CN-121119841-B - Intelligent teaching supervision evaluation method and system based on large model

CN121119841BCN 121119841 BCN121119841 BCN 121119841BCN-121119841-B

Abstract

The invention relates to the technical field of neural networks, in particular to a large model-based intelligent teaching supervision evaluation method and system, which comprises the following steps of collecting teacher multi-source data and labeling normalization processing, constructing a time sequence behavior set and a label ordering sequence, clustering high-frequency concurrent behaviors, marking conflicts, extracting a change trend adjustment boundary, ordering and grading jump leading labels, establishing a grading linkage path by connecting behaviors, and outputting a teaching supervision evaluation scheme. According to the invention, through construction and time sequence integration of the multi-source data tag, structural recombination and dynamic expression of teaching behaviors are realized, behavior classification accuracy is improved by combining behavior frequency clustering and conflict recognition, key behavior features are extracted by means of trend ratio and score jump alignment, a score influence factor ordering and behavior flow direction chain is established, chain mapping of teaching behaviors and score change is realized, evaluation discrimination capability and data response efficiency are enhanced, and accuracy of teaching behavior recognition and resolution of a scoring mechanism are improved.

Inventors

XIAO YU
DU JUAN
LIU XIN
LIU KAN

Assignees

武汉天天互动科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251021

Claims (7)

1. The intelligent teaching supervision evaluation method based on the large model is characterized by comprising the following steps of: S1, collecting teacher voice, operation behaviors and action track data, labeling the behaviors according to start and stop time of an event, calling voice duration, track change intervals and instruction quantity to perform normalization processing, and splicing the voice duration, the track change intervals and the instruction quantity into a behavior set according to time sequence to construct a label sequencing sequence; s2, counting the occurrence frequency of tags in adjacent periods according to the continuous time position of the tag sequencing sequence, clustering high-frequency concurrency behaviors in the periods, extracting conflict behavior combination and marking positions according to mutual exclusion rules, and outputting conflict shielding tag fragment groups; s3, constructing a ratio characteristic of tag change frequency and pause behavior quantity by utilizing the time behavior content of the conflict shielding tag fragment group, extracting behavior change trend and comparing the behavior change trend with adjacent time periods, aligning through jump change node positions, and adjusting a boundary to form a time period boundary structure; S4, extracting behaviors and scoring sequences from the time interval boundary structure, calculating a scoring curve fluctuation range, positioning jump points, extracting all behavior labels in the range, sorting according to the frequency of occurrence in a section and the difference value of the whole frequency, and outputting a jump fragment leading behavior set; The step of obtaining the time interval boundary structure comprises the following steps: S301, based on time behavior content in the conflict shielding tag fragment group, counting the change frequency and the pause behavior quantity of each type of tag in a section, calculating a ratio relation between the change frequency and the pause behavior quantity, and generating a tag pause ratio sequence; s302, calling the label pause ratio sequence, extracting a change trend curve, comparing the change trend with the change trend of the front and rear time periods, calculating the difference degree of the continuous trend, and obtaining a behavior change trend difference value; s303, calling a jump change node position sequence according to the behavior change trend difference value, judging an alignment offset value of a trend jump position, synchronously adjusting the section boundary position, acquiring a section change amplitude sequence, and establishing a time period boundary structure.
2. The large model-based intelligent teaching supervision evaluation method according to claim 1, wherein the tag ordering sequence comprises a behavior type tag sequence, a standardized time node sequence and a periodic behavior pattern, the conflict shielding tag fragment group comprises a structure conflict tag group, a shielding identification site set and a concurrent behavior clustering result, the time period boundary structure comprises a behavior density characteristic parameter, a jumping node alignment boundary and an adjusted section division, and the jumping fragment dominant behavior set comprises a grading mutation region tag set, a dominant behavior type list and a tag frequency difference ordering result.
3. The method for evaluating intelligent teaching supervision based on a large model according to claim 2, wherein the step of obtaining the tag ordering sequence comprises the following steps: s101, labeling and marking each type of behavior according to the start-stop time of an event based on voice, operation behaviors and action track data of a teacher in a class, extracting the duration of the voice, the track change interval and the number of instruction events, and generating a behavior parameter set; s102, carrying out normalization processing on the duration of the voice, the track change interval and the number of instruction events according to the behavior parameter set to generate a periodic parameter normalization sequence; S103, calling the periodic parameter normalization sequence, splicing according to time sequence to construct a multidimensional vector set, and carrying out combination analysis according to the relevance among the behavior parameter ordering trend, the track change and the voice span to obtain a label ordering sequence.
4. The intelligent teaching supervision evaluation method based on the large model according to claim 3, wherein the step of obtaining the conflict shielding label segment group is as follows: S201, calculating the occurrence frequency of the labels in adjacent periods based on continuous time positions in the label sorting sequence, identifying the labels with frequency difference change, and obtaining a label frequency difference sequence; S202, calling the tag frequency difference sequence, extracting tag combinations which occur in a period in a high-frequency concurrency manner, and carrying out type clustering division according to the tag co-occurrence relationship to generate a tag combination cluster group; S203, according to the label combination cluster group, calling a teaching behavior mutual exclusion judging rule, identifying label combinations with structural conflict characteristics, and marking corresponding periodic positions to obtain a conflict shielding label fragment group.
5. The method for evaluating intelligent teaching supervision based on a large model according to claim 1, wherein the step of obtaining the jump segment dominant behavior set is: S401, based on the time interval boundary structure, extracting a behavior label sequence and a scoring sequence in a section, calculating a continuous fluctuation range between node values of a scoring curve, judging the position of a jump point in a fluctuation interval, and obtaining a scoring jump position set; S402, invoking the scoring jump position set, extracting all teacher behavior type labels in the corresponding jump section, and counting the occurrence frequency of the labels in the section to obtain a jump section behavior frequency distribution result; s403, according to the distribution result of the hopping section behavior frequency, comparing the difference value of the label frequency in the section with the overall label frequency, calculating a difference value sorting index, carrying out weighting judgment by combining the behavior distribution characteristics, calculating to obtain the frequency offset intensity value of the label, sorting, screening the leading item, and obtaining the hopping fragment leading behavior set.
6. The large model-based intelligent teaching supervision evaluation method according to claim 1, further comprising: s5, according to the time positions of the jump fragment leading behavior sets, arranging and linearly connecting adjacent behaviors according to appear for the first time sequences, establishing a teaching action flow direction path sequence on a time axis, carrying out chain linkage mapping of action content and grading change, and outputting an intelligent teaching supervision evaluation scheme; the intelligent teaching supervision evaluation scheme comprises a teaching action flow direction map, a grading change linkage mapping path and a time sequence event chain structure; the intelligent teaching supervision evaluation scheme comprises the following acquisition steps: S501, extracting the first appearance position of each type of tag in a time axis based on the jump fragment leading behavior set, arranging all tag items according to time sequence, constructing a tag time propulsion sequence, and acquiring behavior tag time sequencing; S502, calling the time sequence of the behavior labels, constructing linear connection paths according to the time intervals of adjacent labels, and sequentially combining the linear connection paths into a continuous behavior flow path to generate a teaching action flow path sequence; S503, according to the teaching action flow path sequence, tracking the time mapping relation between the labels and the scoring jump points in the path, analyzing the linkage degree of the labels and the scoring response, calculating the scoring linkage response value of the acquired path, and constructing a dynamic mapping mechanism between the labels and the scoring based on the linkage degree to obtain an intelligent teaching supervision evaluation scheme.
7. A large model-based intelligent teaching supervision evaluation system for executing the large model-based intelligent teaching supervision evaluation method as set forth in any one of claims 1 to 6, comprising: the behavior period splicing module acquires the length of teacher voice, the number of operation instructions and a track interval, intercepts the voice according to energy after segmenting according to event time, orders the operation according to accumulated numbers, calculates an offset accumulated value according to the track, and splices according to time after labeling to obtain a label ordering sequence; the concurrency conflict detection module calls the label sequencing sequence to count the label times of the time interval, compares the periodic distribution and judges the label combination by using a mutual exclusion rule, screens the time to be shielded and marks the time, and obtains a conflict shielding label fragment group; the trend jump alignment module extracts the tag change based on the conflict shielding tag fragment group, calculates the change times and pause ratio sequence, compares adjacent difference values and adjusts the position according to jump nodes to obtain a time period boundary structure; The scoring waveband extraction module invokes a label and scoring sequence in the time interval boundary structure, measures scoring fluctuation and locates hopping, extracts hopping range labels to sort according to the difference value of the times and the total number, and obtains a hopping fragment leading behavior set; And the flow direction link establishment module is used for connecting the tag time sequences to form a link based on the first time sequencing in the jump fragment leading behavior set, so as to obtain an intelligent teaching supervision evaluation scheme.

Description

Intelligent teaching supervision evaluation method and system based on large model Technical Field The invention relates to the technical field of neural networks, in particular to a large-model-based intelligent teaching supervision evaluation method and system. Background The core matters comprise construction, training and application of an artificial neural network, and cover specific implementation modes of a deep neural network, a convolutional neural network, a cyclic neural network, a graph neural network and the aspects of image recognition, natural language processing, predictive modeling, intelligent control and the like, and the intelligent neural network belongs to an intelligent computing system for simulating neurons and connection relations thereof to process complex data and abstract characteristics. The development of the technical field promotes the wide application of a large-scale pre-training model, a big data driving intelligent system and a self-adaptive learning mechanism, and gradually expands to the industries of education, medical treatment, finance and the like. The traditional intelligent teaching supervision evaluation method is used for solving the subjectivity and limitation problems of teaching evaluation by adopting a mode based on combination of index scoring and expert manual observation in the teaching quality monitoring and teacher behavior evaluation process, and is generally carried out by means of teaching process record acquisition, evaluation dimension index setting, expert scoring quantification and the like. The intelligent judgment and the structural expression of teaching supervision data are completed by adopting a large-scale neural network model to perform feature extraction and semantic analysis on teaching behavior data and combining data sources such as audio and video of a teacher class, teaching text content and the like, so that the association relation analysis between complex cognitive behavior modeling and multidimensional data in the teaching process is realized. In the prior art, on the basis of index setting and manual judgment, the acquisition record and the manual scoring quantification are relied on in behavior recognition, and because the scoring depends on subjective standards, behavior labels do not have clear structural sequence relations, evaluation contents are difficult to cover continuous changes and concurrent situations of teaching behaviors, when complex scenes such as high-density behavior aggregation or scoring jump are faced, key behavior characteristics and time node association relations are difficult to effectively extract, evaluation ambiguity is easily caused under a scoring greatly-fluctuating scene, the behavior mode lacks clear pointing, and the establishment of a refined analysis and causality chain cannot be supported, so that evaluation conclusion generalization, behavior analysis one-sided and feedback suggestion pertinence are insufficient. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a large-model-based intelligent teaching supervision evaluation method. In order to achieve the purpose, the invention adopts the following technical scheme that the intelligent teaching supervision evaluation method based on the large model comprises the following steps: S1, collecting teacher voice, operation behaviors and action track data, labeling the behaviors according to start and stop time of an event, calling voice duration, track change intervals and instruction quantity to perform normalization processing, and splicing the voice duration, the track change intervals and the instruction quantity into a behavior set according to time sequence to construct a label sequencing sequence; s2, counting the occurrence frequency of tags in adjacent periods according to the continuous time position of the tag sequencing sequence, clustering high-frequency concurrency behaviors in the periods, extracting conflict behavior combination and marking positions according to mutual exclusion rules, and outputting conflict shielding tag fragment groups; s3, constructing a ratio characteristic of tag change frequency and pause behavior quantity by utilizing the time behavior content of the conflict shielding tag fragment group, extracting behavior change trend and comparing the behavior change trend with adjacent time periods, aligning through jump change node positions, and adjusting a boundary to form a time period boundary structure; and S4, extracting behaviors and scoring sequences from the time interval boundary structure, calculating the fluctuation range of a scoring curve, positioning the jumping points, extracting all behavior labels in the range, sorting according to the difference value between the occurrence frequency in the section and the overall frequency, and outputting a jumping fragment dominant behavior set. As a further scheme of the invention, the tag ordering sequence compr