, the language unit verification task includes: Identify the language unit type of the candidate language units to be verified; The verification task format is determined according to the language unit type. The verification task format includes recognition verification tasks based on audiovisual matching or sound-semantic matching, fill-in-the-blank and selection verification tasks based on contextual understanding, and judgment and output verification tasks based on semantic processing. If the candidate language unit to be verified is a concrete noun or an action word, then the recognition verification task based on audiovisual matching or sound-meaning matching is generated. If the candidate language unit to be verified is a function word, copula, preposition, auxiliary word, grammatical word, or abstract word, then the context-understanding-based fill-in-the-blank and selection verification task or the semantic processing-based judgment and output verification task is generated.

[0022] Optionally, the contextual understanding-based fill-in-the-blank and selection-type verification task generates the question stem context based on the set of recognized language units or the set of proficient language units in the user's language unit status data, so that the question stem context language units other than the candidate language units to be verified and the candidate options meet the preset low-load conditions.

[0023] Optionally, the context-understanding-based fill-in-the-blank and selection-type verification task includes: generating at least two sets of verification tasks under different contexts for the candidate language unit to be verified, wherein the first set of contexts is the original context of the candidate language unit to be verified in the target sentence or target video segment, and the second set of contexts is a transfer context different from the original context. The candidate language units to be verified are verified in both the original context and the transferred context. The status data of the candidate language units to be verified are updated based on the user's responses in the two contexts.

[0024] Optionally, it also includes: a translation or retelling positioning step, which involves obtaining the translation or retelling results submitted by the user in voice or text form at a preset time when the user is watching the video data; The translation or restatement result is compared with the corresponding standard meaning to determine the content that the user did not express correctly. Based on the correspondence between the incorrectly expressed content and the language units in the video data, a candidate language unit to be verified is identified or the verification priority of an existing candidate language unit to be verified is increased.

[0025] Optionally, it also includes: an interactive confirmation step, which, after determining the target statement or target video segment, presents one or more language units contained in the target statement or target video segment in an interactive form, and provides at least one of the following auxiliary content for the user to select: the audio of the statement corresponding to the language unit, the audio of the language unit, the translation of the statement, or the translation of the language unit. Obtain the auxiliary content selected by the user, and determine the language unit corresponding to the auxiliary content selected by the user as a candidate language unit to be verified or increase its verification priority.

[0026] Optionally, it also includes: a comprehension verification step, generating a language dependency comprehension verification task corresponding to the video data and outputting it to the user, wherein the language dependency comprehension verification task is a verification task that requires a response based on the dialogue, audio, target language units or sentence semantics in the video data; Obtain the user's feedback information regarding the language dependency comprehension verification task, and determine the comprehension result; If a language unit passes the validation but the language dependency comprehension validation related to that language unit fails, the weight of that language unit's validation result is reduced.

[0027] According to a fifth aspect of this disclosure, a language unit verification system based on interactive behavior is also provided, comprising: The language unit status data acquisition module is used to acquire the user's language unit status data. The language unit status data is recorded in units of language units. The language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The video playback module is used to play video data that meets preset recommendation criteria to users; An interactive behavior monitoring module is used to monitor preset interactive operations performed by the user on the video playback interface during the playback of the video data. The help request event recognition module is used to determine the target statement or target video segment corresponding to the preset interactive operation when the preset interactive operation is detected, based on the time correspondence between the occurrence time of the preset interactive operation and the video playback timeline. The video metadata acquisition module is used to acquire language unit metadata of the video data, wherein the language unit metadata includes at least the language unit information appearing in the corresponding video data; The candidate language unit determination module is used to determine at least one language unit appearing in the target sentence or target video segment as a candidate language unit to be verified based on the language unit metadata of the video data. The verification task generation module is used to generate language unit verification tasks for the candidate language units to be verified and output them to the user. The verification result processing module is used to obtain feedback information from the user regarding the language unit verification task, and update the user's language unit status data based on the feedback information.

[0028] Optionally, the preset interactive operation includes at least one of the following: triggering the help interaction control set in the video playback interface, and determining the corresponding current statement, previous statement or most recently played statement based on the current playback time point, subtitle timestamp, speech recognition timeline, statement start and end time or statement identifier; Trigger sentence-level playback control operation, determine the corresponding target sentence according to the target sentence identifier, the target sentence identifier includes one or more of the following: previous sentence operation, next sentence operation, replay current sentence operation, jump to the start of current sentence operation, or jump to the start of previous sentence operation; Trigger a playback progress bar drag operation, determine the corresponding target statement or semantic segment based on the playback progress position after the operation, and drag the playback progress back to the beginning of the statement or semantic segment.

[0029] Optionally, in the verification task generation module, the language unit verification task includes: Identify the language unit type of the candidate language units to be verified; The verification task format is determined according to the language unit type. The verification task format includes recognition verification tasks based on audiovisual matching or sound-semantic matching, fill-in-the-blank and selection verification tasks based on contextual understanding, and judgment and output verification tasks based on semantic processing. If the candidate language unit to be verified is a concrete noun or an action word, then the recognition verification task based on audiovisual matching or sound-meaning matching is generated. If the candidate language unit to be verified is a function word, copula, preposition, auxiliary word, grammatical word, or abstract word, then the context-understanding-based fill-in-the-blank and selection verification task or the semantic processing-based judgment and output verification task is generated.

[0030] Optionally, the contextual understanding-based fill-in-the-blank and selection-type verification task generates the question stem context based on the set of recognized language units or the set of proficient language units in the user's language unit status data, so that the question stem context language units other than the candidate language units to be verified and the candidate options meet the preset low-load conditions.

[0031] Optionally, the context-understanding-based fill-in-the-blank and selection-type verification task includes: generating at least two sets of verification tasks under different contexts for the candidate language unit to be verified, wherein the first set of contexts is the original context of the candidate language unit to be verified in the target sentence or target video segment, and the second set of contexts is a transfer context different from the original context. The candidate language units to be verified are verified in both the original context and the transferred context. The status data of the candidate language units to be verified are updated based on the user's responses in the two contexts.

[0032] According to a sixth aspect of this disclosure, a computer program product is also provided, including a computer program that, when executed by a processor, implements the method described in a fourth aspect of this disclosure.

[0033] According to the seventh aspect of this disclosure, a forgetting risk update method with multimodal feedback is provided, comprising: The language unit status data maintenance steps involve recording and updating in real time one or more of the following in the language unit status data: recognition status, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time. The language unit status data is recorded on a per-language-unit basis, and the language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The multimodal perception data acquisition steps involve acquiring the user's multimodal perception data through one or more of an image acquisition device and an audio acquisition device while the user is watching video data. The potential unresolved event identification step involves extracting and identifying features from the multimodal perception data to determine whether the user has generated a potential unresolved feedback event. The forgetting risk calculation step involves calculating the forgetting risk value of a language unit that meets the preset recognition conditions in the language unit status data, based on one or more of the following: the recognition status of the language unit, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time, as well as the potential ununderstood feedback events. The status update step involves adjusting the language unit status data of language units that meet the preset update conditions based on the forgotten risk value. The recommended impact step involves recalculating the language unit load of subsequent video data relative to the user based on the updated language unit state data, and updating the subsequent recommendation results.

[0034] Optionally, the multimodal perception data includes at least one of the following: user facial expression data, head posture data, gaze direction data, or gaze duration data acquired by an image acquisition device; User voice data collected by an audio acquisition device, the voice data including one or more of the following: questioning tone, voice indicating a request for help, or sound feedback related to video content.

[0035] Optionally, it also includes: when the potential ununderstood feedback event is identified, determining the corresponding current statement or video segment based on the time correspondence between the occurrence time of the potential ununderstood feedback event and the video playback timeline; Based on the language unit metadata of the current statement or video segment, determine at least one language unit that appears therein as a candidate language unit to be verified or increase its verification priority. The language unit status data is updated only after the candidate language units to be verified are confirmed in the subsequent verification process.

[0036] Optionally, in the forgetting risk calculation step, the forgetting risk value is calculated in the following ways: Obtain at least one of the following as a time reference: the most recent valid verification time, the most recent consolidation time, the most recent valid occurrence time, the most recent valid contact time, or the reference time when the language unit was marked as a recognition state; The time factor is determined based on the time interval between the current time and the time base. The mastery strength factor is determined based on all or part of the historical successful verification count, consecutive successful verification count, and valid occurrence count of the language unit. The forgetting risk value is calculated based on the time factor and the mastery strength factor.

[0037] Optionally, the time factor is determined according to a preset days-weight function; in the days-weight function, the more successful the historical verification of a language unit, the more gradual the rate of weight growth over time; the fewer successful the historical verification of a language unit, the steeper the rate of weight growth over time.

[0038] Optionally, it also includes: a valid occurrence count tracking step, which records the current valid occurrence count of each language unit and the baseline valid occurrence count when the language unit enters the cognitive state; When the difference between the current valid occurrence count and the baseline valid occurrence count reaches a preset threshold, the language unit is added to the consolidation queue. During the consolidation phase, the language units in the queue to be consolidated are subjected to consolidation verification. If the consolidation check passes, maintain or improve the state of the language unit; if the consolidation check fails, reset the baseline valid occurrence count of the language unit to the current valid occurrence count, and recalculate the subsequent consolidation trigger conditions.

[0039] Optionally, it also includes: an abnormal segment identification step, which identifies abnormal viewing segments based on at least one feedback behavior generated by the user during the viewing of video data, such as pausing, rewinding, listening again, asking for help, confused expressions or questioning voices; The reference segment determination step involves identifying segments in which the user did not exhibit the aforementioned feedback behavior and did not display any other potentially uncomprehended signals as reference viewing segments. The comparative analysis step involves identifying the language units appearing in the abnormal viewing segment and the reference viewing segment based on the language unit metadata of the video data, and then comparing the two. When a target language unit appears more frequently in the abnormal viewing segment than it appears more frequently in the reference viewing segment, it is identified as a candidate language unit to be verified or the verification priority of an existing candidate language unit is increased.

[0040] Optionally, it also includes: a consolidation extraction step, maintaining at least two sets of language units that can be consolidated and extracted, including a set of known language units and a set of mastered language units; Language units are extracted from different sets according to a preset ratio for reinforcement and verification; For language units in the known language unit set, the extraction weight is determined based on the number of days elapsed since the most recent verification and the historical verification results.

[0041] Optionally, it also includes: recommending parameter adjustment steps and calculating the success rate of the user's most recent preset language unit verifications; When the success rate is lower than the preset success rate threshold, the language unit load threshold for subsequent video data recommendations is adjusted to reduce the difficulty of subsequent recommended content.

[0042] According to the eighth aspect of this disclosure, a forgetting risk update system with multimodal feedback is also provided, comprising: The language unit status data maintenance module is used to record and update one or more of the following in real time: the recognition status of each language unit, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time. The language unit status data is recorded on a per-language-unit basis. The language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The multimodal perception data acquisition module is used to acquire the user's multimodal perception data through one or more of the image acquisition device and audio acquisition device during the user's viewing of video data. The potential unresolved event identification module is used to extract and identify features from the multimodal perception data to determine whether the user has generated a potential unresolved feedback event. The forgetting risk calculation module is used to calculate the forgetting risk value of a language unit that meets the preset recognition conditions in the language unit status data, based on one or more of the following: the recognition status of the language unit, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time, as well as the potential ununderstood feedback event. The status update module is used to adjust the status data of language units that meet the preset update conditions based on the forgetting risk value. The recommendation impact module is used to recalculate the language unit load of subsequent video data relative to the user based on the updated language unit state data, and update the subsequent recommendation results.

[0043] Optionally, the multimodal perception data includes at least one of the following: user facial expression data, head posture data, gaze direction data, or gaze duration data acquired by an image acquisition device; User voice data collected by an audio acquisition device, the voice data including one or more of the following: questioning tone, voice indicating a request for help, or sound feedback related to video content.

[0044] Optionally, in the forgetting risk calculation module, the forgetting risk value is calculated in the following ways: Obtain at least one of the following as a time reference: the most recent valid verification time, the most recent consolidation time, the most recent valid occurrence time, the most recent valid contact time, or the reference time when the language unit was marked as a recognition state; The time factor is determined based on the time interval between the current time and the time base. The mastery strength factor is determined based on all or part of the historical successful verification count, consecutive successful verification count, and valid occurrence count of the language unit. The forgetting risk value is calculated based on the time factor and the mastery strength factor.

[0045] Optionally, it also includes: a valid occurrence count tracking module, which records the current valid occurrence count of each language unit and the baseline valid occurrence count when the language unit enters the cognitive state; When the difference between the current valid occurrence count and the baseline valid occurrence count reaches a preset threshold, the language unit is added to the consolidation queue. During the consolidation phase, the language units in the queue to be consolidated are subjected to consolidation verification. If the consolidation check passes, maintain or improve the state of the language unit; if the consolidation check fails, reset the baseline valid occurrence count of the language unit to the current valid occurrence count, and recalculate the subsequent consolidation trigger conditions.

[0046] According to a ninth aspect of this disclosure, a computer program product is also provided, comprising a computer program that, when executed by a processor, implements the method described in a seventh aspect of this disclosure.

[0047] This embodiment of the disclosure accurately records the user's language unit status data on a language unit basis. Based on the language unit status data and the language unit metadata of the video data, it calculates the language unit load of each candidate video data relative to the user, filters out video data that meets the preset recommendation conditions, and forms or updates a sequence of videos to be played containing one or more video data. This allows the user to continuously obtain video data of appropriate difficulty, avoids the situation where the system overestimates the user's language ability and the recommended content is too difficult, and improves the matching degree between the difficulty of video data recommendation and the user's actual comprehension ability.

[0048] Other features and advantages of the embodiments of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0049] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the disclosure.

[0050] Figure 1 This is a flowchart illustrating an adaptive streaming method according to some embodiments; Figure 2 This is a block diagram of an adaptive streaming system according to some embodiments; Figure 3 This is a flowchart illustrating a language unit verification method based on interactive behavior according to some embodiments; Figure 4 Here is a block diagram of a language unit verification system based on interactive behavior according to some embodiments; Figure 5 This is a flowchart illustrating a forgetting risk update method with multimodal feedback according to some embodiments; Figure 6 Here is a structural block diagram of a forgetting risk update system with multimodal feedback according to some embodiments; Figure 7 This is a day-weight function curve based on some embodiments. Detailed Implementation

[0051] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0052] The following description of at least one exemplary embodiment is merely illustrative and is not intended to limit the scope of this disclosure or its application or use.

[0053] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0054] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0055] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0056] First Embodiment Figure 1 This is a flowchart illustrating an adaptive streaming method according to some embodiments. The execution entity in this embodiment can be a server, a user-end device with corresponding processing capabilities, or a collaborative combination of a server and a user-end device. For example... Figure 1 As shown, the adaptive streaming method of this embodiment may include the following steps S110 to S150.

[0057] Step S110: Obtain the user's language unit status data.

[0058] This step is used to obtain the user's language unit status data for each language unit to be tested through testing. This data serves as the user's initial language unit status data and is used as the basic input for subsequent language unit load calculation and video recommendation.

[0059] Language units are the core processing objects of this solution, and can include at least one of the following: words, phrases, idioms, characters, words, symbols, syllables, sentence fragments, grammatical structures, or fixed expressions. In the context of children's English listening learning, language units are usually based on words or phrases. Language unit status data refers to user-related data recorded at the language unit level, which can include at least one of the following: whether the user recognizes the language unit, whether they have mastered it, level of recognition, whether it needs reinforcement, number of occurrences, historical time of each occurrence, number of checks, historical time of each check, historical check result, number of successful checks, number of failed checks, number of consecutive correct answers, number of consecutive incorrect answers, and the most recent check result or time. The level of recognition can be a numerical value or percentage between 0 and 1, used to represent the user's mastery of the language unit.

[0060] In some examples, the system acquires the user's initial state data for each language unit through a precise initial test upon first access. For different application scenarios, the system can employ various testing methods such as listening and selecting images, listening and selecting meanings, listening and selecting shapes, listening and selecting native language interpretations, listening and selecting native language interpretation audio, listening and selecting sentences, contextualized listening fill-in-the-blank, true / false questions, matching questions, oral responses, translation, retelling, and shadowing. Taking children's English listening learning as an example, the system can use the listening and selecting image method to test each unit individually: for example, the system plays the pronunciation of the target word while simultaneously displaying multiple candidate images or videos to the user, including one image or video corresponding to the target word and multiple images or videos corresponding to confused words. The user selects the image or video corresponding to the pronunciation. If the user selects correctly, the recognition status of that language unit is updated to "recognized," or the recognition level is adjusted to exceed a preset threshold; if the user selects incorrectly, the language unit is recorded as "unrecognized" or "low recognition level." The system does not attempt to test all language units at once. Instead, it prioritizes acquiring a batch of verified language units. For untested language units, the system adopts a conservative approach during recommendation and load calculation, such as treating them as unfamiliar, to reduce the risk of overestimating user abilities. For function words, linking verbs, prepositions, particles, or grammatical words such as "the," "to," "am," "is," "are," and "be," which are difficult to reliably express in a single image, the system may temporarily exclude them from the initial listening and image selection test. Instead, they will be verified in subsequent viewing and verification stages using methods such as contextualized listening comprehension, sentence selection, or semantic relationship judgment. In these examples, by accurately testing each language unit individually to obtain initial state data, the system avoids overestimating user abilities based on coarse-grained indicators such as age and grade level, providing a reliable foundation of language unit states for subsequent recommendations.

[0061] In other examples, the system can control the termination or skipping of the test process based on at least one of the following: test quantity threshold, test time threshold, cumulative number of errors, consecutive number of errors, user skip behavior, number of user skips, user timeout, or user attributes. For example, the test can be terminated if a user answers 5 questions incorrectly in a row, answers 10 questions incorrectly in a total of, or actively chooses to skip, to prevent the user from becoming continuously frustrated or losing patience.

[0062] Additionally, it's necessary to control the number of tests and the testing time during the testing process to avoid excessive testing or long testing times, which could lead to user boredom or abandonment. The number of tests and testing time are preset values ​​that the system can adjust. For example, we can select the top 200-500 high-frequency language units as the test set. This ensures an appropriate number of units while allowing users with basic language skills to start with video data of a certain difficulty. Although untested language units don't necessarily mean they are unfamiliar, it's not necessary to exhaustively test all language units. It's sufficient to ensure users start with a certain level of difficulty, as the video data itself is interesting enough that users can quickly correct their language unit status data while enjoying watching the videos.

[0063] In other examples, the system can also be set to not require initial testing, or to pause the current round of initial testing and continue to supplement and verify during or after watching subsequent video data, gradually improving the user's language unit status data.

[0064] Step S120: Obtain multiple candidate video data and the language unit metadata of each candidate video data.

[0065] This step involves the server reading candidate video data and its corresponding language unit metadata from the video database. The language unit metadata includes at least the language unit information that appears in the corresponding video data.

[0066] In some examples, language unit metadata may also include at least one of the following: the time of appearance of the language unit, its order of appearance, frequency of appearance, corresponding sentence, corresponding audio segment, corresponding subtitle or dialogue, the video segment to which it belongs, and target semantic information. Target semantic information refers to the plot meaning, event relationships, or key information that the user needs to understand in the overall video data or a segment, such as characters, actions, objects, locations, times, event sequences, causes, results, emotions, or intentions. The system can also record the distribution of language units in consecutive time windows or consecutive language unit windows to determine whether a large number of language units that the user does not recognize appear in a particular segment. In these examples, by pre-establishing language unit metadata containing appearance time, sentence correspondence, and target semantic information, the system can accurately calculate the local language unit load in subsequent steps and quickly locate candidate language units to be verified in the target sentence during the verification process.

[0067] Step S130: Calculate the language unit load of each candidate video data relative to the user based on the language unit status data and language unit metadata.

[0068] This step is performed by the server. Taking the language unit status data from step S110 and the language unit metadata from step S120 as input, the language unit load of each candidate video data relative to the current user is calculated and used as the basis for recommendation determination in step S140. The language unit load represents the language understanding burden caused by the candidate video data to the user.

[0069] In one implementation, the system calculates the overall language unit load, which may include at least one of the following: the number of unrecognized language units in the candidate video data, the proportion of recognized or unrecognized language units in a single candidate video data, the number of language units in the candidate video data with a recognition level below a preset threshold, and the cumulative or weighted average of the recognition or non-recognition levels of each language unit in the candidate video data. Here, non-recognition level can be understood as 1 minus recognition level. The cumulative value can be obtained by summing the products of the frequency of occurrence of each language unit and the non-recognition level. The weighted average is obtained by dividing the cumulative value of the recognition or non-recognition levels of each language unit by the total frequency of occurrence of each language unit in the video data or the total duration of the video data, for example: R = Σ(frequency of occurrence × recognition level of the language unit) / Σ frequency of occurrence.

[0070] In some examples, the system also calculates local language unit load. The system calculates the number of unfamiliar language units, the number of language units with a recognition level below a preset threshold, or the cumulative value of unfamiliarity within each continuous playback time window or continuous language unit window. Local language unit load is used to avoid situations where the overall number of unfamiliar language units in the candidate video data is not high, but multiple unfamiliar language units appear in a short segment, leading to local comprehension difficulties for the user. In these examples, by simultaneously calculating the overall load and local load, the system can dually control the overall difficulty and local difficulty peaks of the candidate video data, more accurately reflecting the user's actual comprehension burden.

[0071] In other examples, the allowable language unit load threshold is determined based on the playback duration of the candidate video data. The shorter the playback duration of the candidate video data, the lower the allowable language unit load threshold; the longer the playback duration of the candidate video data, the higher the allowable language unit load threshold. By dynamically linking playback duration with the load threshold, the system can adopt differentiated difficulty control strategies for video data of different durations, avoiding setting excessively high tolerances for unfamiliar words in short videos.

[0072] In other examples, when multiple video data belong to the same continuous story, the same thematic unit, or the same preset content set, these multiple video data are considered as aggregation candidate content. The system calculates the playback duration, total language unit load, episode language unit load, and local window language unit load of the aggregation candidate content. Based on these metrics, it determines whether to recommend the aggregation candidate content. For example, if an animated series has 10 episodes, each 6 minutes long, the system can consider it as a total of 60 minutes of aggregation content, perform an overall evaluation based on the corresponding duration load threshold, and simultaneously verify whether the episode load and local window load of each episode meet the requirements.

[0073] In another implementation, the adaptive streaming method further includes a language unit set determination step, which determines one or more language unit sets based on language unit status data and preset filtering conditions, and calculates the language unit load of each candidate video data relative to the user based on one or more language unit sets and language unit metadata, or one or more language unit sets, language unit status data, and language unit metadata.

[0074] The determination of the language unit set in this step is performed by the server. The input is the language unit status data and preset filtering conditions obtained in step S110, and the output is one or more language unit sets for use in the calculation of language unit load in step S130.

[0075] A language unit set is a categorized set formed based on language unit status data and preset conditions. It includes at least a recognition set, and may also include a mastery set and / or a reinforcement set. The recognition set consists of language units whose recognition status is "recognized" or whose recognition level exceeds a preset threshold. The mastery set consists of language units that have consistently passed multiple tests, such as those with a preset number of consecutive correct answers. The reinforcement set consists of language units that are recognized but whose recent consecutive test success count has not reached the mastery standard, and is used for continuous reinforcement in subsequent viewing and testing sessions. A language unit can belong to multiple sets simultaneously; for example, a language unit may belong to both the recognition set and the reinforcement set due to insufficient consecutive success counts.

[0076] The "recognition" in the above example is an exemplary description of the state of a language unit. For example, it can indicate that the user has established a correspondence between sound and semantics, sound and appearance, and character shape and semantics that meets at least one preset condition for the language unit. In the context of listening learning, the recognition state can at least indicate that after hearing the language unit, the user is able to understand its semantics and / or recognize its appearance.

[0077] In another implementation, the adaptive streaming method may further include a dynamic adjustment step, dynamically determining the language unit load threshold of the candidate video data based on the user's current language proficiency. Additionally, a playback duration threshold or language density threshold for the candidate video data may also be dynamically determined based on the user's current language proficiency. The user's current language proficiency can be determined based on at least one of the following: the number of recognized language units, the number of language units mastered, historical viewing time, historical language unit verification success rate, or historical comprehension verification success rate. For users with lower language proficiency, the system can set a stricter load threshold, recommending only short videos containing very few unfamiliar language units; for users with higher language proficiency, the system can allow more unfamiliar language units and longer video durations, but still must meet the overall and local language unit load requirements.

[0078] Step S140: Based on the language unit load, determine the video data that meets the preset recommendation conditions from multiple candidate video data.

[0079] This step is executed by the server, taking the language unit load calculated in step S130 as input and outputting video data to be recommended that meets the preset recommendation conditions.

[0080] Preset recommendation criteria may include: the overall language unit load is not higher than a preset overall load threshold; and / or the language unit load of each local window is not higher than a preset local load threshold. The system first filters out candidate video data whose overall load meets the criteria, and then further evaluates whether the load of each local window meets the requirements; if a candidate video data is not difficult overall, but a certain segment contains a large number of unfamiliar language units, the system may lower its recommendation priority or temporarily not recommend it.

[0081] In some examples, if multiple candidate video data meet the preset recommendation criteria, the system can adjust the recommendation order based on at least one of the following: the number of unfamiliar language units, the number of language units with recognition levels below a preset threshold, cumulative unfamiliarity, weighted average recognition or unfamiliarity, and viewing history. Prioritize recommending candidate video data that places the least burden on the user's language comprehension and contains an appropriate amount of new language units. In these examples, by further sorting the candidate video data that meets the criteria, the system can prioritize pushing content most conducive to language unit accumulation to the user while ensuring appropriate difficulty.

[0082] Step S150: Based on video data that meets preset recommendation conditions, form or update a sequence of videos to be played that contains one or more video data.

[0083] In some examples, users can manually select which video data to play in the sequence.

[0084] In other embodiments, the system may automatically play video data in the sequence to be played, for example, when the current video data finishes playing or when a preset playback trigger condition is met, the system may automatically play subsequent video data in the sequence to be played.

[0085] Specifically, the system automatically plays the video data in the playback sequence continuously, achieving automatic and continuous streaming of video data without requiring the user to manually return to the list and reselect each time. When the number of videos in the continuously recommended sequence is insufficient, the system continues to supplement video data that meets preset recommendation criteria until the user stops watching or the preset stopping criteria are met.

[0086] In some examples, before the user manually selects or the system automatically plays subsequent video data, the language unit status data and / or language unit set are updated by verification, and then the language unit load of the subsequent video data relative to the user is recalculated to determine whether the subsequent video data still meets the preset recommendation conditions; if it no longer meets the conditions, the subsequent video data is deleted from the sequence to be played, replaced, or paused.

[0087] This mechanism ensures that each video in the subsequent video data matches the user's current actual language proficiency, avoiding situations where the user's language unit status changes during viewing (e.g., verification results indicate that a language unit has been forgotten), leading to videos that are too difficult. In these examples, by recalculating the load in real time and dynamically verifying the recommendation conditions before playback, the system can adjust the recommended video data based on the latest language unit status data, forming a dynamic and adaptive video data recommendation mechanism.

[0088] The validation in the above example can be performed in the following way: During the process of a user watching video data in the sequence to be played, or at a preset time after watching, language units that meet preset conditions are verified. Based on the verification results, update at least one of the user's language unit status data and / or one or more language unit sets; Based on the updated language unit status data and / or the updated set of one or more language units, the system recalculates or updates the language unit load of subsequent video data relative to the user, and updates the subsequent recommendation results and / or the sequence to be played according to the recalculated or updated language unit load.

[0089] In the above verification method, the system verifies at least some language units appearing in the video data. Before watching video data that meets the preset recommendation conditions, the system verifies that the language unit status data is language units that are not recognized or whose recognition level is lower than a preset threshold (such as new words). Based on the verification results, the system dynamically updates the language unit status data or set affiliation, recalculates or updates the language unit load of subsequent video data relative to the user, and updates the list or sorting of subsequent video data in the sequence to be played.

[0090] In some examples, the system can also verify that language units are recognized or have a recognition level higher than a preset threshold before watching video data that meets preset recommendation conditions. This allows the system to identify language units that were previously considered recognized but may no longer be recognized. Based on the verification results, the system dynamically updates the language unit status data or set affiliation, recalculates the language unit load of subsequent video data relative to the user, and updates the list or sorting of subsequent video data. This helps to reduce the difficulty of subsequent video data when the user forgets it, preventing the difficulty of subsequent video data from becoming too high.

[0091] In other examples, the system can further validate language units that do not appear in the video data according to preset conditions. For example, by introducing language units to be reinforced, the system can subtly and continuously track changes in the state data of more language units that the user has encountered in the past, both during and after watching the video data. This can help to promptly identify forgotten language units and avoid the accumulation of too many language units to be reinforced in the independent reinforcement stages of the system, which could put too much pressure on the user.

[0092] Meanwhile, to avoid user frustration due to an excessive number of language unit verifications or a high failure rate, the system can set a verification budget, which includes: quantity, duration, number of consecutive correct verifications, number of consecutive incorrect verifications, and user skipping behavior.

[0093] When the verification budget is limited, the system can further set verification priorities. For example, the system can set a fixed number of verifications and prioritize verifying language units that appear in the video data and whose language unit status data is "unrecognized" or "recognition level is below a preset threshold" before watching the current video data. For language units that appear in the video data and whose language unit status data is "recognized" or "recognition level is above a preset threshold" before watching the video data, as well as language units that have not appeared in the video data, the verification priority can be sorted based on conditions such as whether they appear in the current episode, recognition level, reinforcement weight, most recent verification time, most recent verification result, historical verification time, historical verification result, most recent consecutive successful verifications, most recent consecutive failed verifications, number of valid appearances, most recent contact time, set affiliation, pause, replay, re-listen, request for explanation, and text annotation click behavior during viewing. Entries exceeding the budget will not be verified in the current session.

[0094] In existing language learning scenarios, if the system relies solely on coarse-grained indicators such as age, grade, or course level to recommend content when users watch video data, it is difficult to accurately reflect the user's true mastery of the language units in the specific video data, which can easily lead to recommended content that is too difficult. As can be seen from steps S110 to S150 above in the embodiments of this disclosure, this embodiment accurately acquires and maintains the user's language unit status data on a per-language-unit basis, calculates the overall language unit load and local language unit load of the candidate video data based on the language unit metadata, filters out video data that meets the preset recommendation conditions and adds it to the playback sequence, updates the user's language unit status data through the verification method in the aforementioned example, and recalculates the load in real time before playback to dynamically verify the recommendation conditions, so that users can continuously obtain video data of appropriate difficulty and avoid poor learning results due to recommended content exceeding the user's actual comprehension ability.

[0095] Figure 2 This is a block diagram of an adaptive streaming system according to some embodiments. For example... Figure 2 As shown, the adaptive streaming system includes: a language unit status data acquisition module, used to acquire the user's language unit status data, which is recorded in units of language units. A language unit includes at least one of words, phrases, idioms, characters, words, syllables, sentence fragments, grammatical structures, or fixed expressions; a video metadata acquisition module, used to acquire multiple candidate video data and the language unit metadata of each candidate video data, where the language unit metadata includes at least the language unit information appearing in the corresponding video data; a language unit load calculation module, used to calculate the language unit load of each candidate video data relative to the user based on the language unit status data and / or one or more sets of language units and the language unit metadata; a recommendation determination module, used to determine the video data to be recommended from multiple candidate video data that meets preset recommendation conditions based on the language unit load; and a push output module, used to form or update a sequence of videos to be played containing one or more video data based on the video data that meets the preset recommendation conditions. The functions of each module correspond to the functions of each step in the aforementioned example, and the data flow relationship between modules is explained in steps S110 to S150.

[0096] In addition, the adaptive streaming system also includes a language unit set determination module and a level dynamic adjustment module. The language unit set determination module is used to determine one or more language unit sets based on language unit status data and preset filtering conditions. It then calculates the language unit load of each candidate video data relative to the user based on one or more language unit sets and language unit metadata, or one or more language unit sets, language unit status data, and language unit metadata. The level dynamic adjustment module is used to dynamically determine the language unit load threshold, playback duration threshold, or language density threshold of the candidate video data based on the user's current language level. The user's current language level is determined based on at least one of the following: the number of language units recognized, the number of language units mastered, historical viewing time, historical language unit verification success rate, or historical comprehension verification success rate.

[0097] The workflow relationships of all modules in the adaptive streaming system in the above embodiments are as follows: Figure 2 As shown.

[0098] In language learning scenarios, a user's language ability is not static. On the one hand, after watching new video data, a user may master new language units that appear in it. On the other hand, language units that the user previously recorded as known or mastered by the system may be forgotten over time, due to changes in context, or lack of repetition, or may only be in a state of unstable mastery. If the system only records a single test result or a single learning result, without continuously tracking the mastery of new language units and the forgetting or unstable mastery of old language units, it is easy to overestimate the user's ability in the long run.

[0099] Second Embodiment Figure 3 This is a flowchart illustrating a language unit verification method based on interactive behavior according to some embodiments. The execution entity in this embodiment can be a collaborative combination of a server and a user-end device. The user-end device is responsible for collecting user interaction operations and reporting them to the server, while the server is responsible for identifying help requests, determining candidate language units to be verified, generating verification tasks, and processing the verification results. Figure 3 As shown, the language unit verification method based on interactive behavior in this embodiment may include steps S310 to S380.

[0100] Step S310: Obtain the user's language unit status data.

[0101] This step is executed in the same way as step S110 in the first embodiment. The system obtains the language unit status data of each language unit to be tested by the user through testing, which serves as a reference for the subsequent candidate language unit determination and verification tasks. For the meaning and acquisition method of the language unit status data, please refer to the description of step S110.

[0102] Step S320: Play video data that meets the preset recommendation criteria to the user.

[0103] This step is performed by the user-end device, which plays video data that meets the preset recommendation conditions as determined by the recommendation mechanism of the first embodiment to the user. During the playback of the video data, the user-end device synchronously records the playback timeline information, including the current playback time, the start and end times of each sentence, and the subtitle timestamp, for use in the time correspondence calculation in the subsequent step S340.

[0104] Step S330: During the playback of video data, monitor the preset interactive operations performed by the user on the video playback interface.

[0105] This step is continuously monitored by the user's terminal device, and all kinds of interactive operations performed by the user on the video playback interface are reported to the server in real time.

[0106] Preset interactive operations can include at least one of the following types.

[0107] The first type is triggering the help-seeking interactive controls set in the video playback interface. For example, clicking interactive controls such as "Didn't understand", "Help me", "Request explanation", or "Request to listen again" will be regarded by the system as the user's help-seeking behavior.

[0108] The second category is triggering sentence-level playback control operations, including one or more of the following: previous sentence operation, next sentence operation, replay current sentence operation, jump to the start of current sentence operation, or jump to the start of previous sentence operation. The system pre-generates the start and end times of sentences and sentence start markers based on video audio, subtitles, dialogue text, manually annotated information, or automatic speech recognition results, and forms snap-in sentence-level anchors in the playback progress bar. When the user triggers the above sentence-level operations, the system records the operation as a help request behavior related to the corresponding sentence or segment.

[0109] The third type is triggering the playback progress bar drag operation. When the user drags the playback progress back to a certain position, the system determines the corresponding target statement or semantic segment based on the playback progress position after the operation, and snaps the playback progress to the beginning of the statement or semantic segment.

[0110] In other examples, under the default listening training scenario for video data, the system does not display subtitles for the target language or translations in the user's native language, allowing the user to focus on improving their listening and speaking skills.

[0111] In addition, the system can conditionally control the appearance of subtitles and user's native language translation based on the current viewing stage (e.g., first viewing, rewatch, verification, reinforcement, or other stages), number of views, user attributes, language unit status data, language unit verification results, comprehension verification results, and other factors. For example, the system defaults to not displaying subtitles during the first viewing, while allowing subtitles to appear during subsequent viewings if all language units in the video data content are in a comprehension state or during specific verification or reinforcement stages.

[0112] When subtitles or text annotations are displayed conditionally, if a user clicks on a subtitle or a language unit within a given sentence, the system determines the target sentence based on the sentence identifier or timestamp of the clicked subtitle. This click is treated as a signal of potential forgetting or unstable mastery of the relevant language unit, rather than simply a word lookup. In other words, the preset interactive operations, under certain circumstances, also include a fourth type of clicking on subtitles or language units within subtitles. In this case, the system can treat the relevant language unit as a candidate for verification in the current video data, increase its verification or reinforcement priority, or adjust its recognition level or set affiliation after meeting preset conditions.

[0113] Step S340: When a preset interactive operation is detected, the target statement or target video segment corresponding to the preset interactive operation is determined based on the time correspondence between the occurrence time of the preset interactive operation and the video playback timeline.

[0114] This step is executed by the server. Taking the type of interactive operation and its occurrence time reported in step S330 as input, and combining it with the video data playback timeline, sentence start and end times, subtitle timestamps, speech recognition timeline, or sentence identifier, the system determines the current sentence, previous sentence, most recently played sentence, or corresponding video segment corresponding to the interactive operation. This serves as the location basis for determining candidate language units in step S360. The system can also determine the corresponding sentence or segment within a preset time range before and / or after the interactive operation to account for the time delay between the user's reaction and the playback of the dialogue.

[0115] Step S350: Obtain the language unit metadata of the video data.

[0116] This step involves the server reading the language unit metadata of the video data being played in step S320 from the video database. This metadata includes at least the language unit information appearing in the video data, the statement identifier corresponding to each language unit, and the occurrence time, for use in step S360.

[0117] Step S360: Based on the language unit metadata of the video data, determine at least one language unit appearing in the target statement or target video segment as a candidate language unit to be verified.

[0118] This step is executed by the server. Taking the target statement or video clip determined in step S340 and the language unit metadata obtained in step S350 as input, it extracts the language units appearing in the target statement or clip as candidate language units to be verified, and outputs them to step S370 for verification task generation. The system does not directly determine candidate language units as those the user does not recognize or whose recognition is reduced; instead, it treats them as potential candidates for forgetting or unstable mastery, further confirming them through subsequent verification tasks.

[0119] In some examples, the language unit verification method may also include an interactive confirmation step. After identifying the target sentence or video segment, the system presents one or more language units contained in the sentence or segment to the user in an interactive format, and provides at least one type of auxiliary content—audio of the sentence, audio of the language unit, translation of the sentence, or translation of the language unit—for the user to select. The system obtains the auxiliary content selected by the user and identifies the language unit corresponding to the user-selected auxiliary content as a candidate language unit to be verified or increases its verification priority. Through this interactive confirmation step, the system can further narrow down the range of candidate language units to be verified based on the user's active request for explanation, thereby improving the targeting of the verification.

[0120] In addition, during the above-mentioned candidate language unit determination step, the system can also verify unfamiliar words or words that the system considers to need reinforcement before the user watches the video data. For example, language units whose status data is unknown or whose recognition level is lower than a preset threshold before watching the video data, or language units whose status data is known or whose recognition level is higher than a preset threshold before watching the video data. The candidate language units to be verified are not all language units that appear in the target sentences or target video segments corresponding to the preset interactive operations.

[0121] Step S370: Generate a language unit verification task for the candidate language units to be verified and output it to the user.

[0122] This step is executed by the server. It takes the candidate language units to be verified as determined in step S360 as input, determines the verification task format according to the language unit type, generates the corresponding verification task, and outputs it to the user through the user terminal device.

[0123] In some examples, the system first identifies the language unit type of the candidate language units to be verified, and then determines the verification task format based on the type. The verification task formats include recognition-based verification tasks based on audiovisual matching or sound-meaning matching, fill-in-the-blank and selection-based verification tasks based on contextual understanding, and judgment and output-based verification tasks based on semantic processing. Specifically, these include at least one of the following: listening to and selecting images, listening to and selecting meanings, listening to and selecting shapes, listening to and selecting native language interpretations, listening to and selecting native language interpretations (audio), listening to and selecting sentences, contextualized listening fill-in-the-blank, true / false questions, matching questions, oral responses, translation, retelling, and shadowing.

[0124] When the candidate language unit to be verified is a concrete noun or action word, a recognition-based verification task (e.g., a sound-based image selection task) is generated, using images or audiovisual materials as the matching objects. Specifically, if the candidate language unit to be verified is a concrete noun or action word, such as "dog," "jump," or "table," which can be stably expressed by images, the system generates a sound-based image selection verification task: the system plays the pronunciation of the target language unit and simultaneously displays multiple candidate images or videos to the user, including one image or video corresponding to the target word and multiple images or videos corresponding to the confused words. The user selects the image or video that corresponds to the pronunciation.

[0125] When the candidate language unit to be verified is a function word, copula, preposition, auxiliary word, grammatical word or abstract word, a fill-in-the-blank and selection verification task based on contextual understanding or a judgment and output verification task based on semantic processing is generated. This verification task requires fill-in-the-blank, selection or judgment verification tasks that rely on contextual information for listening or semantic processing (such as one or more of the following: contextualized listening fill-in-the-blank task, listening and selecting sentence task or semantic relationship judgment task).

[0126] Specifically, semantic processing-based judgment and output verification tasks include semantic relationship judgment, oral paraphrasing tasks, or translation tasks.

[0127] Specifically, if the candidate language unit to be verified is a function word, copula, preposition, auxiliary word, grammatical word, or abstract word, such as are, am, is, be, the, to, etc., which are difficult to be stably represented by images, the system generates one or more of the following tasks: contextualized listening comprehension task, listening and selecting sentences task, or semantic relationship judgment task. In these examples, the verification task format is adaptively selected according to the type of language unit, avoiding the misjudgment problem caused by using the listening and selecting image method for language units such as function words that are difficult to be stably represented by images.

[0128] Among them, the candidate options, interpretation objects or feedback objects involved in the verification task can be presented using multimedia or multimodal information, including but not limited to images, text, audio, video clips, target language interpretation, user's native language interpretation, user's native language interpretation audio or a combination thereof.

[0129] In other examples, for contextualized listening comprehension tasks, the question stem context is generated based on constraints from the user's language unit status data, specifically the set of language units they already know or are proficient in. This ensures that the language units in the question stem context, excluding the candidate language units to be verified and the candidate options, meet a preset low-load condition. For example, when testing "are," the system can play the audio question stem "We __ happy," along with candidate audios "am," "is," "are," and "be." The "we" and "happy" in the question stem are determined by constraints from the user's already mastered language unit set, ensuring the user can understand the question stem itself. Candidate options may include at least two distracting language units belonging to the same grammatical function category or the same confusion set as the language unit to be verified. By constraining the language units in the question stem context to all come from the user's already mastered set, it ensures that the user will not be misjudged as not recognizing the target language unit due to not understanding the question stem itself.

[0130] In some examples, the system generates at least two sets of verification tasks in different contexts for candidate language units to be verified. The first set of contexts is the original context of the language unit in the target sentence or target video clip, and the second set of contexts is a transfer context different from the original context. The system verifies the language unit in both the original and transfer contexts and updates the state data of the language unit based on the user's answers in both contexts. For example, after a user hears "are" in a video containing "Where are my shoes?", the system can design a task around the original sentence or test "are" in a new sentence composed of language units that the user has already mastered. If the user can correctly select in both the original and transfer contexts, the system adjusts "are" to indicate recognition or increases its recognition level. If the user only answers correctly in the original context but remains confused in the transfer context, the system can consider that the user is still in a state of not recognizing or having low recognition of the language unit. By introducing transfer context verification, it is possible to distinguish whether the user has truly mastered the language unit or is only answering based on short-term memory of the original sentence, thereby improving the accuracy of state updates.

[0131] In another implementation, the language unit verification method may further include a translation or paraphrase positioning step, through which the system determines candidate language units to be verified. At a preset time when the user is watching video data, the system acquires the translation or paraphrase results submitted by the user in voice or text format for sentences, segments, or the entire content; compares the translation or paraphrase results with the corresponding standard meaning to determine the content that the user has not expressed correctly; and, based on the correspondence between the incorrectly expressed content and the language units in the video data, determines it as a candidate language unit to be verified or increases the verification priority of existing candidate language units. The translation or paraphrase results are primarily used to determine candidate language units to be verified, but are not directly used as the sole basis for lowering the status of related language units.

[0132] Step S380: Obtain user feedback information regarding the language unit verification task, and update the user's language unit status data based on the feedback information.

[0133] This step is executed by the server. It takes the verification answer submitted by the user through their client device as input, updates the status data of the corresponding language unit based on the correctness of the answer, and writes the update result to the language unit status database. If the user answers correctly, the recognition status of the language unit is updated to "recognized" or the recognition level is positively adjusted, and the unit can be added to the recognition set. If the user answers incorrectly, the recognition status is updated to "unrecognized" or the recognition level is negatively adjusted, and the unit can be removed from the recognition set. The system then recalculates the language unit load of subsequent candidate video data based on the updated language unit status data and updates the subsequent recommendation results.

[0134] In some examples, the language unit verification method may also include a comprehension verification step. When performing this step, the system generates a language-dependent comprehension verification task corresponding to the video data and outputs it to the user. This task requires answering based on the dialogue, audio, target language unit, or sentence semantics in the video data, rather than simply inferring the correct answer from the visuals or plot. For example, if the question is "Why did the little frog jump into the pond?", the correct answer depends on the dialogue "because it was scared," and the user must understand the causal semantics to answer correctly. However, if the question is "Where is the puppy?" and the video clearly shows the puppy under the table, the user can answer correctly without understanding the dialogue; this question does not belong to the language-dependent comprehension verification task. The system obtains user feedback on the language-dependent comprehension verification task to determine the comprehension result. When a language unit verification result passes but the related language-dependent comprehension verification result fails, the system reduces the weight of the current verification result for that language unit, not directly using the passing result as the basis for status improvement. This avoids the system misjudging the user's mastery of the language unit based solely on visual information or rote memorization. In these examples, by combining the language unit verification results with the language dependency comprehension verification results, the system is able to more accurately determine whether the user truly understands the meaning of the target language unit in the context.

[0135] As can be seen from steps S310 to S380 above, this embodiment accurately maps the user's interactive operations during video playback to corresponding sentences or segments, locates candidate language units to be verified by combining language unit metadata, and adaptively generates recognition-type verification tasks based on audiovisual matching or sound-semantic matching, fill-in-the-blank and selection-type verification tasks based on contextual understanding, and judgment and output-type verification tasks based on semantic processing according to the language unit type. It also comprehensively updates the language unit state data by combining the language dependency comprehension verification results, thereby realizing dynamic calibration of language unit state based on real viewing behavior and reducing the risk that the system will misjudge that the user has mastered the knowledge because they only answered correctly based on the picture or mechanical memory.

[0136] Figure 4 This is a block diagram of a language unit verification system based on interactive behavior, according to some embodiments. For example... Figure 4 As shown, the language unit verification system includes: a language unit status data acquisition module for acquiring the user's language unit status data; a video playback module for playing video data that meets preset recommendation conditions to the user; an interactive behavior monitoring module for monitoring preset interactive operations performed by the user on the video playback interface during video playback; a help-seeking event recognition module for determining the target statement or target video segment corresponding to the preset interactive operation based on the time correspondence between the occurrence time of the preset interactive operation and the video playback timeline when a preset interactive operation is detected; a video metadata acquisition module for acquiring language unit metadata of the video data, which includes at least the language unit information appearing in the corresponding video data; and a candidate language unit determination module. The system is used to determine at least one language unit appearing in the target sentence or target video segment as a candidate language unit to be verified based on the language unit metadata of the video data; the verification task generation module is used to generate language unit verification tasks for the candidate language units to be verified and output them to the user. The form of the verification task is determined according to the type of language unit. For concrete nouns or action words, it generates recognition-type verification tasks based on audiovisual matching or sound-meaning matching. For function words, copulas, prepositions, auxiliary words, grammatical words, or abstract words, it generates one or more of the following: fill-in-the-blank and selection-type verification tasks based on context understanding or judgment and output-type verification tasks based on semantic processing; the verification result processing module is used to obtain the user's feedback information on the language unit verification tasks and update the user's language unit status data according to the feedback information. The functions of the above modules correspond to the functions of the steps in the second embodiment, see the description of steps S310 to S380.

[0137] In addition, the language unit verification system may also include a translation or paraphrase positioning module, an interactive confirmation module, and a comprehension verification module. The translation or paraphrase positioning module is used to acquire the translation or paraphrase results submitted by the user in voice or text form at a preset time when the user is watching video data; compare the translation or paraphrase results with the corresponding standard meaning to determine the content that the user has not expressed correctly; and determine it as a candidate language unit to be verified or increase the verification priority of existing candidate language units based on the correspondence between the incorrectly expressed content and the language units in the video data. The interactive confirmation module is used to present one or more language units contained in the target sentence or target video segment in an interactive form after determining the target sentence or target video segment, and provide at least one of the following auxiliary content for the user to choose from: audio of the corresponding sentence, audio of the language unit, translation of the sentence, or translation of the language unit; acquire the auxiliary content selected by the user, and determine the language unit corresponding to the user-selected auxiliary content as a candidate language unit to be verified or increase its verification priority. The comprehension verification module generates and outputs language dependency comprehension verification tasks corresponding to video data to the user. These tasks require answers based on dialogue, audio, target language units, or sentence semantics within the video data. The module obtains user feedback on the language dependency comprehension verification tasks and determines the comprehension result. When a language unit passes the verification but the related language dependency comprehension verification fails, the weight of that language unit's current verification result is reduced.

[0138] The workflow of all modules in the language unit verification system in the above embodiments is as follows: Figure 4 As shown.

[0139] In existing language learning products, vocabulary tests, post-viewing Q&A, and reinforcement mechanisms are often disconnected from the difficulty control of video recommendations. Even if the system records that a user has answered or learned certain language units correctly, it struggles to determine whether the unit has been stably mastered, forgotten, recognized only in specific contexts, or answered correctly solely based on visual information, rote memorization, or answer clues. Consequently, the system may consistently treat unstable language units as already recognized or mastered, failing to identify forgotten or unstable units. This leads to an underestimation of the number of truly unknown language units in subsequent video data, resulting in increasingly difficult recommended content.

[0140] Third Embodiment Figure 5This is a flowchart illustrating a multimodal feedback-based forgetting risk update method according to some embodiments. The execution entity in this embodiment can be a collaborative combination of a server and a user-end device. The user-end device is responsible for collecting multimodal perception data and reporting it to the server, while the server is responsible for identifying potential ununderstood feedback events, calculating forgetting risk values, updating language unit state data, and recalculating subsequent recommendation results. Figure 5 As shown, the forgetting risk update method with multimodal feedback in this embodiment may include the following steps S510 to S560.

[0141] Step S510: Record and update in real time one or more of the following in the status data of each language unit: recognition status, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time.

[0142] Specifically, the status data of each language unit is recorded and updated in real time, including at least one of the following: whether the language unit is recognized, whether it is mastered, level of recognition, whether it needs to be reinforced, number of occurrences, historical time of each occurrence, number of verifications, historical time of each verification, historical result of each verification, number of successful verifications, number of failed verifications, number of consecutive correct answers, number of consecutive incorrect answers, most recent verification result, or most recent verification time.

[0143] This step is continuously executed by the server, maintaining the state data of each language unit in the language unit state database in real time. The recording unit for the language unit state data is the same as in the first embodiment, based on the language unit. A language unit can include at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. Whenever a user completes a verification, watches a video clip containing a language unit, or triggers other preset update conditions, the server updates fields such as the recognition status, historical verification results, number of consecutive successful verifications, number of valid occurrences, and most recent contact time for that language unit, providing a real-time data basis for calculating the forgetting risk value in step S540.

[0144] Step S520: During the process of the user watching video data, the user's multimodal perception data is collected through one or more of the image acquisition device and audio acquisition device.

[0145] This step is performed by the user-end device under the condition of user authorization and meeting the preset collection conditions. It collects multimodal perception data generated by the user during the viewing of video data through image acquisition device and / or audio acquisition device, and reports the collected data to the server for processing.

[0146] In some examples, multimodal perception data may include at least one of the following: user facial expression data, head posture data, gaze direction data, or gaze-off-screen duration data acquired through an image acquisition device; and user voice data acquired through an audio acquisition device, which includes one or more of the following: interrogative tones, voices conveying a request for help, or sound feedback related to the video content, such as the user saying "What does that mean?", "I don't understand", or "Say it again," or making emotional sounds such as confusion or surprise. In these examples, by fusing image and audio perception data, the system can capture potential uncomprehended signals generated by the user during viewing from multiple dimensions.

[0147] Step S530: Perform feature extraction and identification on the multimodal perception data to determine whether the user has generated a potential unresolved feedback event.

[0148] This step is executed by the server. It takes the multimodal perception data collected and reported in step S520 as input, processes the data through at least one of image recognition, posture recognition, gaze recognition, speech recognition, audio feature recognition or multimodal information analysis, identifies whether the user has generated viewing feedback events such as frowning, shaking head, looking away from the screen, making a questioning tone or saying a voice with the meaning of asking for help, and outputs the viewing feedback events that are judged to be potential ununderstood feedback events to step S540.

[0149] It should be noted that potential uncomprehended feedback events serve only as weak feedback signals indicating that the user may not have understood, may have forgotten, or has unstable mastery of the language unit. They are not used directly as a basis for lowering the recognition level of a particular language unit. The system does not directly reduce the recognition level of a related language unit based on a single frown or questioning tone. Instead, it treats them as weak signals that trigger the determination of candidate language units to be verified, and further confirms them through subsequent verification tasks.

[0150] Step S540: For language units in the language unit status data that meet the preset recognition conditions, calculate the forgetting risk value of the language unit based on one or more of the following: recognition status, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time, as well as potential unresolved feedback events.

[0151] This step is executed by the server. It takes the language unit status data maintained in step S510 and the potential ununderstood feedback events identified in step S530 as inputs, calculates the forgetting risk value for language units that meet the preset recognition conditions (e.g., the current recognition status is recognition or the recognition level exceeds the preset threshold), and outputs it to step S550 for status update.

[0152] The calculation method for the forgetting risk value may include: first, obtaining at least one of the following: the most recent valid verification time, the most recent consolidation time, the most recent valid occurrence time, the most recent valid contact time, or the baseline time when the language unit was marked as a recognition state, as a time baseline; determining a time factor based on the time interval between the current time and the time baseline; determining a mastery strength factor based on all or part of the historical successful verification count, consecutive successful verification count, and valid occurrence count of the language unit; and calculating the forgetting risk value based on the time factor and the mastery strength factor.

[0153] Among them, the number of successful historical verifications of language units is the number of correct results in the historical verification results.

[0154] In some examples, the time factor is determined based on a preset days-weighting function. For example... Figure 7 As shown, the function plots curves with the number of days since the last verification on the horizontal axis and the corresponding forgetting risk weight on the vertical axis for different historical verification success counts. Figure 7 The diagram illustrates three representative curves: Language units with fewer correct answers (e.g., only 1 correct answer) correspond to a steep curve, meaning their weight increases rapidly within a short period, leading to a rapid increase in the risk of forgetting. These language units are easily selected for reinforcement testing within a short time. Language units with a moderate number of correct answers (e.g., 3 correct answers) correspond to a medium-slope curve, with their weight increasing at a moderate rate over time. Language units with more correct answers (e.g., 5 or more correct answers) correspond to a flat curve, meaning their weight increases significantly only after a longer period, with a slow increase in the risk of forgetting. These language units are considered stable over a longer period and are less likely to be prioritized for selection. The core characteristic of this function is that for the same language unit, the more successful the historical tests, the more gradual the rate of weight increase over time; conversely, the fewer successful historical tests, the steeper the rate of weight increase over time; and the less stable the historical tests, the significantly higher the risk of forgetting within a short period. By introducing a non-linear days-weighting function that is related to the degree of mastery, the calculation of forgetting risk is made more in line with the actual laws of human memory. The language units with more unstable mastery are assigned a higher weight for forgetting risk, so that the review and verification of these language units are given priority in the consolidation extraction stage.

[0155] When step S530 identifies a potential unresolved feedback event, the system treats the event as an additional forgetting risk signal and takes it into account when calculating the forgetting risk value of the relevant language unit, but the event itself does not directly trigger a state downgrade.

[0156] Step S550: Adjust the language unit status data of language units that meet the preset update conditions according to the forgetting risk value.

[0157] This step is performed by the server. Taking the forgetting risk value calculated in step S540 as input, the state of language units whose forgetting risk value exceeds the preset update condition is adjusted, such as reducing the recognition level, moving them from the recognition set to the set to be consolidated, or adjusting their set affiliation. The adjustment results are written into the language unit state database.

[0158] In some examples, when a potentially ununderstood feedback event is identified, the system determines the corresponding current statement or video segment based on the time correspondence between the event's occurrence and the video playback timeline. Based on the language unit metadata of that statement or segment, it identifies at least one language unit appearing therein as a candidate language unit to be verified or increases its verification priority. Only after confirming the candidate language units in subsequent verification stages is the language unit status data updated based on the confirmation result. By treating potentially ununderstood feedback events generated by multimodal perception data only as weak signals to trigger the determination of candidate language units to be verified, rather than directly lowering the language unit status, misjudgments of status due to occasional user behavior are effectively avoided.

[0159] In other examples, this forgetting risk update method may also include a valid occurrence count tracking step. When performing this step, the system records the current valid occurrence count N of each language unit and the baseline valid occurrence count M when the language unit enters the recognition state. When the difference between N and M reaches a preset threshold, the system adds the language unit to the consolidation queue. After the user enters the consolidation phase, the system performs consolidation verification on the language units in the consolidation queue. If the consolidation verification passes, the state of the language unit is maintained or improved, for example, adjusted to a proficient state. If the consolidation verification fails, the baseline valid occurrence count M of the language unit is reset to the current valid occurrence count N, i.e., the subsequent consolidation triggering conditions are recalculated from the new viewing baseline, and the number of successful consolidation counts is reset to zero. By recording the valid occurrence count and setting a difference threshold, the system can naturally trigger consolidation verification during the user's continuous video viewing, avoiding excessive backlog of consolidation tasks that could stress the user.

[0160] In some other examples, the forgetting risk update method may further include an abnormal segment identification step, a reference segment determination step, and a comparative analysis step. When performing the abnormal segment identification step, the system identifies abnormal viewing segments based on at least one feedback behavior generated by the user during video viewing, such as pausing, rewinding, replaying, asking for help, confused expressions, or questioning voices. When performing the reference segment determination step, the system identifies segments where the user does not exhibit the aforementioned feedback behavior and no other potentially uncomprehended signals as reference viewing segments. When performing the comparative analysis step, the system identifies language units appearing in both the abnormal viewing segments and the reference viewing segments based on the language unit metadata of the video data, and compares the two: when the target language unit appears more frequently in the abnormal viewing segment than in the reference viewing segment, it is identified as a candidate language unit to be verified or the verification priority of existing candidate language units is increased. When a language unit appears in an abnormal viewing segment but is viewed normally multiple times by the user or passes relevant verifications in the reference viewing segment, the verification priority of that language unit is reduced. For example, if a user rewinds and asks for help while watching "The dog is under the table.", the system identifies this sentence as an abnormal viewing segment and initially obtains candidate language units such as dog, is, under, the, and table. In the user's viewing history, "dog" and "table" have appeared multiple times in other sentences, and the user has completed the viewing normally each time. However, "under" has appeared multiple times in the segment where the rewind or help request occurred. Based on this, the system lowers the verification priority of "dog" and "table" and increases the verification priority of "under". By comparing the distribution of language units in the abnormal viewing segment with that in the reference viewing segment, the system can identify the language units that truly cause comprehension difficulties from the user's overall viewing behavior pattern, rather than relying solely on a single feedback behavior.

[0161] In another implementation, the forgetting risk update method identifies language units to be reinforced based on their long-term absence. When a language unit has entered a state of recognition, mastery, or meets preset recognition conditions, the system records at least one of the following: the most recent effective appearance time, the most recent effective verification time, the most recent reinforcement time, or the most recent effective contact time. If the language unit does not appear effectively again within a preset time interval, preset number of episodes watched, preset viewing duration, or preset date interval, or is not effectively verified again, the system identifies the language unit as a language unit to be reinforced, or increases its priority. This allows for the discovery of language units previously considered recognized but potentially forgotten or poorly mastered due to long-term absence.

[0162] In another implementation, the forgetting risk update method may further include a consolidation extraction step. During this step, the system maintains at least two sets of language units that can be consolidated and extracted: a set of recognized language units and a set of mastered language units. When a user enters the consolidation phase, the system extracts language units from different sets according to a preset ratio for consolidation verification. For example, it extracts 8 recognized language units and 2 mastered language units from 10 consolidation words. For language units in the recognized language unit set, the system determines the extraction weight based on the number of days since the most recent verification and historical verification results: the more correct answers in the past, the flatter the curve of weight change over time; the fewer correct answers in the past, the faster the weight increases in a shorter period, making it easier to be extracted. Even if a language unit has already entered the mastered set, the system can still perform a random check on that language unit based on conditions such as low-probability sampling, effective exposure interval, or the most recent verification time. If the sampling result fails, the system can adjust the language unit to be unrecognized or reduce its recognition level and remove it from the mastered set.

[0163] Specifically, consolidation extraction refers to the process of selecting language units from the set of known or mastered language units according to preset rules for consolidation and verification.

[0164] Step S560: Recalculate the language unit load of subsequent video data relative to the user based on the updated language unit state data, and update the subsequent recommendation results.

[0165] This step is performed by the server. Taking the language unit status data adjusted in step S550 as input, the server recalculates the overall language unit load and local language unit load of the subsequent candidate video data in accordance with the method of step S130 in the first embodiment, and re-screens and sorts the subsequent recommendation results according to the recommendation conditions in step S140, and sends the updated recommendation results to the user terminal device.

[0166] In some examples, the system also performs a recommendation parameter adjustment step: the system calculates the success rate of the user's most recent preset number of language unit verifications; when this success rate is lower than a preset success rate threshold, the system determines that the current recommendation difficulty may be too high and adjusts the language unit load threshold for subsequent video data recommendations to reduce the difficulty of subsequent recommended content. For example, when a higher load value indicates higher difficulty, the system lowers the maximum allowed language unit load threshold; when a higher proportion of recognized language units indicates lower difficulty, the system raises the minimum allowed proportion of recognized language units threshold for recommendations. By dynamically linking the language unit verification success rate with recommendation parameters, the system can promptly reduce the recommendation difficulty when the user's forgetting becomes more severe, avoiding the continuous pushing of video data that exceeds the user's current comprehension ability.

[0167] As can be seen from steps S510 to S560 above, this embodiment identifies potential unresolved feedback events generated by the user during viewing by fusing multimodal perception data collected by the image acquisition device and the audio acquisition device. It combines the historical verification records of language units, the number of consecutive successful verifications, the number of effective occurrences, and the most recent contact time, and dynamically calculates the forgetting risk value through time factors and mastery strength factors. It also adjusts the state of language units that meet the preset update conditions. At the same time, it continuously tracks the forgetting and unstable mastery of the recognized language units through effective occurrence tracking, abnormal segment comparison analysis, and consolidation extraction mechanisms. Based on the updated language unit state data, it recalculates the subsequent recommendation results, realizing continuous tracking and active calibration of the user's language unit forgetting state, and avoiding the system's long-term overestimation of the user's forgotten or unstable mastery of language units.

[0168] Figure 6 This is a structural block diagram of a forgetting risk update system with multimodal feedback, based on some embodiments. For example... Figure 6 As shown, the system includes: a language unit status data maintenance module, used to record and update in real time one or more of the following in the language unit status data: recognition status, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time; a multimodal perception data acquisition module, used to acquire the user's multimodal perception data through one or more of the following during the user's viewing of video data: image acquisition device and audio acquisition device; a potential unresolved event identification module, used to extract and identify features from the multimodal perception data to determine whether the user has generated a potential unresolved feedback event; a forgetting risk calculation module, used to calculate the forgetting risk value of language units that meet the preset recognition conditions in the language unit status data, based on one or more of the following: recognition status, historical verification results, number of consecutive successful verifications, number of valid occurrences, and recent contact time, as well as potential unresolved feedback events; a status update module, used to adjust the language unit status data of language units that meet the preset update conditions based on the forgetting risk value; and a recommendation impact module, used to recalculate the language unit load of subsequent video data relative to the user based on the updated language unit status data, and update the subsequent recommendation results. The functions of the above modules correspond to the functions of each step in the third embodiment, see the description of steps S510 to S560.

[0169] In addition, the forgetting risk update system may also include an abnormal segment identification module, a reference segment determination and comparison analysis module, a consolidation extraction module, and a valid occurrence frequency tracking module. The abnormal segment identification module identifies abnormal viewing segments based on at least one feedback behavior generated by the user during video viewing, such as pausing, rewinding, replaying, asking for help, confused expressions, or questioning voices. The reference segment determination module identifies segments where the user does not exhibit any feedback behavior and no other potentially uncomprehended signals as reference viewing segments. The comparison analysis module identifies language units appearing in both abnormal and reference viewing segments based on the language unit metadata of the video data, and compares the two; when the frequency of a target language unit appearing in an abnormal viewing segment is higher than its frequency in a reference viewing segment, it is identified as a candidate language unit to be verified or the verification priority of existing candidate language units is increased. The consolidation extraction module maintains at least two sets of language units that can be consolidated and extracted, including a set of recognized language units and a set of mastered language units. Language units are extracted from different sets according to a preset ratio for consolidation verification. For language units in the recognized language unit set, the extraction weight is determined based on the number of days since the most recent verification and historical verification results. The effective occurrence count tracking module is used to record the current effective occurrence count of each language unit and the baseline effective occurrence count when the language unit enters the recognition state. When the difference between the current effective occurrence count and the baseline effective occurrence count reaches a preset threshold, the language unit is added to the consolidation queue. In the consolidation process, the language units in the consolidation queue are consolidated and verified.

[0170] The workflow of all modules in the forgetting risk update system in the above embodiments is as follows: Figure 6 As shown.

[0171] In some examples, during the initial user test, while watching video data, or in the language unit verification and reinforcement stages after watching video data, in order to increase user interest and improve users' willingness to complete the corresponding tasks, the system can configure virtual assessment objects for the process to guide users to complete more language unit tests as much as possible.

[0172] For example, for language unit verification and consolidation stages such as initial testing, viewing video data, or viewing video data afterward, the system can package these stages as tasks to be completed collaboratively or competitively with a virtual object. The virtual object can be a virtual partner, a virtual opponent, a human-computer role, or a challenge role. The system determines the performance parameters of the virtual assessment object based on at least one of the following: the user's language unit status, current test, verification or consolidation round, answer accuracy rate, answer score, response time, cumulative correct answers, consecutive correct answers, cumulative incorrect answers, consecutive incorrect answers, user skipping behavior, and remaining test quantity. Performance parameters include at least one of the following: the virtual object's answering speed, answer accuracy rate, answer score, simulated correct or incorrect answers, cumulative correct answers, consecutive correct answers, cumulative incorrect answers, consecutive incorrect answers, lead or lag, interim win / loss results, encouraging feedback, challenging feedback, and reward feedback.

[0173] The system matches the performance of the virtual assessment subject to the user's current assessment level, allowing the user to experience leading, catching up, or winning in most assessment stages, and providing a moderate challenge in some stages. For example, if a user needs to consolidate multiple language units, the system generates a listening and picture selection consolidation task and configures a virtual opponent. When the user's recent accuracy rate is low, the system makes the virtual opponent's answering speed and accuracy slightly slower, allowing the user to lead most of the time; when the user answers correctly consecutively, the system increases the virtual opponent's accuracy or speed, keeping the task moderately challenging; when the user answers incorrectly consecutively or shows a tendency to quit, the system reduces the intensity of the virtual opponent's performance and outputs encouraging feedback.

[0174] The following describes a complete process of adaptive video data streaming, language unit verification, and forgetting risk update, in conjunction with the first to third embodiments, to demonstrate the collaborative operation of the three embodiments in a real-world application scenario.

[0175] I. Initialization Phase: Acquisition of Language Unit State Data When a user first enters the system, the system obtains the user's initial state data for each language unit through a precise initial test. Taking children's English listening learning as an example, the system uses a listen-and-select image method to test each unit: the system plays the pronunciation of the target word and simultaneously displays a candidate group containing the image or video corresponding to the target word and multiple images or videos corresponding to confusing words. The user selects the image or video corresponding to the pronunciation. If the answer is correct, the language unit is marked as recognized; if the answer is incorrect, it is marked as unrecognized or with low recognition. Function words, linking verbs, or prepositions such as "the," "to," "am," "is," and "are" are not included in the initial listen-and-select image test and are handled in the subsequent viewing and verification stage. After the initial test is completed, the system establishes a user language unit state database based on the test results, forming a recognized set, a set to be consolidated, and a set to be mastered, which serve as the data basis for subsequent recommendations and verification.

[0176] II. Recommendation Phase: Adaptive Streaming Based on user language unit status data, the system selects video data that meets preset recommendation conditions from the candidate video database and adds it to the playback sequence according to the adaptive streaming method in the first embodiment. The specific process is as follows: (1) The system reads the user's current language unit status data from the language unit status database and determines the knowledge set, the set to be consolidated set, and the set to be mastered set.

[0177] (2) The system reads multiple candidate video data and their language unit metadata from the video database, and calculates the overall language unit load and local language unit load of each candidate video data relative to the user in turn. The overall language unit load is calculated by counting the number of language units that are not recognized or whose recognition is below a preset threshold in the candidate video data or by the cumulative number of unrecognized language units; the local language unit load is evaluated by assessing the peak local comprehension burden in each segment in units of continuous playback time window or continuous language unit window.

[0178] (3) The system selects candidate video data whose overall language unit load is not higher than the preset overall load threshold and whose language unit load in each local window is not higher than the preset local load threshold, and adds them to the playback sequence. If multiple candidate video data meet the conditions, the system adjusts the recommendation order based on factors such as the number of unfamiliar language units, cumulative unfamiliarity or unfamiliarity, and viewing history.

[0179] (4) Before playing subsequent videos in the sequence to be played, the system recalculates the language unit load of the video based on the latest language unit status data and determines whether it still meets the preset recommendation conditions. If it no longer meets the conditions, the video is deleted or replaced from the sequence to ensure that subsequent videos in the continuously streamed sequence match the user's current real language level.

[0180] III. Viewing and Verification Phase: Multipath Language Unit Verification During or at a preset time after a user watches recommended video data, the system can also verify at least some language units. The language units to be verified can be determined directly according to preset verification rules, or based on the user's interactive behavior during video playback. The following explanation uses the language unit verification based on interactive behavior as an example from the second embodiment.

[0181] The specific process is as follows: (1) During video playback, the user terminal device continuously monitors the user's preset interactive operations, including clicking on help controls such as "didn't understand", triggering sentence-level playback control operations such as previous sentence / re-listen to current sentence, and dragging the playback progress bar to go back.

[0182] (2) When the above-mentioned preset interactive operation is detected, the system locates the corresponding target statement or target video segment according to the time correspondence between the time of the operation and the time of the video playback timeline; and extracts the language units appearing in the target statement or segment as candidate language units to be verified by combining the language unit metadata of the video data.

[0183] (3) The system adaptively generates corresponding verification tasks based on the language unit type of the candidate language units to be verified: for concrete nouns or action words, it generates recognition-type verification tasks based on audiovisual matching or sound-meaning matching; for function words, copulas, prepositions, or abstract words, it generates fill-in-the-blank and selection-type verification tasks based on contextual understanding or judgment and output-type verification tasks based on semantic processing. The contextualized listening fill-in-the-blank task's question stem is generated by constraints from the user's already mastered set of language units, ensuring that the user can understand the question stem itself and avoiding misjudgments due to the language units in the question stem exceeding the user's ability.

[0184] (4) The system obtains the user's verification answer and updates the status data of the corresponding language unit according to whether the answer is correct or incorrect. At the same time, the system can also generate a language dependency comprehension verification task. When the language unit verification passes but the comprehension verification fails, the weight of the passing result is reduced to avoid the user being misjudged as having mastered the language just because they answered correctly based on the picture or mechanical memory.

[0185] (5) After each verification is completed, the system writes the updated language unit status data into the language unit status database and triggers the recommended load recalculation process in the first embodiment to dynamically verify the subsequent video data in the playback sequence to ensure that the difficulty of the subsequent streaming content matches the user's latest language level.

[0186] IV. Forgetting Risk Tracking Phase: Forgetting Risk Updates Using Multimodal Feedback As users continuously watch video data, the system synchronously follows the forgetting risk update method in the third embodiment, continuously tracking the forgetting risk of recognized language units through multimodal perception data collection and analysis, and dynamically adjusting the language unit status data and subsequent recommendation results. The specific process is as follows: (1) The system continuously maintains the language unit status database and records in real time at least one of the following: whether each language unit is recognized, whether it is mastered, degree of recognition, whether it needs to be consolidated, number of occurrences, historical time of each occurrence, number of verifications, historical time of each verification, historical result of each verification, number of successful verifications, number of failed verifications, number of consecutive correct answers, number of consecutive incorrect answers, most recent verification result or most recent verification time, so as to provide a real-time data basis for forgetting risk calculation.

[0187] (2) When authorized by the user and meeting the preset collection conditions, the user terminal device collects the user's multimodal perception data through an image acquisition device and / or an audio acquisition device, including facial expression data, head posture data, gaze direction data, or user voice data (such as voice expressions expressing help, such as "What do you mean?" or "I don't understand"). The server performs feature extraction and recognition on the above multimodal perception data to determine whether the user has generated potential uncomprehended feedback events such as frowning, shaking their head, looking away from the screen, or making questioning remarks.

[0188] (3) The system calculates the forgetting risk value for language units that meet the preset recognition conditions, combining the time factor and the mastery strength factor. The time factor is based on a preset number of days-weighting function (e.g., Figure 7 As shown in the figure: language units with fewer historical correct answers correspond to a steep curve, and the risk of forgetting increases significantly in a short period of time; language units with more historical correct answers correspond to a flat curve, and the risk of forgetting increases significantly only after a longer period of time. The mastery intensity factor is determined by comprehensively considering the number of historical successful verifications, the number of consecutive successful verifications, and the number of valid occurrences. When a potential ununderstood feedback event is identified in step (2), the system includes it in the calculation as an additional forgetting risk signal, but does not directly trigger a state downgrade.

[0189] (4) For language units whose forgetting risk value exceeds the preset update condition, the system reduces their recognition level or moves them from the recognition set to the consolidation set. At the same time, the system performs abnormal segment identification and comparative analysis: based on the user's feedback behavior such as pausing, rewinding, and asking for help, abnormal viewing segments are identified and compared with the distribution of language units in the reference viewing segments. The system increases the verification priority of language units that appear frequently in abnormal viewing segments and have a low frequency of appearance in the reference viewing segments.

[0190] (5) In the consolidation phase, the system extracts language units from the sets of known and mastered language units according to a preset ratio for consolidation and verification. For known language units, the system determines the extraction weight based on the number of days since the most recent verification and the historical number of correct answers (see [link]). Figure 7 As shown in the days-weight function curve, language units with fewer correct answers have a faster weight increase in a shorter period of time and are more likely to be selected for reinforcement and verification.

[0191] (6) Based on the updated language unit status data, the system recalculates the language unit load of subsequent candidate video data in accordance with the method of the first embodiment, and re-filters and sorts the subsequent recommendation results. At the same time, the system counts the success rate of the user's most recent preset number of language unit verifications: when the success rate is lower than the preset threshold, the system judges that the current recommendation difficulty may be too high, and dynamically adjusts the language unit load threshold to reduce the difficulty of subsequent recommended content and avoid continuously pushing video data that exceeds the user's current comprehension ability.

[0192] V. Closed-loop operation The four stages described above operate cyclically as the user continues to use the system, forming a continuously self-calibrating recommendation loop: The recommendation stage filters video data of appropriate difficulty based on the user's current language unit status data and continuously streams it; the viewing verification stage directly determines the language units to be verified through preset verification rules, and / or accurately locates candidate language units to be verified through user interaction, and dynamically updates the language unit status data based on the verification results; the forgetting risk tracking stage continuously tracks the forgetting of learned language units through multimodal perception data and a days-weighting function, and feeds the updated results back to the recommendation stage. The methods in the three embodiments work together to enable the system to continuously obtain video data recommendations that match the user's actual language level, reducing the risk of overly difficult recommendations due to the system overestimating the user's language ability, improving the match between the difficulty of video data recommendations and the user's actual comprehension ability, and creating an efficient adaptive language learning experience.

[0193] This embodiment provides an electronic device including a processor and a memory. The memory can be used to store a computer program, and the processor can be used to retrieve the computer program from the memory to execute any of the method embodiments of this disclosure, including the adaptive streaming method in the first embodiment, the language unit verification method based on interactive behavior in the second embodiment, or the forgetting risk update method with multimodal feedback in the third embodiment. The processor can be one or more, and these processors can execute instructions individually or jointly; the memory can also be one or more, and these memories can store the aforementioned computer program individually or jointly.

[0194] This disclosure also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the adaptive streaming method, the language unit verification method based on interactive behavior, or the forgetting risk update method with multimodal feedback as described in the foregoing embodiments of this disclosure. Optionally, the computer-readable storage medium may be a non-transitory storage medium, but is not limited thereto; it may also be a temporary storage medium.

[0195] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0196] This disclosure may be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having a computer-readable program loaded thereon for causing a processor to implement any of the methods in the foregoing embodiments of this disclosure.

[0197] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media may include, for example, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0198] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include one or more of copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to computer-readable storage media in the respective computing / processing device.

[0199] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source or object programs written in any combination of one or more programming languages, including object-oriented programming languages ​​(e.g., Smalltalk, C++, etc.) and conventional procedural programming languages ​​(e.g., the "C" language or similar programming languages). The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network (e.g., a local area network or a wide area network), or it may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays, or programmable logic arrays, is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the embodiments of this disclosure.

[0200] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions. These computer-readable program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer-implemented process, thereby causing the instructions that execute on the computer, other programmable data processing apparatus or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0201] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It should be noted that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are all equivalent.

[0202] The various embodiments of this disclosure have been described above. These descriptions are exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of this disclosure is defined by the appended claims.

Claims

1. An adaptive streaming method, characterized in that, include: The language unit status data acquisition step involves acquiring the user's language unit status data. The language unit status data is recorded in units of language units, and the language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The video metadata acquisition step involves acquiring multiple candidate video data and language unit metadata for each candidate video data. The language unit metadata includes at least the language unit information appearing in the corresponding video data. The language unit load calculation step involves calculating the language unit load of each candidate video data relative to the user based on the language unit status data and the language unit metadata. The recommended determination step involves determining video data that meets preset recommendation conditions from the plurality of candidate video data based on the language unit load. The push output step involves forming or updating a sequence of videos to be played that includes one or more of the video data, based on the video data that meets the preset recommendation conditions.

2. The method according to claim 1, characterized in that, In the language unit load calculation step, the language unit load includes the overall language unit load and / or the local language unit load; The overall language unit load includes at least one of the following: the number of unrecognized language units in a single candidate video data, the ratio of recognized to unrecognized language units in a single candidate video data, the number of language units in a single candidate video data with a recognition level below a preset threshold, and the cumulative or weighted average value of the recognition or non-recognition levels of each language unit in a single candidate video data. The local language unit load is calculated in units of continuous playback time windows or continuous language unit windows within a single candidate video data.

3. The method according to claim 2, characterized in that, In the recommendation determination step, the preset recommendation conditions include: The overall language unit load is not higher than a preset overall load threshold; and / or The load of each local window language unit does not exceed the preset local load threshold.

4. The method according to claim 1, characterized in that, The language unit load calculation step also includes: The allowable language unit load threshold is determined based on the playback duration of the candidate video data. The shorter the playback duration of the candidate video data, the lower the allowable language unit load threshold, and the longer the playback duration of the candidate video data, the higher the allowable language unit load threshold.

5. The method according to claim 1, characterized in that, The language unit load calculation step also includes: When multiple video data belong to the same continuous story, the same theme unit, or the same preset content set, the multiple video data are used as aggregation candidate content, and the playback duration of the aggregation candidate content, the total language unit load of the aggregation candidate content, the diversity language unit load, and the local window language unit load are calculated. Whether to recommend the aggregated candidate content is determined based on the playback duration of the aggregated candidate content, the total language unit load of the aggregated candidate content, the language unit load of the episodes, and the language unit load of the local window.

6. The method according to claim 1, characterized in that, Also includes: The language unit set determination step involves determining one or more language unit sets based on the language unit status data and preset filtering conditions, and calculating the language unit load of each candidate video data relative to the user based on the one or more language unit sets and the language unit metadata, or the one or more language unit sets, the language unit status data, and the language unit metadata.

7. The method according to claim 6, characterized in that, The push output step also includes: Before playing subsequent video data in the sequence to be played, the language unit load of the subsequent video data relative to the user is recalculated based on the updated language unit status data and / or language unit set. Determine whether the subsequent video data still meets the preset recommendation conditions; if it no longer does, delete, replace, or pause the playback of the subsequent video data from the sequence to be played.

8. The method according to claim 1, characterized in that, Also includes: The dynamic adjustment process dynamically determines the language unit load threshold of candidate video data based on the user's current language proficiency. The user's current language proficiency is determined based on at least one of the following: the number of language units recognized, the number of language units mastered, the historical viewing time, the historical language unit verification success rate, or the historical comprehension verification success rate.

9. The method according to claim 1, characterized in that, In the step of acquiring language unit status data, the language unit status data includes at least one of the following: whether the user recognizes the language, whether the user has mastered it, the degree of recognition, whether the user needs to be reinforced, the number of times the language unit appears, the time of each historical appearance, the number of times the language unit is verified, the time of each historical verification, the result of each historical verification, the number of successful verifications, the number of failed verifications, the number of consecutive correct answers, the number of consecutive incorrect answers, the most recent verification result, or the most recent verification time.

10. The method according to claim 1, characterized in that, In the recommendation determination step, if multiple candidate video data all meet the preset recommendation conditions, the recommendation order is adjusted according to at least one of the following: the number of unfamiliar language units, the number of language units with recognition level below a preset threshold, the cumulative degree of unfamiliarity, the weighted average degree of recognition or degree of unfamiliarity, and the viewing history.

11. An adaptive push-stream system, characterized in that, include: The language unit status data acquisition module is used to acquire the user's language unit status data. The language unit status data is recorded in units of language units. The language unit includes at least one of the following: word, phrase, idiom, character, word, syllable, sentence fragment, grammatical structure, or fixed expression. The video metadata acquisition module is used to acquire multiple candidate video data and language unit metadata of each candidate video data. The language unit metadata includes at least the language unit information appearing in the corresponding video data. The language unit load calculation module is used to calculate the language unit load of each candidate video data relative to the user based on the language unit status data and the language unit metadata. The recommendation determination module is used to determine, based on the language unit load, video data to be recommended that meets preset recommendation conditions from the plurality of candidate video data; The push output module is used to form or update a sequence of videos to be played, which includes one or more of the video data, based on the video data that meets the preset recommendation conditions.

12. The system according to claim 11, characterized in that, The language unit load calculation module is also used for: The allowable language unit load threshold is determined based on the playback duration of the candidate video data. The shorter the playback duration of the candidate video data, the lower the allowable language unit load threshold, and the longer the playback duration of the candidate video data, the higher the allowable language unit load threshold.

13. The system according to claim 11, characterized in that, The language unit load calculation module is also used for: When multiple video data belong to the same continuous story, the same theme unit, or the same preset content set, the multiple video data are used as aggregation candidate content, and the playback duration of the aggregation candidate content, the total language unit load of the aggregation candidate content, the diversity language unit load, and the local window language unit load are calculated. Whether to recommend the aggregated candidate content is determined based on the playback duration of the aggregated candidate content, the total language unit load of the aggregated candidate content, the language unit load of the episodes, and the language unit load of the local window.

14. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1-10.