Mutual evaluation management system and method among students in school based on campus points

Abstract

The invention discloses a system and a method for mutual evaluation management among students in school based on campus points. The system comprises an identity recognition module which is used for recognizing the identity of a user; the mutual evaluation coupon generation module is used for generating mutual evaluation coupon data according to the demand information; the mutual evaluation coupon point acquisition module is used for acquiring information in the mutual evaluation coupons after mutual evaluation and confirming evaluation types and point numbers; the integral storage module is used for storing integral information of each student; the point exchange module is used for selectively exchanging prizes according to the points acquired by the students; and the display module is used for displaying the points of each student in a visual mode. According to the mutual evaluation management system among the students in the school based on the campus points, a complete mutual evaluation coupon online generation mode is provided, so that the students can perform online mutual evaluation, and the problem that paper mutual evaluation coupons need to be used in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of peer review management technology, specifically to a peer review management system and a peer review management method based on campus points among on-campus students. Background Technology

[0002] In current school student management, paper-based peer evaluation forms are still used for mutual evaluation. This method is wasteful of paper and inconvenient to manage. Summary of the Invention

[0003] The purpose of this invention is to provide a peer review management system for on-campus students based on campus points to at least solve one of the above-mentioned technical problems.

[0004] One aspect of the present invention provides a peer review management system for on-campus students based on campus points, the peer review management system for on-campus students based on campus points comprising: The identity recognition module is used to acquire user image information and identify the user's identity based on the user image information; The mutual evaluation coupon generation module is used to obtain the user's input demand information after being identified by the identity recognition module, and generate mutual evaluation coupon data based on the demand information.

[0005] Optionally, the identity recognition module includes: An image acquisition module, wherein the image acquisition module is used to acquire a face image; An image recognition module is used to recognize the facial image, thereby identifying the user's identity.

[0006] Optionally, the image recognition module includes: A face region detection module is used to locate and extract face regions from a face image using a face detection algorithm; The preprocessing module is used to preprocess the face region to obtain a preprocessed image; The feature extraction module is used to extract image features from the preprocessed image; The similarity calculation module is used to perform similarity calculation between the image features and pre-stored face images, thereby obtaining the user identity information corresponding to the pre-stored face image with the highest similarity and exceeding a preset threshold.

[0007] Optionally, the face region detection module includes: A backbone network is used to obtain scale feature maps at various scales from a face image. The scale feature maps at various scales include a first scale feature map, a second scale feature map, a third scale feature map, and a fourth scale feature map. A scale-aware feature fusion module is used to upsample the fourth-scale feature map to make its scale consistent with the third-scale feature map, thereby obtaining the first pyramid feature; and to upsample the first pyramid feature to make its scale consistent with the second-scale feature map. Figure 1 The second pyramid feature is obtained; the second pyramid feature is then upsampled to match the scale of the first pyramid feature. Figure 1 Thus, the characteristics of the third pyramid were obtained; A scale-aware attention map generation module is used to generate a first scale-aware attention map based on the first pyramid features, a second scale-aware attention map based on the second pyramid features, and a third scale-aware attention map based on the third pyramid features. The feature weighting fusion module is used to generate a first weighted feature map based on the first pyramid features and the first scale perceptual attention map; generate a second weighted feature map based on the second pyramid features and the second scale perceptual attention map; and generate a third weighted feature map based on the third pyramid features and the third scale perceptual attention map. The feature fusion module is used to fuse the first weighted feature map, the second weighted feature map, and the third weighted feature map to obtain fused features; A dynamic scale-aware attention mechanism module is used to process the fused features through three dimensions: space, channel, and scale, thereby obtaining an enhanced feature map. The face detection and alignment module is used to perform region detection on the enhanced feature map, thereby obtaining the position, size and key point information of the face region.

[0008] Optionally, the scale-aware attention map generation module includes: A spatial attention module is used to perform spatial dimension attention weighting on the fused features to generate a spatial attention map; The channel attention module is used to perform channel-dimensional attention weighting on the fused features to generate a channel attention map; A scale attention module is used to perform scale-dimensional attention weighting on the fused features to generate a scale attention map. An attention fusion module is used to fuse spatial attention maps, channel attention maps, and scale attention maps to obtain a dynamic scale-aware attention map. The feature enhancement module is used to perform element-wise multiplication of the dynamic scale-aware attention map and the fused features to obtain the enhanced feature map.

[0009] Optionally, the face detection and alignment module includes: The preliminary face candidate box acquisition module is used to acquire a set of face candidate boxes, the face probability corresponding to each candidate box, and the bounding box regression vector based on the enhanced feature map. The secondary face candidate box acquisition module is used to perform preliminary screening of candidate boxes based on the face probability corresponding to each candidate box, and retain candidate boxes with higher probabilities, thereby obtaining a set of screened face candidate boxes, the face probability corresponding to each screened face candidate box, and the bounding box regression vector. The final face region acquisition module is used to obtain the final face region set, the face probability, the bounding box regression vector, and the coordinates of the face key points for each region in the final face region set based on the filtered face candidate box set, the face probability corresponding to each filtered face candidate box, and the bounding box regression vector.

[0010] Optionally, the preliminary face candidate box acquisition module includes: A dynamically enhanced convolution module is provided, which performs a dynamically enhanced convolution operation on the enhanced feature map; wherein the dynamically enhanced convolution operation is performed using the following formula: ;in, The intermediate result of the convolution operation is called the preliminary feature map; This is a dynamic enhancement factor, ranging from (0,1); The Obtain it using the following formula: ;in, This is the convolution kernel weight matrix, used to extract local features from the enhanced feature map; The enhanced feature map; The Obtain it using the following formula: ; in, This is the Sigmoid activation function, used to restrict the dynamic enhancement factor to the range (0,1); It is a learnable weight matrix used to map the statistical information of the input feature map to the enhancement factor; For a learnable bias vector, and Together they form an affine transformation, used to adjust the baseline value of the enhancement factor; The height of the input feature map; The width of the input feature map; The number of channels in the input feature map; The value of the enhanced feature map at position (i,j,k); The index representing the height of the input feature map is used to traverse every position of the feature map in the vertical direction; The index representing the width of the input feature map is used to traverse every position of the feature map in the horizontal direction; The index represents the number of channels in the input feature map, used to traverse each channel of the feature map, ensuring that information from all channels is considered.

[0011] Optionally, the mutual rating coupon generation module includes: The requirement parsing module is used to obtain the requirement information input by the user and obtain a structured requirement vector based on the requirement information. The NLP model acquisition module is used to acquire trained NLP models. A question generation module is used to input the structured demand vector into the NLP model to generate a high-quality question list; The peer review voucher generation module is used to generate peer review voucher data based on the question list.

[0012] Optionally, the total loss function of the trained NLP model is as follows: ;in, Based on the generation of loss; Domain adaptation loss; Loss due to difficulty adaptation; Loss due to increased diversity; Loss due to increased fun; To adapt to cognitive development loss.

[0013] This application also provides a method for managing peer review among enrolled students based on campus points, characterized in that the method includes: Acquire user image information and identify the user's identity based on the user image information; The system obtains the user's input requirements after the identity recognition module has identified them, and generates mutual evaluation coupon data based on the requirements.

[0014] Beneficial effects: The peer review management system based on campus points proposed in this application provides a complete online generation method for peer review vouchers, enabling students to conduct peer reviews online and solving the problem of existing technologies requiring the use of paper peer review vouchers. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the structure of a peer review management system for on-campus students based on campus points, according to an embodiment of this application.

[0016] Figure 2 This is a schematic diagram of the peer review record in one embodiment of this application.

[0017] Figure 3 This is a schematic diagram of the leaderboard metadata in one embodiment of this application.

[0018] Figure 4 This is a schematic diagram of a leaderboard content list in one embodiment of this application.

[0019] Figure 5 This is a schematic diagram of the personal data structure relationship in one embodiment of this application.

[0020] Figure 6 This is a schematic diagram of a behavior state node in one embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0022] like Figure 1 The peer review management system for on-campus students based on campus points, as shown, includes: The identity recognition module is used to acquire user image information and identify the user's identity based on the user image information; The mutual evaluation coupon generation module is used to obtain the user's input demand information after being identified by the identity recognition module, and generate mutual evaluation coupon data based on the demand information.

[0023] In this embodiment, the application may further include the following modules: Evaluation Distribution and Points Recording Module: Students can give away peer evaluation coupons to other students during the evaluation distribution process. Each student being evaluated can scan the QR code on the received peer evaluation coupon to access the system and redeem points. Each student's points can be recorded under two indicators: "Gifts" and "Received".

[0024] The peer review system displays each student's individual points, including current "gift" and "receive" points. Students can view their evaluation contributions for the current period and understand their current point ranking. Furthermore, the system displays the "receive" and "gift" point rankings across the entire school, increasing interactivity and transparency.

[0025] 5. Points Calculation and Ranking Module: Points are accumulated, and the redemption of each peer review voucher is based on a pre-set points value. For example, a "gratitude" type peer review voucher may earn fewer points, while a "praise" type may earn more points. The system dynamically calculates each student's total points for a semester or a specific period based on their evaluation activities, and displays the school's points ranking according to the total points.

[0026] Data storage and security module: All student evaluation data and points records are stored in the backend database. The data is protected by encryption technology to prevent malicious modification. The system backs up data regularly to ensure data integrity and reliability.

[0027] The specific usage process is as follows: Students log in to the peer review management system based on campus points using facial recognition. In the peer review management system interface based on campus points, students can select a preset template to obtain peer review coupons, choose one of four evaluation types, and determine the number of coupons to print. For example, in this embodiment, students select one or more evaluation types on the peer review device: gratitude, appreciation, encouragement, and praise.

[0028] Understandably, one can also design their own template for peer review coupons and the specific questions they want to ask.

[0029] Once a peer review coupon is generated, it can be sent directly to other students, who can then provide feedback.

[0030] The students being evaluated use the peer review machine to scan the peer review coupon QR code to complete the points redemption.

[0031] Students can view their current points and their ranking in terms of points given and received.

[0032] Regularly review the school-wide rankings to encourage positive feedback and mutual assistance among students.

[0033] Incentive mechanism: The points system encourages students to support, encourage, and recognize each other, which helps to enhance the campus culture.

[0034] Highly operable: Students can complete peer review simply by scanning a QR code, making the process concise and easy to use.

[0035] High transparency: The points system is open and transparent, and students can check their points and the school's overall evaluation at any time.

[0036] Data security: Facial recognition and encryption technologies are used to ensure data security and prevent malicious behavior and data tampering.

[0037] Students can distribute the generated peer review coupons to other students. Each peer review coupon can be given to one or more students as a form of evaluation.

[0038] It is understandable that the peer review vouchers in this application can also be distributed in paper form by printing.

[0039] After receiving the peer review voucher, the student being evaluated scans it using the peer review machine. After scanning, the system reads the information in the voucher, confirms the evaluation type and points, and completes the redemption and recording of points according to the pre-set rules.

[0040] Points Recording and Display: The system records each student's "gifted" and "received" points. Students can view their points on the peer review platform, including current and historical points. The system also displays school-wide point rankings for students to check, encouraging healthy competition.

[0041] Periodic settlement and data statistics: The system settles points and generates point statistics reports according to a preset period (such as monthly or per semester). Point data can be exported for student behavior analysis and reward system development.

[0042] In this embodiment, the identity recognition module includes: An image acquisition module, wherein the image acquisition module is used to acquire a face image; An image recognition module is used to recognize the facial image, thereby identifying the user's identity.

[0043] In this embodiment, the image recognition module includes: A face region detection module is used to locate and extract face regions from a face image using a face detection algorithm; The preprocessing module is used to preprocess the face region to obtain a preprocessed image; The feature extraction module is used to extract image features from the preprocessed image; The similarity calculation module is used to perform similarity calculation between the image features and pre-stored face images, thereby obtaining the user identity information corresponding to the pre-stored face image with the highest similarity and exceeding a preset threshold.

[0044] In this embodiment, the face region detection module includes: A backbone network is used to obtain scale feature maps at various scales from a face image. These scale feature maps include a first scale feature map, a second scale feature map, a third scale feature map, and a fourth scale feature map. In this embodiment, a deep convolutional neural network (such as ResNet, EfficientNet, etc.) is used as the backbone network to extract features from the input image, resulting in multi-scale feature maps: first scale feature map C2, second scale feature map C3, third scale feature map C4, and fourth scale feature map C5 (the subscript indicates the scale of the feature map relative to the input image; for example, the scale of C5 is 1 / 32 of the input image).

[0045] A scale-aware feature fusion module is used to upsample the fourth-scale feature map to make its scale consistent with the third-scale feature map, thereby obtaining the first pyramid feature; and to upsample the first pyramid feature to make its scale consistent with the second-scale feature map. Figure 1 The second pyramid feature is obtained; the second pyramid feature is then upsampled to match the scale of the first pyramid feature. Figure 1 The process is repeated to obtain the third pyramid feature. Specifically, C5 is upsampled to match the scale of C4, resulting in the first pyramid feature P5. P5 is then upsampled to match the scale of C3, resulting in the second pyramid feature P4. Finally, P4 is upsampled to match the scale of C2, resulting in the third pyramid feature P3. The final feature pyramids P3, P4, and P5 correspond to different scales.

[0046] A scale-aware attention map generation module is used to generate a first scale-aware attention map based on the features of the first pyramid, a second scale-aware attention map based on the features of the second pyramid, and a third scale-aware attention map based on the features of the third pyramid. Specifically, a lightweight scale-aware attention generator (SAG) is used. The SAG takes feature pyramids P3, P4, and P5 as input and generates a scale-aware attention map (first scale-aware attention map A3, second scale-aware attention map A4, and third scale-aware attention map A5) for each scale of feature map through convolutional layers with shared weights and global average pooling layers. Each element value in the graph represents the corresponding scale feature map. The importance of this positional feature.

[0047] The feature weighting fusion module is used to generate a first weighted feature map based on the first pyramid features and the first scale perceptual attention map; generate a second weighted feature map based on the second pyramid features and the second scale perceptual attention map; and generate a third weighted feature map based on the third pyramid features and the third scale perceptual attention map. The feature fusion module is used to fuse the first weighted feature map, the second weighted feature map, and the third weighted feature map to obtain fused features; A dynamic scale-aware attention mechanism module is used to process the fused features through three dimensions: space, channel, and scale, thereby obtaining an enhanced feature map. The face detection and alignment module is used to perform region detection on the enhanced feature map, thereby obtaining the position, size and key point information of the face region.

[0048] In this embodiment, the scale-aware attention map generation module includes: A spatial attention module is used to perform spatial dimension attention weighting on the fused features to generate a spatial attention map; The channel attention module is used to perform channel-dimensional attention weighting on the fused features to generate a channel attention map; A scale attention module is used to perform scale-dimensional attention weighting on the fused features to generate a scale attention map. An attention fusion module is used to fuse spatial attention maps, channel attention maps, and scale attention maps to obtain a dynamic scale-aware attention map. The feature enhancement module is used to perform element-wise multiplication of the dynamic scale-aware attention map and the fused features to obtain the enhanced feature map. .

[0049] In this embodiment, the face detection and alignment module includes: The preliminary face candidate box acquisition module is used to acquire a set of face candidate boxes, the face probability corresponding to each candidate box, and the bounding box regression vector based on the enhanced feature map. The secondary face candidate box acquisition module is used to perform preliminary screening of candidate boxes based on the face probability corresponding to each candidate box, and retain candidate boxes with higher probabilities, thereby obtaining a set of screened face candidate boxes, the face probability corresponding to each screened face candidate box, and the bounding box regression vector. The final face region acquisition module is used to obtain the final face region set, the face probability, the bounding box regression vector, and the coordinates of the face key points for each region in the final face region set based on the filtered face candidate box set, the face probability corresponding to each filtered face candidate box, and the bounding box regression vector.

[0050] In this embodiment, the preliminary face candidate box acquisition module (dynamically enhanced P-Net) includes: A dynamically enhanced convolution module is provided, which performs a dynamically enhanced convolution operation on the enhanced feature map; wherein the dynamically enhanced convolution operation is performed using the following formula: ;in, The intermediate result of the convolution operation is called the preliminary feature map; This is a dynamic enhancement factor, ranging from (0,1); The Obtain it using the following formula: ;in, This is the convolution kernel weight matrix, used to extract local features from the enhanced feature map; The enhanced feature map; The Obtain it using the following formula: ; in, This is the Sigmoid activation function, used to restrict the dynamic enhancement factor to the range (0,1); It is a learnable weight matrix used to map the statistical information of the input feature map to the enhancement factor; For a learnable bias vector, and Together they form an affine transformation, used to adjust the baseline value of the enhancement factor; The height of the input feature map; The width of the input feature map; The number of channels in the input feature map; The value of the enhanced feature map at position (i,j,k); The index representing the height of the input feature map is used to traverse every position of the feature map in the vertical direction; The index representing the width of the input feature map is used to traverse every position of the feature map in the horizontal direction; The index represents the number of channels in the input feature map, used to traverse each channel of the feature map, ensuring that information from all channels is considered.

[0051] In this embodiment, the preliminary face candidate box acquisition module further includes: The face classification and bounding box regression module is used to dynamically enhance the feature map after convolution. Face classification and bounding box regression are performed in parallel. Specifically, face classification is performed by using a convolutional layer (with 2 output channels) and a softmax activation function to generate a probability map. , which represents the probability that each position on the input feature map is a face.

[0052] Bounding box regression generates a bounding box regression vector map through a convolutional layer (with 4 output channels). , which represents the coordinate offset of the face candidate box at each location on the input feature map relative to the default anchor box.

[0053] In this embodiment, the preliminary face candidate box acquisition module further includes: A candidate box generation module, which is used to generate candidate boxes based on the probability map of face classification. and bounding box regression vector map Generate a series of face candidate boxes Specifically, for locations on the probability map where the probability value is greater than a preset threshold, the position and size of the default anchor box are adjusted based on the corresponding bounding box regression vector to generate face candidate boxes.

[0054] In this embodiment, the secondary face candidate box acquisition module includes: A candidate box filtering module, which is used to filter candidates based on a probability map of face classification. Perform initial screening on the candidate boxes, and retain the candidate boxes with higher probability (e.g., the first two).

[0055] The retained candidate boxes are mapped back to the original image, and the corresponding image patches are cropped. Feature extraction is performed on the cropped image patches, and the Dynamic Enhancement Convolution (DBConv) operation is also used to obtain the input to the Dynamic Enhancement R-Net, thereby obtaining the set of filtered face candidate boxes, the face probability corresponding to each filtered face candidate box, and the bounding box regression vector.

[0056] In this embodiment, the final face region acquisition module (dynamically enhanced O-Net) includes: Based on the bounding box regression vectors output by R-Net, the candidate boxes are adjusted to obtain more accurate face candidate boxes.

[0057] The adjusted candidate boxes are mapped back to the original image, the corresponding image patches are cropped, and adjusted to the input size of O-Net.

[0058] Feature extraction is performed on the cropped image patches using the same dynamically enhanced convolution (DBConv) operation.

[0059] Face classification, bounding box regression, and key point localization are performed in parallel on the extracted feature maps.

[0060] Face classification and bounding box regression: The computation method is similar to P-Net and R-Net, but uses O-Net-specific network parameters.

[0061] A keypoint coordinate map is generated by a convolutional layer (with 2×N output channels, where N represents the number of facial keypoints), and the coordinates of N facial keypoints are represented (each keypoint contains two coordinates, x and y).

[0062] The Non-Maximum Suppression (NMS) algorithm is used to perform final filtering on the face classification probabilities and bounding box regression results, removing redundant candidate boxes. Post-processing (such as coordinate transformation and filtering) is then applied to the keypoint coordinates to obtain more accurate keypoint locations.

[0063] In this embodiment, by improving the convolutional operations of P-Net, R-Net, and O-Net to dynamically enhanced convolutions and introducing a dynamic enhancement factor into the loss function, DBC-Net can significantly improve the accuracy and robustness of face detection, especially in complex scenes. This improvement complements the scale-aware attention mechanism in the SA-APN framework, together forming an efficient and accurate face detection system.

[0064] In this embodiment, the preprocessing module includes: aligning the detected faces (e.g., using affine transformation to align key points such as eyes, nose and mouth to predetermined positions) and normalizing (e.g., adjusting image size, grayscale, histogram equalization, etc.) to eliminate the influence of changes in lighting, pose and expression on the recognition results.

[0065] In this embodiment, the similarity calculation module can use cosine similarity, Euclidean distance, or other similarity metrics to calculate the similarity between two facial feature vectors. A similarity threshold is set according to the needs of the actual application scenario. When the similarity between two faces exceeds this threshold, they are considered to belong to the same person; otherwise, they are considered to belong to different people.

[0066] In this embodiment, Scale-Aware Feature Fusion (SAF) can adaptively learn the importance of features at different scales and achieve more effective multi-scale feature fusion through feature weighted fusion, thereby improving the network's ability to detect faces at multiple scales.

[0067] The Dynamic Scale Aware Attention (DSA) mechanism innovatively considers three dimensions simultaneously: space, channel, and scale. It dynamically learns the importance of facial features at different scales and adaptively adjusts the network's attention to features at different scales, significantly improving the network's anti-interference ability and face detection accuracy in complex backgrounds.

[0068] Adaptive Feature Fusion (AFF) learns fusion weights adaptively based on the statistical information of the input feature map, enabling more flexible multi-scale feature fusion and further improving the richness and discriminativeness of feature representation.

[0069] Using the above image recognition method for student identity authentication can more accurately locate and extract the student's face region from the image captured by the camera. Even under complex lighting conditions, background interference, large changes in face size, and partial occlusion, it can provide a more reliable foundation for subsequent face recognition and identity authentication, significantly improving the robustness, accuracy, and user experience of the system.

[0070] In this embodiment, the mutual rating coupon generation module includes: The requirement parsing module is used to acquire the requirement information input by the user and obtain a structured requirement vector based on the requirement information. Specifically, it cleans the text requirement provided by the user (such as "generate peer review questions about a certain subject") by removing irrelevant symbols and stop words. Using the encoder part of a pre-trained language model (such as BERT or RoBERTa), the requirement text is converted into a high-dimensional semantic vector. Requirement keywords are automatically expanded through word list matching or semantic similarity calculation.

[0071] The NLP model acquisition module is used to acquire trained NLP models (such as BERT, RoBERTa, T5, Flan-T5, GPT-2, Llama, Falcon, etc.). A question generation module is used to input the structured demand vector into the NLP model to generate a high-quality question list; The peer review voucher generation module is used to generate peer review voucher data based on the question list.

[0072] In this embodiment, the total loss function of the trained NLP model is as follows: ;in, Based on the generation of loss; Domain adaptation loss; Loss due to difficulty adaptation; Loss due to increased diversity; Loss due to increased fun; To adapt to cognitive development loss.

[0073] In this embodiment, Obtain it using the following formula: ;in, The basic generation loss measures the fundamental quality of the sequence generated by the model. T is the sequence length, i.e., the total number of characters or words used to generate the question. t is the current time step, representing the t-th element in the sequence being generated. The actual word or character generated at time step t. It is a sequence of all words or characters generated before time step t. The input sequence is the original requirement text provided by the user. To in a given input sequence and the previously generated sequence Under the condition of generating words The probability of.

[0074] In this embodiment, Obtain it using the following formula: ;in, The domain-adaptive loss ensures that the generated questions cover the preset evaluation domains. D is the preset domain set, such as [learning habits], [classroom participation], [collaborative behavior], etc. d is a specific domain within set D. The dynamic weight of the domain d is adjusted based on historical generation records. Let d be a uniform distribution of the domain, indicating that ideally each domain should have balanced coverage. The actual distribution of the domain d in the generation problem is calculated using a domain classifier. This is the Kullback-Leibler divergence, which measures the difference between the actual distribution Pd and the ideal distribution Ud.

[0075] In this embodiment, Obtain it using the following formula: ;in, A difficulty-fit loss is used to ensure that the difficulty of the generated questions matches the target grade level. α and β are learnable weight parameters that are dynamically adjusted through meta-learning to balance the impact of the two losses. MSE is the mean squared error, which measures the difference between the Lexile text difficulty value and the target grade level difficulty value. This is the Lexile text difficulty value, calculated using the Lexile analysis framework. Set the difficulty value corresponding to the target grade level, such as 3 for third grade. The Jensen-Shannon divergence measures the difference between the distribution of syntactic complexity and the typical syntactic distribution for the target grade level. The syntactic complexity distribution is obtained based on dependency parsing. The typical syntactic distribution for the target grade level is obtained by statistically analyzing text data from the target grade level.

[0076] In this embodiment, Obtain it using the following formula: ;in, To enhance diversity, the generated questions are encouraged to maintain semantic and syntactic diversity. λ is a dynamic balancing coefficient that is automatically adjusted based on the generation batch, controlling the relative importance of semantic and syntactic diversity. To address semantic diversity loss, cosine similarity is calculated based on sentence vectors encoded by BERT to measure the semantic similarity between generated questions. To address the loss of syntactic diversity, we measure the syntactic similarity between generation questions based on the edit distance of the dependency tree.

[0077] In this embodiment, Obtain it using the following formula: ;in, Increased interest leads to a loss of engagement, thereby enhancing the attractiveness of the generated questions. These are learnable weight parameters that are dynamically adjusted through reinforcement learning to balance the impact of various interesting metrics. Apply a rationality loss to emojis, calculated based on a pre-trained emoji classifier, to measure whether the use of emojis in the problem is appropriate. To mitigate the loss of interrogative word diversity, the use of a variety of open-ended interrogative words, such as "how" and "why," is encouraged. The loss of story context integration measures whether the questions incorporate story contexts that appeal to children.

[0078] In this embodiment, Obtain it using the following formula: ;in, To adapt to cognitive developmental loss, the generated questions are designed to align with the cognitive developmental stage of elementary school students. C represents the set of cognitive features, such as [concrete thinking], [conservation concepts], and [classification ability]. c represents a specific cognitive feature within set C. The dynamic feature weights are automatically adjusted based on the student's age to control the importance of each cognitive feature. BCE is the binary cross-entropy loss, which measures the difference between the detection probability of cognitive feature c in the generated question and the target threshold. The detection probability of cognitive feature c in the problem is calculated using a cognitive feature detector. The threshold for target cognitive features, such as concrete thinking features, should be greater than 0.8.

[0079] This application achieves a comprehensive quality improvement across the entire generation process by jointly optimizing six loss functions: basic generation, domain adaptation, difficulty adaptation, diversity enhancement, fun enhancement, and cognitive development adaptation.

[0080] We construct hierarchical optimization goals from grammatical correctness (basic level) to educational value (cognitive level) to avoid suboptimal solutions caused by single-dimensional optimization.

[0081] Ensure that the generated questions not only conform to language norms, but also possess domain representativeness, appropriate difficulty, diverse formats, and cognitive inspiration.

[0082] This application also provides a method for managing peer review among enrolled students based on campus points, characterized in that the method includes: Acquire user image information and identify the user's identity based on the user image information; The system obtains the user's input requirements after the identity recognition module has identified them, and generates mutual evaluation coupon data based on the requirements.

[0083] The following example further illustrates the peer review management system for on-campus students based on campus points, and it is understood that this example does not constitute any limitation on this application.

[0084] The peer review management system of this invention mainly consists of intelligent peer review devices, a backend server, and a database. The system is functionally divided into several sub-modules, including a user authentication module (facial recognition), a peer review coupon distribution module, an evaluation and points management module, a leaderboard calculation and display module, and a data storage and security module. The intelligent peer review devices communicate with the backend data center via wired or wireless networks, and all evaluation behaviors and points records are synchronized to the backend database in real time.

[0085] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A peer review management system and method based on campus points among on-campus students, comprising the following steps: This solution constructs a three-tiered structure consisting of "mutual evaluation behavior rule set + mutual evaluation change + mutual evaluation behavior record model" to realize the technical management system architecture and workflow configuration for the entire mutual evaluation process. Students: can provide feedback, assign feedback and a number of feedbacks to other students, and receive feedback from other students.

[0086] Teachers / Class Teachers: Responsible for supervising the evaluation process and handling any abnormal situations during the evaluation.

[0087] Administrator: Responsible for system management, points rule settings, data statistics, etc.

[0088] Furthermore, the system configures project information, including: Define the system evaluation project type, and set the project evaluation type as Evaluation=(E1,E2,E3,……,En), where E1,E2,E3,……,En represent the project evaluation type category respectively; Define student information labels, let Student = (S1, S2, S3, ..., Sn), where S1, S2, S3, ..., Sn represent a specific student; Define the number of system evaluation items, setting the number of evaluation items to 1, 2, 3, ..., N; Define a leaderboard label, set the leaderboard label to Ranking=(R1,R2,R3,...Rn), where R1,R2,R3,...Rn represent the corresponding rankings; Furthermore, the student peer review system uses facial recognition algorithms for access, including: An edge recognition model component is deployed on the peer-evaluation device terminal. This component integrates face capture, feature extraction, and vector comparison functions, and its structure is as follows: Image acquisition submodule: calls the device's camera to acquire images in real time; Feature vector calculation module: calls the local model and uses a deep neural network to align facial features and extract 128-dimensional vectors from the image; Feature matching engine: It calls the embedded database to compare vector information and uses the cosine similarity algorithm to complete identity matching; Authentication result output module: After a successful match, an access confirmation command is generated and a login log is recorded.

[0089] Data security and access control All facial feature vectors are encrypted and stored using an irreversible encryption algorithm; The system sets an identification threshold (default ≥0.85) and liveness detection to avoid false identification or the use of photos or other materials for identification and verification; Peer review records are modeled in the form of structured behavioral logs, with fields such as Figure 2 As shown; The system dynamically loads the corresponding coupon template based on the peer review type selected by the student (such as gratitude, appreciation, encouragement, praise). Furthermore, the printing of points coupons adopts an asynchronous queue scheduling mechanism. When the student selects the coupon type and quantity, the system checks the currently available points, freezes the points to be printed, and determines whether the current student's points to be printed have reached the available points threshold. If they have, printing can proceed; otherwise, an insufficient points result prompt is returned. Furthermore, after the points vouchers are used, an ordered set structure is used to store the student ID and their score (or other ranking factors), and to update the leaderboard data and the data on individual acquisition and gifting. The leaderboard metadata area is as follows Figure 3 As shown.

[0090] Furthermore, the leaderboard list, where each item represents a student's ranking performance in that dimension, is as follows: Figure 4 As shown. The personal data structure relationship is as follows. Figure 5 As shown.

[0091] The leaderboard data is cached and stored using "dimension + time + class" as the key-value rule, and is read in the system through logical binding and pagination mechanisms using a structured list or ordered set approach.

[0092] Furthermore, in each peer review activity, the system constructs a finite-state process for each student's peer review behavior, defining the behavior state nodes as follows: Figure 6 As shown.

[0093] The student peer assessment status transition path is described as "State A - Condition / Action - State B - Condition / Action - State C".

[0094] The description refers to the normal transition logic between states, where: S0-S1: Students log into the system and access the peer review interface, triggering a state transition. S1-S2: Select the evaluation dimensions and quantity, print the evaluation coupons, and trigger the state transition; S2-S3: Give the person being evaluated an evaluation voucher (to express gratitude / appreciation / encouragement / praise, etc.), triggering a state change; S3-S4, the person being evaluated scans the points voucher on the mutual evaluation machine, triggering a state transition; The described transition logic between states includes: S1-S2: If the cumulative number of prints exceeds the limit set for the current semester, a print failure will be returned; if the total number of prints exceeds the total number of prints for the current student, a print failure will be returned, triggering a state transition. S3-S4: If the mutual evaluation coupon has been used, return "points failed" and trigger a state transition.

[0095] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A peer review management system for on-campus students based on campus points, characterized in that, The peer review management system for on-campus students based on campus points includes: The identity recognition module is used to acquire user image information and identify the user's identity based on the user image information; The mutual evaluation coupon generation module is used to obtain the user's input demand information after being identified by the identity recognition module, and generate mutual evaluation coupon data based on the demand information. The mutual evaluation voucher points acquisition module is used to acquire information from the mutual evaluation vouchers after mutual evaluation, and to confirm the evaluation type and points. An points storage module, which is used to store the points information of each student separately; The points redemption module is used to selectively redeem prizes based on the points earned by students. The display module is used to visually display the scores of each student.

2. The peer review management system for on-campus students based on campus points as described in claim 1, characterized in that, The identity recognition module includes: An image acquisition module, wherein the image acquisition module is used to acquire a face image; An image recognition module is used to recognize the facial image, thereby identifying the user's identity.

3. The peer review management system for on-campus students based on campus points as described in claim 2, characterized in that, The image recognition module includes: A face region detection module is used to locate and extract face regions from a face image using a face detection algorithm; The preprocessing module is used to preprocess the face region to obtain a preprocessed image; The feature extraction module is used to extract image features from the preprocessed image; The similarity calculation module is used to perform similarity calculation between the image features and pre-stored face images, thereby obtaining the user identity information corresponding to the pre-stored face image with the highest similarity and exceeding a preset threshold.

4. The peer review management system for on-campus students based on campus points as described in claim 3, characterized in that, The face region detection module includes: A backbone network is used to obtain scale feature maps at various scales from a face image. The scale feature maps at various scales include a first scale feature map, a second scale feature map, a third scale feature map, and a fourth scale feature map. The scale-aware feature fusion module is used to upsample the fourth-scale feature map to make its scale consistent with the third-scale feature map to obtain the first pyramid feature; upsample the first pyramid feature to make its scale consistent with the second-scale feature map to obtain the second pyramid feature; and upsample the second pyramid feature to make its scale consistent with the first-scale feature map to obtain the third pyramid feature. A scale-aware attention map generation module is used to generate a first scale-aware attention map based on the first pyramid features, a second scale-aware attention map based on the second pyramid features, and a third scale-aware attention map based on the third pyramid features. The feature weighting fusion module is used to generate a first weighted feature map based on the first pyramid features and the first scale perceptual attention map; generate a second weighted feature map based on the second pyramid features and the second scale perceptual attention map; and generate a third weighted feature map based on the third pyramid features and the third scale perceptual attention map. The feature fusion module is used to fuse the first weighted feature map, the second weighted feature map, and the third weighted feature map to obtain fused features; A dynamic scale-aware attention mechanism module is used to process the fused features through three dimensions: space, channel, and scale, thereby obtaining an enhanced feature map. The face detection and alignment module is used to perform region detection on the enhanced feature map, thereby obtaining the position, size and key point information of the face region.

5. The peer review management system for on-campus students based on campus points as described in claim 4, characterized in that, The scale-aware attention map generation module includes: A spatial attention module is used to perform spatial dimension attention weighting on the fused features to generate a spatial attention map; The channel attention module is used to perform channel-dimensional attention weighting on the fused features to generate a channel attention map; A scale attention module is used to perform scale-dimensional attention weighting on the fused features to generate a scale attention map. An attention fusion module is used to fuse spatial attention maps, channel attention maps, and scale attention maps to obtain a dynamic scale-aware attention map. The feature enhancement module is used to perform element-wise multiplication of the dynamic scale-aware attention map and the fused features to obtain the enhanced feature map.

6. The peer review management system for on-campus students based on campus points as described in claim 5, characterized in that, The face detection and alignment module includes: The preliminary face candidate box acquisition module is used to acquire a set of face candidate boxes, the face probability corresponding to each candidate box, and the bounding box regression vector based on the enhanced feature map. The secondary face candidate box acquisition module is used to perform preliminary screening of candidate boxes based on the face probability corresponding to each candidate box, and retain candidate boxes with higher probabilities, thereby obtaining a set of screened face candidate boxes, the face probability corresponding to each screened face candidate box, and the bounding box regression vector. The final face region acquisition module is used to obtain the final face region set, the face probability, the bounding box regression vector, and the coordinates of the face key points for each region in the final face region set based on the filtered face candidate box set, the face probability corresponding to each filtered face candidate box, and the bounding box regression vector.

7. The peer review management system for on-campus students based on campus points as described in claim 6, characterized in that, The preliminary face candidate box acquisition module includes: A dynamically enhanced convolution module is provided, which performs a dynamically enhanced convolution operation on the enhanced feature map; wherein the dynamically enhanced convolution operation is performed using the following formula: ;in, The intermediate result of the convolution operation is called the preliminary feature map; This is a dynamic enhancement factor, ranging from (0,1); The Obtain it using the following formula: ;in, This is the convolution kernel weight matrix, used to extract local features from the enhanced feature map; The enhanced feature map; The Obtain it using the following formula: ； in, This is the Sigmoid activation function, used to restrict the dynamic enhancement factor to the range (0,1); It is a learnable weight matrix used to map the statistical information of the input feature map to the enhancement factor; For a learnable bias vector, and Together they form an affine transformation, used to adjust the baseline value of the enhancement factor; The height of the input feature map; The width of the input feature map; The number of channels in the input feature map; The value of the enhanced feature map at position (i,j,k); The index representing the height of the input feature map is used to traverse every position of the feature map in the vertical direction; The index representing the width of the input feature map is used to traverse every position of the feature map in the horizontal direction; The index represents the number of channels in the input feature map, used to traverse each channel of the feature map, ensuring that information from all channels is considered.

8. The peer review management system for on-campus students based on campus points as described in claim 7, characterized in that, The mutual evaluation coupon generation module includes: The requirement parsing module is used to obtain the requirement information input by the user and obtain a structured requirement vector based on the requirement information. The NLP model acquisition module is used to acquire trained NLP models. A question generation module is used to input the structured demand vector into the NLP model to generate a high-quality question list; The peer review voucher generation module is used to generate peer review voucher data based on the question list.

9. The peer review management system for on-campus students based on campus points as described in claim 8, characterized in that, The total loss function of the trained NLP model is as follows: ;in, Based on the generation of loss; Domain adaptation loss; Loss due to difficulty adaptation; Loss due to increased diversity; Loss due to increased fun; To adapt to cognitive development loss.

10. A method for managing peer evaluation among on-campus students based on campus points, characterized in that, The peer review management method among on-campus students based on campus points includes: Acquire user image information and identify the user's identity based on the user image information; The system obtains the user's input requirements after the identity recognition module has identified them, and generates mutual evaluation coupon data based on the requirements.

Images