CN-121971861-A - Double-layer self-adaptive training recommendation system and method based on standardized performance increment and dynamic baseline

CN121971861ACN 121971861 ACN121971861 ACN 121971861ACN-121971861-A

Abstract

The invention discloses a double-layer self-adaptive training recommendation system and a double-layer self-adaptive training recommendation method based on standardized performance increment and dynamic baselines, which relate to the technical fields of artificial intelligence technology, education technology and cognitive science, wherein the system is provided with a double-state switching mechanism, a clear initialization path is provided for new users, user experience from 0 to N is guaranteed, a baseline storage module introduces a standardized performance score SPS, the defect that cross-game types are incomparable is overcome through mean value and variance calibration, a game fatigue GFS model can avoid boredom of children due to repetition, an initialization state or a dynamic self-adaptive state is determined by detecting total game times of users, training loads are set, SPS sequences and NPD data of a latest k-field game are used for calculating SPS_level and SPS_ trend, training load N is used for generating training class packets, finally, in-game difficulty regulation and control are realized based on DGB, fatigue switching can be performed, and training enthusiasm is guaranteed.

Inventors

Mu Ai
GAO JUNHUI
MA SHUAI
YANG ZHICHENG
Lei Zhongliu
CAO JIE

Assignees

深圳鹏脑科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (10)

1. A dual-layer adaptive training recommendation system based on standardized performance increments and a dynamic baseline, comprising: The dynamic game baseline module is used for storing and acquiring corresponding average scores of the histories of the same-aged people as a dynamic game baseline DGB according to the ages of the users, the game IDs and the level of the level; The standardized performance evaluation module is used for calculating standardized performance increment NPD according to the game score of the user and the corresponding DGB, and calculating standardized performance score SPS according to the NPD and the historical performance score of the game; the state switching module is used for switching between an initialization state and a dynamic self-adaptive state according to the number of the game records completed by the user; The self-adaptive training decision module calculates a standardized expression level SPS_level through weighted average based on an SPS sequence of a user nearest continuous multi-game when the self-adaptive training decision module is in a dynamic self-adaptive state, calculates a standardized expression trend SPS_ trend through linear regression, and outputs a variable training load value N according to the combination condition of the SPS_level and the SPS_ trend; The game fatigue management module is used for maintaining a game fatigue GFS score for each game, wherein the GFS score is updated according to the number of times the game is played, a time attenuation rule and a finish cooking penalty rule when a user reaches a preset finish cooking threshold value; The lesson generating module is used for selecting game types according to the training load value N and the capability short board of the user, preferentially selecting games with the lowest GFS score from the corresponding game types, and generating training lesson bags containing N games; And the in-game regulation and control module is used for dynamically regulating the difficulty level of the next level according to the score of the current level of the user and the comparison result of struggle line and finish line calculated based on DGB.
2. The dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baseline of claim 1, wherein: In the standardized performance evaluation module, the calculation mode of the standardized performance increment NPD is NPD= (user score-DGB)/DGB; The calculation mode of the standardized expression score SPS is SPS= (user NPD-mu)/sigma; wherein mu represents the average value of NPD values generated by all users in the history database in a specific game type, and sigma represents the standard deviation of NPD values generated by all users in the history database in the specific game type.
3. The dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baseline of claim 1, wherein: SPS_level is a weighted average of the last K SPS values, and the weight increases from far to near along with the game time; SPS trend is the unitary linear regression slope m of the last K NPD data points; K represents a preset game number threshold.
4. A dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baselines as claimed in claim 3, wherein: the unitary linear regression slope m represents the amplified ratio of each game played, relative to the average, introducing a constant ψ, where: m > ψ represents that the recent comprehensive cognitive state of the user is in a significant progress trend; m < - ψ represents that the recent comprehensive cognitive state of the user is in a obviously descending trend; - ψ is less than or equal to m is less than or equal to ψ, representing that the recent comprehensive cognitive state of the user is in a consolidation and stabilization period.
5. The dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baseline of claim 1, wherein: the state switching module is in an initialized state when the number K of the game records of the user is smaller than a preset threshold value K, and is switched to a dynamic self-adaptive state otherwise; In the initialized state, the system uses a fixed training load, and preferably selects a game with a GFS score of zero, and performs difficulty regulation and control in the game according to DGB.
6. The dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baseline of claim 1, wherein: In the adaptive training decision module, the training load value N is determined according to the following logic: if the SPS_level is higher than a first threshold, outputting a load reduction training load; if SPS_level is lower than the second threshold and SPS_ trend is greater than zero, outputting an added training load; if SPS_level is lower than a second threshold and SPS_ trend is less than zero, outputting a load-reducing training load; The rest of the cases output standard training loads.
7. The dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baseline of claim 1, wherein: In the game fatigue management module, the GFS score updating rule includes: rule A, increase the first preset value after every game is played; Rule B, decreasing the games with GFS greater than zero by a second preset value every preset time; and C, when the user reaches the preset finish threshold value for the first time in the game, adding a third preset value as the finish penalty, and not triggering the rule A any more.
8. The dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baseline of claim 1, wherein: in the intra-game regulation and control module, struggle lines and finish-maturing lines are obtained through calculation of DGB and preset floating percentage P: The struggle line is calculated in a mode that struggle line=dgb (1-P); The fine cooked line is calculated by the method that fine cooked line=dgb (1+P).
9. The dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baselines of claim 8, wherein: The regulation logic of the intra-game regulation module comprises: if the score is higher than the accurate maturation line, starting from the current checkpoint jump level next time; If the score is between struggle line and finish-maturing line, starting from current checkpoint upgrading next time; if the score is lower than the struggling line, the next time the current checkpoint is downgraded; if the score is lower than the struggle line a plurality of times in succession, the current game is aborted and the user is guided to the next game of the class bag.
10. A double-layer self-adaptive training recommendation method based on standardized performance increment and dynamic base line is characterized by comprising the following steps: step S1, judging that the system is in an initialized state or a dynamic self-adaptive state according to the number of history records of the total games of the user; S2, when the game is in a dynamic self-adaptive state, extracting an SPS sequence of a user nearest continuous K-field game, calculating a performance level SPS_level through weighted average, and calculating a performance trend SPS_ trend through linear regression; step S3, determining and outputting a training load value N according to the combination condition of the SPS_level and the SPS_ trend; Step S4, according to the game fatigue GFS score, training load value N and user capacity short board of each game, preferentially selecting the game with the lowest GFS score from the games with the corresponding types, and generating a training class packet containing a plurality of games; and S5, adjusting the difficulty level of the next checkpoint in real time according to the comparison result of the current checkpoint score of the user and the struggling line and the finish line calculated based on the DGB.

Description

Double-layer self-adaptive training recommendation system and method based on standardized performance increment and dynamic baseline Technical Field The invention relates to the technical fields of artificial intelligence technology, education technology and cognitive science, in particular to a double-layer self-adaptive training recommendation system and method based on standardized expression increment and dynamic base line, which are a personalized recommendation system and training method for training cognitive ability of children (such as self-control ability, reaction ability, memory and the like), and can dynamically adapt to training load, content type and difficulty level based on instant expression state and long-term historical trend of users so as to realize accurate and efficient cognitive training. Background The cognitive ability training of children is a key way for promoting the brain development and comprehensive quality improvement of children, and the personalized recommendation technology is a core support for guaranteeing the training effect and the user participation. However, the conventional cognitive training system has a plurality of defects in the personalized recommendation link, and seriously affects the training quality and the user experience: 1. The existing system generally relies on historical performance data of users to construct a recommendation model, when a new user uses the system for the first time, individual recommendation cannot be obtained due to no data accumulation, only fixed standardized content can be received, initial experience is poor, and training interests are difficult to build. 2. The fixed score threshold (such as 50 scores and 90 scores are excellent) is adopted as an evaluation standard, the difference of difficulty curves and the difference of score upper limits of different training games are ignored, the performance evaluation result of the cross-game is distorted, and the real cognitive level of a user cannot be reflected. 3. The user with higher level but in fatigue descending trend and the user with lower level but in rapid progress trend may have the same average value, but the required training scheme is completely opposite, so that the recommended result is not matched with the actual requirement of the user. 4. The traditional short-board training logic machinery repeatedly recommends games with poor user performance, lacks of content rotation and freshness design, and particularly aims at children users, contradiction psychology is easily caused in the boring repeated training, and training persistence and participation enthusiasm are reduced. Therefore, there is a need for a cognitive training recommendation system and method that can address the above-mentioned drawbacks, and achieve smooth transition, scientific evaluation, personalized content recommendation, and dynamic load adjustment for new users. Disclosure of Invention The invention provides a technical scheme capable of solving the problems in order to overcome the defects of the prior art. A dual-layer adaptive training recommendation system based on normalized performance delta and dynamic baseline, comprising: The dynamic game baseline module is used for storing and acquiring corresponding average scores of the histories of the same-aged people as a dynamic game baseline DGB according to the ages of the users, the game IDs and the level of the level; The standardized performance evaluation module is used for calculating standardized performance increment NPD according to the game score of the user and the corresponding DGB, and calculating standardized performance score SPS according to the NPD and the historical performance score of the game; the state switching module is used for switching between an initialization state and a dynamic self-adaptive state according to the number of the game records completed by the user; The self-adaptive training decision module calculates a standardized expression level SPS_level through weighted average based on an SPS sequence of a user nearest continuous multi-game when the self-adaptive training decision module is in a dynamic self-adaptive state, calculates a standardized expression trend SPS_ trend through linear regression, and outputs a variable training load value N according to the combination condition of the SPS_level and the SPS_ trend; The game fatigue management module is used for maintaining a game fatigue GFS score for each game, wherein the GFS score is updated according to the number of times the game is played, a time attenuation rule and a finish cooking penalty rule when a user reaches a preset finish cooking threshold value; The lesson generating module is used for selecting game types according to the training load value N and the capability short board of the user, preferentially selecting games with the lowest GFS score from the corresponding game types, and generating training lesson bags containing N games;