An intelligent evaluation system for study tours based on an agent

Abstract

This invention discloses an intelligent evaluation system for study tours based on intelligent agents, including a multimodal data acquisition module, a data intelligent analysis module, a comprehensive analysis and processing module, and an intelligent evaluation output module. This invention relates to the field of intelligent evaluation technology and solves the technical problem that study tour evaluation is currently hampered by an overemphasis on results and a neglect of process and effectiveness, making it difficult to meet the needs of quality education for process-oriented, comprehensive, and personalized evaluation. This invention, by relying on a knowledge graph and dynamic weight model in the field of study tour evaluation, achieves the adaptation of indicator dimensions to themes / grades, the adjustment of granularity to data richness, and the dynamic optimization of weights to objectives / features. It completely solves the problem of the rigid, one-size-fits-all approach of traditional indicators. Through a three-dimensional trajectory model of space-time-behavior, it depicts the entire study tour process and combines a hidden Markov model and the Viterbi algorithm to decode behavioral sequences, achieving intelligent identification and quantitative analysis of behavioral patterns and improving the accuracy of behavioral pattern recognition.

Description

Technical Field

[0001] This invention relates to the field of intelligent evaluation technology, specifically to an intelligent evaluation system for study tours based on intelligent agents. Background Technology

[0002] As a core practical vehicle for quality education, the scientific nature and comprehensiveness of the evaluation system of study tours directly determine the effectiveness of the implementation of educational goals.

[0003] With the development of educational informatization, existing technologies have attempted to introduce smart devices and data analysis into research-based learning evaluation, but many technical bottlenecks remain that are difficult to overcome: Existing systems mostly rely on single electronic attendance or paper records, collecting only structured data such as sign-in and test scores, while omitting unstructured data such as voice interaction, collaborative behavior, and experimental operations, resulting in a one-sided evaluation dimension; at the same time, the lack of standardized de-identification mechanisms in data preprocessing makes it easy to leak students' personal information, which does not meet the requirements of data security regulations. Traditional evaluation uses a fixed indicator framework, which cannot dynamically adjust the dimensions and granularity of indicators according to the research and study theme and the students' learning stage. It has a one-size-fits-all problem and is difficult to match the educational goals of different scenarios. Existing technologies can only record single-point behavior or time information, lacking three-dimensional correlation modeling of space-time-behavior, and cannot fully depict students' learning trajectories; behavior pattern recognition mostly relies on manual rules or single algorithms, which are difficult to handle the uncertainty of behavior sequences, resulting in missed detections and misjudgments. Existing systems rely heavily on teachers' manual observation to identify abnormal student behavior, lacking quantitative judgment standards and dynamic baseline references. This results in significant subjective bias in abnormal identification, a high rate of missed detections, and an inability to trigger timely intervention prompts. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent evaluation system for study tours based on intelligent agents. This system solves the problem that study tour evaluation is currently focused on results rather than process and form rather than effectiveness, making it difficult to meet the needs of quality education for process-oriented, comprehensive, and personalized evaluation.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent evaluation system for study tours based on intelligent agents, comprising: The multimodal data acquisition module is used to collect full-dimensional data through a multimodal acquisition terminal, and transmit the full-dimensional data to the data intelligent analysis module after preprocessing. The data intelligence analysis module is used to extract key features from the preprocessed full-dimensional data, call the knowledge graph of the study tour evaluation field and combine it with the meta-information of the study tour scenario, deconstruct the basic indicator dimensions, adjust the indicator granularity and allocate indicator weights, generate a dynamic indicator system and then transmit it to the comprehensive analysis and processing module. The comprehensive analysis and processing module is used to construct a three-dimensional trajectory model of space-time-behavior based on the dynamic indicator system and preprocessed multi-source time series data. It performs state decoding on the behavior sequence through the hidden Markov model, identifies behavior patterns and calculates quantitative indicators, identifies abnormal patterns based on preset thresholds and dynamic baselines, and transmits abnormal pattern information to the intelligent evaluation output module. The intelligent evaluation output module is used to display abnormal mode information to administrators.

[0006] As a further embodiment of the present invention, the multimodal acquisition terminal includes a student smart terminal, a scene IoT device, an audio and video acquisition device, and a work upload port, wherein the full-dimensional data includes structured data and unstructured data.

[0007] As a further aspect of the present invention, the preprocessing includes: removing outliers and duplicate data, filling in missing values, converting unstructured data into an analyzable format and unifying the data timestamp and spatial coordinate system, extracting key audio and video information, labeling text with thematic tags, and desensitizing data containing personal identifiers.

[0008] As a further embodiment of the present invention, the knowledge graph in the field of study evaluation includes three layers of nodes: literacy dimension, evaluation index, and data feature, as well as the relationships between nodes. The generation of the dynamic indicator system includes matching indicator dimensions with the research and study theme and the students' learning stage, adjusting the indicator granularity based on the richness of data features, and allocating indicator weights through a dynamic weight model.

[0009] As a further aspect of the present invention, the dynamic weight model adopts a goal-oriented and data feedback dual-drive mechanism. Initial weights are generated based on preset educational goals using the analytic hierarchy process. The initial weights are then corrected using a random forest regression model based on the discriminative power of the indicator features. Indicators with higher discriminative power have higher weights.

[0010] As a further aspect of the present invention, the three-dimensional trajectory model uses the time axis as the core driving axis, including: Timeline: Divided into preset time slices, recording the start and end times of each slice and the corresponding study tour segment; Spatial dimension layer: Overlaying location coordinate clustering results with scene semantic labels to identify high-frequency activity areas; Behavior dimension layer: Behavior types are labeled based on multi-source temporal features, and automatic labeling is achieved through behavior-feature mapping rules; The three-dimensional trajectory model forms a dynamic trajectory by continuously stitching together time slices, and short-term abnormal fluctuations are corrected by a smoothing algorithm.

[0011] As a further embodiment of the present invention, the Hidden Markov Model includes a hidden state set, an observation sequence, a state transition probability, and an observation probability. The hidden state set represents typical behavioral patterns in the study tour scenario, and the observation sequence represents the spatial-behavioral-physiological fusion features of each time slice in the three-dimensional trajectory model. The Viterbi algorithm is used to solve for the optimal hidden state sequence and output the behavioral pattern flow trajectory.

[0012] As a further aspect of the present invention, the method for identifying behavioral patterns and calculating quantitative indicators is as follows: Acquire the flow trajectory of behavioral patterns and calculate quantitative indicators through preset rules, including the duration percentage of a certain behavioral pattern S. i Total duration = number of consecutive time slices of this pattern × duration of a single slice, percentage = (S i (Total duration / Total study tour duration) × 100%; Mode conversion frequency: Count the number of times any two different modes are converted. Frequency = Number of conversions / Total number of conversions × 100%; Conversion path characteristics: High-frequency conversion paths with a frequency of ≥10% are extracted and their average interval duration is recorded to form a path preference map.

[0013] As a further aspect of the present invention, the abnormal pattern recognition includes dual determination: Based on preset thresholds, single-pattern anomalies and transition anomalies are identified. Combined with a dynamic baseline formed by the average behavior patterns of students in the same batch, anomalies that deviate from the preset range of the average are identified, anomaly marking is automatically triggered, and anomaly pattern information is generated.

[0014] As a further aspect of the present invention, for a single anomaly, if the percentage of problem stuck mode is ≥30%, or if the hesitant and observing mode appears continuously for ≥5 slices, it is determined to be an anomaly. For a conversion anomaly, if the experimental operation results in problem stuck conversion ≥3 times consecutively, or if a high-frequency path is missing, it is determined to be an anomaly.

[0015] This invention provides an intelligent evaluation system for study tours based on intelligent agents. Compared with existing technologies, it has the following advantages: This invention achieves full-dimensional capture of structured and unstructured data through multimodal terminals. Combined with standardized preprocessing, it not only solves the problem of one-sided data dimensions in traditional methods but also ensures student privacy and security. Relying on knowledge graphs and dynamic weight models in the field of study tour evaluation, it enables indicator dimensions to adapt to themes / grades, granularity to adjust with data richness, and weights to be dynamically optimized according to objectives / features. This completely solves the problem of rigid, one-size-fits-all traditional indicators. It depicts the entire study tour process through a three-dimensional trajectory model of space-time-behavior and combines hidden Markov models and the Viterbi algorithm to decode behavioral sequences, achieving intelligent identification and quantitative analysis of behavioral patterns. This overcomes the fragmented and subjective defects of traditional analysis, improves the accuracy of behavioral pattern identification, and adopts a dual judgment mechanism of preset thresholds and dynamic baselines to quantify anomaly identification standards. It automatically generates early warning information containing anomaly types, time periods, and related features, solving the problems of large subjective biases and delayed intervention in traditional anomaly identification, and providing accurate basis for teachers' timely guidance. Attached Figure Description

[0016] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 This application provides an intelligent evaluation system for study tours based on intelligent agents, including: a multimodal data acquisition module, a data intelligent analysis module, a comprehensive analysis and processing module, and an intelligent evaluation output module, and combined with... Figure 1 It can be seen that the information between the above functional modules is transmitted in one direction only.

[0019] The multimodal data acquisition module is used to collect multi-dimensional data through multimodal acquisition terminals, including student smart terminals, scene IoT devices, audio and video acquisition devices, and work upload ports. The multi-dimensional data includes structured data and unstructured data. The structured data includes attendance information, task completion rate, and test scores, while the unstructured data includes voice interaction, behavioral interaction, team collaboration videos, research reports, and images of handicraft works. The module also preprocesses the obtained multi-dimensional data and transmits the preprocessed multi-dimensional data to the data intelligent analysis module.

[0020] The preprocessing includes removing false alarms and duplicate data from sensors using outlier detection algorithms, and filling in missing values ​​using time-series interpolation or mode imputation; converting unstructured data into an analyzable format and unifying the data timestamp format and spatial coordinate system; extracting key information from audio and video data, performing preliminary word segmentation and topic tagging on text works; blurring faces in image data containing personal identifiers, removing voiceprint information from voice data, and hashing sensitive fields in structured data.

[0021] The data intelligence analysis module is used to process the acquired preprocessed full-dimensional data and extract key features from structured and unstructured data. Key features of structured data include task completion rate time series curves, objective question score distribution, standard deviation of time spent on group collaborative tasks, and percentage of scene dwell time. Key features of unstructured data include keyword frequency in voice interaction, collaborative action recognition results in behavioral videos, theme relevance in text works, and innovative feature vectors of handmade works. Next, the intelligent agent calls the knowledge graph of the research and study evaluation domain, takes core competencies as the top-level framework, and deconstructs the basic indicator dimensions by combining meta-information of research and study scenarios. The knowledge graph consists of three layers of nodes: competency dimension, evaluation indicator, and data feature, as well as the relationship between nodes. Specifically, if the research and study theme is a STEM experiment, the dimensions of scientific inquiry, experimental operation specifications, and data rigor are deconstructed first. If the theme is humanities research, the focus should be on deconstructing dimensions such as the application of historical knowledge, interview skills, and depth of cultural understanding; For lower elementary school students, the indicators are simplified to observable metrics such as participation, completion of basic tasks, and simple collaboration; for high school students, they are refined to higher-order metrics such as critical thinking, quality of innovative proposals, and cross-group resource integration. Simultaneously, based on the feature richness of the preprocessed data, the granularity of the indicators is automatically adjusted. If a certain dimension has rich data features, it is refined into a third-level indicator; if a certain dimension has limited data features, it is simplified to a first-level indicator. Through the dynamic weight model built into the agent, the weights of each indicator are automatically assigned according to the distribution of data features and the priority of educational goals. The specific processing method is as follows: A hybrid model driven by both goal orientation and data feedback is adopted. Based on the preset educational goals, the initial weights are generated by the analytic hierarchy process (AHP). Using the random forest regression model, the initial weights are corrected with the feature discrimination of each indicator in the preprocessed data as input. The weight of the indicator with higher feature discrimination is automatically increased. When the real-time data feature of an indicator deviates from the initial expectation by more than a threshold or new data types are added to supplement the new features, the agent makes fine adjustments to the indicator weights or adds temporary indicators, generating a structured dynamic indicator system. This system includes a hierarchical relationship table of dimension-indicator-sub-indicator-weight-corresponding data feature, and is then transmitted to the comprehensive analysis and processing module.

[0022] The comprehensive analysis and processing module is used to acquire a dynamic indicator system and track students' learning trajectories through time-series data fusion modeling and multimodal intelligent analysis. Based on preprocessed multi-source time-series data, which includes all dimensions of data (specifically location coordinate sequences, device interaction logs, task status change records, and physiological state time-series curves), a three-dimensional trajectory model of space-time-behavior is constructed. The three-dimensional model uses the time axis as the core driving axis, and the spatial and behavioral dimensions are dynamically mapped over time. The specific structure is as follows: Timeline: The timeline is divided into fine-grained segments of 10 seconds each, recording the start / end time of each time segment and the corresponding study tour segment; Spatial dimension layer: On each time slice, the location coordinate clustering results and scene semantic labels are overlaid. The spatial distribution density is visualized through heat map to identify high-frequency activity areas, such as a student spending more than 60% of their time in the instrument area.

[0023] Behavior dimension layer: Based on multi-source temporal features, behavior types are labeled on each time slice, and automatic labeling is achieved through behavior-feature mapping rules.

[0024] By continuously splicing time slices, a dynamic trajectory is formed that shows the spatial location migrating over time and the behavior type changing with space / time. For example, from 09:10 to 09:15, the behavior is to look up information in the data area; from 09:15 to 09:30, the behavior changes to experimental operation at the workbench. Short-term abnormal fluctuations are corrected by the sliding window mean algorithm. Hidden Markov Models are used to decode the state of continuous behavior sequences, identify students' behavioral patterns in study tour scenarios, and calculate the duration and transition frequency of each pattern. The specific processing method is as follows: The design includes a hidden Markov model comprising a hidden state set S, an observation sequence O, state transition probabilities A, and observation probabilities B. The hidden state set S specifically represents typical behavioral patterns defined in the research and study scenario, such as independent exploration, collaborative discussion, task-solving, and observation / wait-and-see. The observation sequence O represents the spatial-behavioral-physiological fusion features of each time slice in the 3D trajectory model as the observed values. The state transition probability A represents the transition probability from pattern Si to Sj, and the observation probability B represents the probability that the hidden state Si generates the observed value Ok. Then, the continuous time slice observation sequence O = [O1, O2, ..., Ok] is input. T Let T be the total number of slices, and let A be the transition probability matrix and B be the observation probability matrix after training. The Viterbi algorithm is used to solve for the optimal hidden state sequence S*=[S1, S2, ..., S*]. T ], calculate the arrival state S in each time slice t i Furthermore, the maximum probability δt(i) of the observation is Ot, and the path pointer ψt(i), i.e. the behavioral pattern sequence with the maximum probability P(S|O), is recorded, and the behavioral pattern flow trajectory of the student is output throughout the entire study tour process; Acquire the flow trajectory of behavioral patterns and calculate quantitative indicators through preset rules, including the duration percentage of a certain behavioral pattern S. i Total duration = number of consecutive time slices of this pattern × duration of a single slice, percentage = (S i (Total duration / Total study tour duration) × 100%, accurate to one decimal place; Mode transition frequency: Count the number of transitions between any two different modes (Si→Sj, i≠j), and the frequency = number of transitions / total number of transitions × 100%; Conversion path characteristics: Extract high-frequency conversion paths (frequency ≥ 10%) and record their average interval duration to form a path preference map; Simultaneously, abnormal pattern recognition is performed based on preset thresholds and dynamic baselines. Regarding the judgment rules for preset thresholds, for a single abnormality, if the proportion of the problem stuck mode is ≥30%, or the hesitant and observing mode appears continuously for ≥5 slices, it is judged as an abnormality. For a conversion abnormality, if the experimental operation leads to the problem stuck conversion ≥3 times, or the high-frequency path is missing, it is judged as an abnormality. Regarding the rules for determining the dynamic baseline, based on the average behavioral patterns of students in the same batch, if a student deviates from the average by ±20 percentage points, an anomaly marker is automatically triggered, and anomaly pattern information is generated. At the same time, the anomaly pattern information is transmitted to the intelligent evaluation output module. After the anomaly pattern is marked, the system automatically generates a prompt message containing the anomaly type, the time period of occurrence, and related characteristics.

[0025] The intelligent evaluation output module is used to display the acquired abnormal pattern information to the corresponding management personnel.

[0026] The data in the above formulas are all calculated using numerical values, without substituting the units of the parameters. In addition, the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0027] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A study tour intelligent evaluation system based on intelligent agents, characterized in that, include: The multimodal data acquisition module is used to collect full-dimensional data through a multimodal acquisition terminal, and transmit the full-dimensional data to the data intelligent analysis module after preprocessing. The data intelligence analysis module is used to extract key features from the preprocessed full-dimensional data, call the knowledge graph of the study tour evaluation field and combine it with the meta-information of the study tour scenario, deconstruct the basic indicator dimensions, adjust the indicator granularity and allocate indicator weights, generate a dynamic indicator system and then transmit it to the comprehensive analysis and processing module. The comprehensive analysis and processing module is used to construct a three-dimensional trajectory model of space-time-behavior based on the dynamic indicator system and preprocessed multi-source time series data. It performs state decoding on the behavior sequence through the hidden Markov model, identifies behavior patterns and calculates quantitative indicators, identifies abnormal patterns based on preset thresholds and dynamic baselines, and transmits abnormal pattern information to the intelligent evaluation output module. The intelligent evaluation output module is used to display abnormal mode information to administrators.

2. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, The multimodal acquisition terminal includes student smart terminals, scene IoT devices, audio and video acquisition devices, and work upload ports. The full-dimensional data includes structured data and unstructured data.

3. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, Preprocessing includes: removing outliers and duplicate data, filling in missing values, converting unstructured data into an analyzable format and unifying data timestamps and spatial coordinate systems, extracting key audio and video information, labeling text with themes, and desensitizing data containing personal identifiers.

4. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, The knowledge graph in the field of study tour evaluation includes three layers of nodes: literacy dimension, evaluation indicators, and data characteristics, as well as the relationships between nodes. The generation of the dynamic indicator system includes matching indicator dimensions with the research and study theme and the students' learning stage, adjusting the indicator granularity based on the richness of data features, and allocating indicator weights through a dynamic weight model.

5. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, The dynamic weight model adopts a goal-oriented and data feedback dual-drive mechanism. It generates initial weights based on preset educational goals using the analytic hierarchy process, and then uses a random forest regression model to adjust the initial weights according to the discriminative power of the indicator features. The higher the discriminative power of the features, the higher the weight of the indicator.

6. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, The 3D trajectory model uses the time axis as its core driving axis and includes: Timeline: Divided into preset time slices, recording the start and end times of each slice and the corresponding study tour segment; Spatial dimension layer: Overlaying location coordinate clustering results with scene semantic labels to identify high-frequency activity areas; Behavior dimension layer: Behavior types are labeled based on multi-source temporal features, and automatic labeling is achieved through behavior-feature mapping rules; The three-dimensional trajectory model forms a dynamic trajectory by continuously stitching together time slices, and short-term abnormal fluctuations are corrected by a smoothing algorithm.

7. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, Hidden Markov models include a hidden state set, an observation sequence, state transition probabilities, and observation probabilities; The hidden state set represents typical behavioral patterns in the study tour scenario, and the observation sequence represents the spatial-behavioral-physiological fusion features of each time slice in the three-dimensional trajectory model. The Viterbi algorithm is used to solve for the optimal hidden state sequence and output the behavioral pattern flow trajectory.

8. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, The method for identifying behavioral patterns and calculating quantitative indicators is as follows: Acquire the flow trajectory of behavioral patterns and calculate quantitative indicators through preset rules, including the duration percentage of a certain behavioral pattern S. i Total duration = number of consecutive time slices of this pattern × duration of a single slice, percentage = (S i (Total duration / Total study tour duration) × 100%; Mode conversion frequency: Count the number of times any two different modes are converted. Frequency = Number of conversions / Total number of conversions × 100%; Conversion path characteristics: High-frequency conversion paths with a frequency of ≥10% are extracted and their average interval duration is recorded to form a path preference map.

9. The intelligent evaluation system for study tours based on intelligent agents according to claim 1, characterized in that, Anomaly pattern recognition includes a two-step process: Based on preset thresholds, single-pattern anomalies and transition anomalies are identified. Combined with a dynamic baseline formed by the average behavior patterns of students in the same batch, anomalies that deviate from the preset range of the average are identified, anomaly marking is automatically triggered, and anomaly pattern information is generated.

10. The intelligent evaluation system for study tours based on intelligent agents according to claim 9, characterized in that, For single anomalies, if the percentage of stuck issues is ≥30%, or if the hovering and observing pattern appears consecutively for ≥5 slices, it is judged as an anomaly. For conversion anomalies, if the experimental operation results in a stuck issue conversion ≥3 times consecutively, or if a high-frequency path is missing, it is judged as an anomaly.

Images