Hybrid teaching evaluation optimization system and method based on real-time communication

Abstract

The invention discloses a hybrid teaching evaluation optimization system and method based on real-time communication, relates to the technical field of intelligent teaching, and can fuse bullet screen text and student face image information in an online teaching process to construct a multi-modal emotion evaluation mechanism. A first credibility index is generated by acquiring a network delay value and performing nonlinear attenuation processing, a second credibility index is generated in combination with response time delay and historical activeness of students to questions asked by teachers, and the second credibility index is coupled with basic biological weights corresponding to course types to generate a dynamic weight coefficient. And weighted fusion of the text and the biological emotion data is realized. The method can effectively adapt to the change of a communication environment, improves the accuracy and timeliness of emotion recognition, achieves the early warning feedback based on a dual-threshold mechanism, and facilitates the improvement of the teaching interaction quality of a large-scale live classroom.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent teaching, in particular to a hybrid teaching evaluation optimization system and method based on real-time communication. BACKGROUND

[0002] With the rapid development of distance education and hybrid teaching, more and more teaching platforms begin to introduce emotional computing and student behavior analysis technology to assist teachers in real-time mastering student status. In the current large class live class scene, common student feedback forms include barrage text, expression image and expression micro-motion biological feature signals. These data are usually used to generate comprehensive emotion indicators to help teachers judge the understanding state or attention concentration of students. However, due to the heterogeneity of multi-source data and the dynamics of the teaching environment, how to evaluate the credibility of various feedback data and effectively fuse them is still a key difficulty to realize high-precision emotion perception.

[0003] Most of the existing multi-modal teaching feedback systems use fixed or experience set weight strategies to fuse text and biological signals, for example, a fixed fusion ratio is uniformly given to biological signals and text signals, which lacks a response mechanism to the dynamic changes of data quality, especially in the case of network delay, untimely response or fluctuation of student participation, it is impossible to effectively model and quantitatively process the credibility of different data channels. SUMMARY

[0004] In view of the deficiencies in the prior art, the present application provides a hybrid teaching evaluation optimization system and method based on real-time communication.

[0005] In order to achieve the above purpose, the technical scheme of the present application is as follows:

[0006] In the first aspect, the present application discloses a hybrid teaching evaluation optimization method based on real-time communication, comprising the following steps:

[0007] Obtain barrage text data of a target student, student facial image and network delay value between teacher end and student end;

[0008] Perform emotional scoring on the barrage text data to generate a text emotion value; extract biological features based on the student facial image data and calculate a biological emotion value;

[0009] Perform nonlinear attenuation processing according to the network delay value to generate a first credibility index;

[0010] Obtain the time interval from the moment the teacher asks the question to the moment the student responds to the barrage, and generate a second credibility index in combination with historical response data;

[0011] The basic biological weight is determined according to a course type, the basic biological weight, a first credibility index and a second credibility index are coupled to generate a biological weight coefficient, and a text weight coefficient is derived based on the biological weight coefficient;

[0012] The biological sentiment value and the text sentiment value are weighted and calculated according to the biological weight coefficient and the text weight coefficient to generate a real-time sentiment value;

[0013] It is judged whether the real-time sentiment value continuously falls below a preset negative sentiment threshold and whether a continuous duration exceeds a preset time threshold; if the judgment result is yes, a warning signal is triggered, and a warning instruction is sent to a teacher end.

[0014] In a second aspect, the application discloses a hybrid teaching evaluation optimization system based on real-time communication, which comprises:

[0015] A data acquisition module is configured to acquire barrage text data of a target student, student facial image data and network delay values between a teacher end and a student end;

[0016] A sentiment value calculation module is configured to perform sentiment scoring on the barrage text data to generate a text sentiment value, and extract biological features based on the student facial image data to calculate a biological sentiment value;

[0017] A credibility index calculation module is configured to perform nonlinear attenuation processing on the network delay values to generate a first credibility index, and acquire a time interval from a teacher questioning moment to a student barrage response moment, and combine historical response data to generate a second credibility index;

[0018] A weight coefficient calculation module is configured to determine a basic biological weight according to a course type, couple the basic biological weight, the first credibility index and the second credibility index to generate a biological weight coefficient, and derive a text weight coefficient based on the biological weight coefficient;

[0019] A weighted fusion module is configured to perform weighted calculation on the biological sentiment value and the text sentiment value according to the biological weight coefficient and the text weight coefficient to generate a real-time sentiment value;

[0020] A warning judgment module is configured to judge whether the real-time sentiment value continuously falls below a preset negative sentiment threshold and whether a continuous duration exceeds a preset time threshold; if the judgment result is yes, a warning signal is triggered, and a warning instruction is sent to a teacher end.

[0021] Compared with the prior art, the application has the following beneficial effects:

[0022] 1. By collecting network delay values and implementing nonlinear attenuation processing, a dynamic modeling of the distortion risk of biological signals is constructed, and further combined with the response delay of students to the teacher's questions and historical interaction data, a second credibility index is calculated to reflect the attention and sustained participation. This dual-index collaborative construction of signal reliability effectively solves the misjudgment problem caused by network fluctuations or insufficient interaction in existing multi-modal fusion schemes;

[0023] 2. Based on the double channels of student barrage text and facial image data, an emotion recognition path is constructed, which can fully cover the emotional response of students in language expression and non-verbal behavior. BRIEF DESCRIPTION OF DRAWINGS

[0024] The disclosure of the present application will be described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present application, and in the drawings, the same reference numerals are used to refer to the same parts. Among them:

[0025] Figure 1 is a flow chart of the method of the present application;

[0026] Figure 2 is a data flow chart of the present application;

[0027] Figure 3 is a system module diagram of the present application. DETAILED DESCRIPTION

[0028] It is easy to understand that according to the technical scheme of the present application, those skilled in the art can propose a variety of structures and implementation methods that can be replaced with each other without changing the essential spirit of the present application. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical scheme of the present application, and should not be considered as the whole or as a limitation or restriction of the technical scheme of the present application.

[0029] SUMMARY

[0030] In traditional multimodal teaching feedback systems, fixed-weight strategies cannot adapt to the dynamic impact of network latency and response time fluctuations on data reliability. This leads to a deviation in the fusion ratio of biometric signals and text feedback from the requirements of real-world scenarios, thereby reducing the reliability of affective computing results. For example, in a real-time interactive classroom with hundreds of students, when a teacher asks a question, the text data from student responses via bullet comments experiences a time shift due to network transmission latency. Simultaneously, facial expression images suffer keyframe loss due to network bandwidth fluctuations. At this point, the system still linearly superimposes biometric and text affective values ​​according to preset weights, ignoring the nonlinear attenuation effect of network latency on the integrity of biometric signal sampling, and the quantitative impact of response time differences on the timeliness of text feedback. The resulting real-time affective values ​​exhibit phase deviation in the time dimension and introduce a decrease in signal-to-noise ratio in the numerical dimension, making it impossible for teachers to accurately identify the true learning status of the student group. If these problems are not addressed, persistent affective computing errors will lead to misjudgments or omissions in the early warning mechanism, preventing teachers from adjusting their teaching strategies in a timely manner, and hindering the effective correction of student distraction and misunderstanding. Ultimately, this results in distorted teaching effectiveness evaluation and a decline in platform service quality.

[0031] To address the aforementioned issues, this application first analyzes the nonlinear impact of latency fluctuations on data reliability, considering the incomplete sampling of biosignals due to network latency, and explores the feasibility of adjusting the reliability index through a dynamic decay function. Simultaneously, to address the interference of response time differences on the timeliness of text feedback, it considers establishing a time-sensitive reliability assessment mechanism by incorporating historical response data. Furthermore, this application finds that fixed-weight strategies cannot adapt to dynamic changes in the quality of multi-source data, and proposes coupling network latency reliability, response time reliability, and course type-based weights to construct a dynamic weight allocation model, enabling the fusion ratio of biosignals and text signals to be adjusted in real time according to data quality.

[0032] In this regard, such as Figure 1 As shown, this application proposes a method for optimizing blended learning evaluation based on real-time communication, including the following steps:

[0033] The system acquires the target student's bullet screen text data, student facial images, and network latency values ​​between the teacher and student ends. The network latency value refers to the time difference in data transmission between the teacher and student ends, which can be obtained by using the timestamp difference calculation method of the network protocol stack. It is used to assess the impact of network fluctuations on the quality of biometric data acquisition.

[0034] The bullet screen text data is subjected to sentiment scoring to generate a text sentiment value; biometric features are extracted based on the student facial image data to calculate a biometric sentiment value; the bullet screen text data refers to the real-time text feedback information sent by students through the bullet screen function, which can be specifically processed using natural language processing technology for word segmentation and sentiment analysis to capture students' immediate emotional expressions during the teaching process. The student facial image data refers to facial biometric information collected through a camera, which can be specifically processed using computer vision technology to extract the intensity values ​​of the electromyographic signals of the brow apex and the corner of the mouth to reflect the student's true emotional state.

[0035] A nonlinear decay process is performed based on the network latency value to generate a first credibility index. The nonlinear decay process refers to an algorithm that dynamically adjusts the credibility based on the degree of network latency fluctuation. Specifically, an exponential decay formula combined with network volatility can be used to calculate the credibility index, which is used to reduce the weight of biometric data when network latency exceeds the standard.

[0036] The time interval between the teacher's question and the student's response in the live chat is obtained, and a second credibility index is generated by combining historical response data. The second credibility index is a dynamic evaluation indicator based on response time and historical data. Specifically, it can be generated by relative difference value comparison and trend analysis methods to quantify the credibility of student interaction behavior.

[0037] The basic biometric weights are determined based on the course type. The basic biometric weights, the first credibility index, and the second credibility index are coupled to generate biometric weight coefficients. The text weight coefficients are then derived based on these biometric weight coefficients. The biometric weight coefficients refer to the dynamic weight values ​​of biometric data in the fusion calculation. Specifically, they can be generated by multiplying the credibility index with the course type parameter. These coefficients are used to dynamically adjust the contribution ratio of biometrics to the final sentiment value based on data quality.

[0038] The biological sentiment value and the text sentiment value are weighted and calculated according to the biological weight coefficient and the text weight coefficient to generate a real-time sentiment value. The real-time sentiment value refers to the weighted calculation result of the combined biological features and text features. Specifically, it can be generated by a linear weighted fusion algorithm to reflect the students' emotional state in the teaching scenario in real time.

[0039] The system determines whether the real-time emotional value remains below a preset negative emotional threshold and whether the duration exceeds a preset time threshold. If the determination is yes, an early warning signal is triggered, and an early warning instruction is sent to the teacher. The preset negative emotional threshold refers to the critical emotional value that triggers teaching intervention. It can be specifically set through historical classroom data statistical analysis methods to identify persistent negative emotional states that require teacher intervention.

[0040] The core innovation of this application lies in constructing a dual credibility assessment mechanism based on network quality and behavioral patterns. By dynamically coupling the weight coefficients of biometric and textual features, it overcomes the shortcomings of traditional fixed-weight strategies that cannot adapt to network fluctuations and changes in student interaction behavior. Simultaneously, through the combined application of a nonlinear decay algorithm and response time trend analysis, it achieves real-time quantitative assessment of the credibility of multi-source data, ensuring the robustness of the affective computing model in complex teaching environments.

[0041] like Figure 2 The diagram shown is a data flow chart of this application; as a preferred embodiment, the solution of this application is specifically implemented as follows:

[0042] First, the system acquires the target students' live chat text data, student facial images, and network latency between the teacher's and student's devices. The chat text data includes real-time comments sent by students in the live class session. Student facial images are captured via the student's camera and pre-processed before being transmitted to the server. Network latency is obtained by measuring the time difference between when the teacher sends information and when the student receives it.

[0043] Next, sentiment scoring is performed on the bullet screen text data. A pre-trained natural language processing model is used to segment the text and perform sentiment analysis, outputting a value between -1 and 1 to represent the text's sentiment score. Simultaneously, biometric features are extracted based on student facial image data. A facial expression recognition algorithm is used to extract the location and changes of key facial feature points, calculating a biometric sentiment score, also between -1 and 1.

[0044] Then, a non-linear decay process is performed based on the network latency value to generate a first credibility index. Specifically, an exponential decay function can be used to calculate the credibility index, ensuring that the greater the latency, the lower the credibility. Simultaneously, the time interval between the teacher's question and the student's response is obtained and compared with the student's historical average response time to generate a second credibility index. The more timely the response, the higher the credibility.

[0045] Furthermore, a basic biological weight is determined based on the course type (e.g., theoretical course, practical course, etc.). This basic biological weight is then coupled with the two aforementioned credibility indices to obtain the final biological weight coefficient. The text weight coefficient is obtained by subtracting the biological weight coefficient from 1, ensuring that the sum of the two is 1.

[0046] Finally, the biological sentiment value and the text sentiment value are weighted and calculated according to the biological weight coefficient and the text weight coefficient to generate a real-time sentiment value. This sentiment value is continuously monitored. If it is continuously lower than the preset negative sentiment threshold and the duration exceeds the preset time threshold, an early warning signal is triggered, and an early warning instruction is sent to the teacher's end, prompting the teacher to pay attention to the students' status and adjust the teaching strategy in a timely manner.

[0047] Through the above-described scheme, this application achieves dynamic perception of network latency and student response time, converting them into a credibility index, effectively solving the problem that fixed-weight strategies cannot adapt to changes in data quality. By dynamically adjusting the fusion ratio of biosignals and text signals, the reliability of sentiment computing results is improved. Simultaneously, an early warning mechanism is introduced, enabling teachers to promptly identify the true learning status of student groups, effectively avoiding distortion in teaching effectiveness evaluation and a decline in platform service quality. In large-scale real-time interactive classrooms, this scheme can accurately capture changes in student emotions, providing strong support for teachers to adjust teaching strategies, thereby improving the overall quality and effectiveness of blended learning.

[0048] This application further proposes performing nonlinear attenuation processing based on the network latency value to generate a first confidence index, including:

[0049] Obtain the network latency value sequence within a preset time window, and use the ratio of the standard deviation to the mean of the network latency value sequence as the network volatility. The network volatility quantifies the dispersion of network latency by the ratio of the standard deviation to the mean, and a preset volatility threshold is used to distinguish between stable and volatile network states.

[0050] Determine whether the network volatility is less than a preset volatility threshold;

[0051] If the judgment result is yes, perform the baseline attenuation process:

[0052] If the network latency value does not exceed the preset latency threshold, set the first confidence index to 1;

[0053] If the network latency value exceeds the preset latency threshold, exponential decay will be performed according to the preset basic decay rate.

[0054] If the judgment result is negative, a volatility amplification factor is generated based on the magnitude of the network volatility exceeding the preset volatility threshold. The actual decay rate is obtained by superimposing the preset base decay rate with the volatility amplification factor. Exponential decay is performed according to the actual decay rate, and the first confidence index is guaranteed to be no lower than the preset lower limit.

[0055] By calculating the volatility of the network latency sequence in real time, a distinction is made between stable and high-volatility scenarios. In stable scenarios, baseline decay is performed solely based on the latency value to avoid over-adjustment. In high-volatility scenarios, a volatility amplification factor is added, causing the credibility index to decay more rapidly as the volatility increases. A preset lower limit constrains the minimum value of the credibility index to prevent it from dropping to zero due to extreme volatility. This scheme allows the credibility index to simultaneously reflect both the absolute value of latency and the degree of network volatility, improving the rationality of biosignal weight allocation and thus optimizing the accuracy of real-time sentiment value calculation.

[0056] Through the above technical solution, this application can dynamically adjust the credibility index according to network latency and fluctuations, effectively improving the accuracy and adaptability of blended learning assessment. It maintains high credibility when network conditions are good, while appropriately reducing credibility when network latency or fluctuations occur, avoiding interference from network factors on the assessment results. Simultaneously, by setting a credibility lower limit, it ensures that the assessment system maintains basic functionality even under extreme network conditions. This adaptive credibility adjustment mechanism can better reflect changes in data quality in the actual teaching environment, improving the overall reliability of blended learning assessment.

[0057] This application further performs exponential attenuation according to a preset base attenuation rate, including:

[0058] The attenuation formula is:

[0059] ,in As the first credibility index, To preset the basic attenuation coefficient, This is the difference between the network latency value and the preset latency threshold.

[0060] The formula for calculating the fluctuation amplification factor is as follows:

[0061] ,in This is the difference between the network volatility and the preset volatility threshold.

[0062] Exponential function Used to calculate the difference between the network latency value and a preset latency threshold. The rate of decay when converted into a credibility index. The preset basic attenuation coefficient controls the steepness of the attenuation curve; the difference... The larger the credibility index, the higher the credibility index. The faster the exponential rate decreases, the more accurately the formula reflects the impact of high latency on reliability through the nonlinear characteristics of the exponential function.

[0063] Fluctuation amplification factor By measuring the difference between network volatility and a preset volatility threshold Linearly superimposed to a baseline value of 1, this achieves dynamic adjustment of the actual decay rate. When the network volatility exceeds a preset volatility threshold, It is a positive value, that is A value greater than 1 amplifies the actual decay rate beyond the base decay rate, accelerating the decay of the credibility index. This factor directly affects the base decay rate, resulting in a higher actual decay rate and a faster decline in the credibility index when network volatility is higher.

[0064] By combining the two sets of formulas, the quantitative impact of latency difference on credibility is ensured, and the dynamic adjustment of network volatility on decay rate is realized, thereby improving the calculation accuracy and scenario adaptability of credibility index.

[0065] For example: preset base attenuation coefficient It can be set to 0.05, and the preset latency threshold can be set to 100. When a network latency value of 150 is detected, .

[0066] Substituting into the attenuation formula, we get: .

[0067] Assuming the preset volatility threshold is 0.2, the currently detected network volatility is 0.3;

[0068] but Substituting into the formula for calculating the fluctuation amplification factor, we get:

[0069] .

[0070] Therefore, the actual decay rate is At this point, the first credibility index The value is: .

[0071] Finally, the calculated first credibility index is judged. Whether the requirement of the lower confidence limit is met, the final output is a definite value.

[0072] Through the above technical solutions, this application can dynamically adjust the credibility index according to network latency and fluctuations, improving its adaptability to changes in network quality. Simultaneously, the exponential decay function can more accurately reflect the non-linear impact of increased latency on credibility, avoiding excessive penalties that might result from simple linear decay. Furthermore, the introduction of a fluctuation amplification factor makes the system more sensitive to network instability, further enhancing the accuracy and robustness of credibility assessment.

[0073] This application further proposes the following process for generating the second credibility index:

[0074] Obtain the historical average response time of the target student and calculate the relative difference between the current response time interval and the historical average response time. Historical average response time is calculated by statistically analyzing the historical response time data of target students within a preset period; relative difference value It is expressed as a percentage of the difference between the current response time and the historical average response time.

[0075] Determine if the current response time exceeds the historical average response time;

[0076] If the judgment result is yes, then a negative attenuation factor is set based on the relative difference value. The preset basic confidence level is reduced by the negative attenuation factor as the second confidence index, and the reduction result is controlled to be no lower than the preset lower limit value.

[0077] The negative attenuation factor is adjusted linearly or non-linearly based on the magnitude of the relative difference value;

[0078] For example: negative attenuation factor based on relative difference value Set to: ;

[0079] Second Credibility Index The calculation formula is:

[0080] ,in To establish a baseline confidence level; in this formula, the relative difference value If it is a positive value, then The value of is greater than 1, that is The second credibility index is less than 1. Less than the preset basic confidence level .

[0081] If the judgment result is negative, the response time change trend within the preset period is analyzed. When there is a continuous acceleration trend, a positive gain factor is set based on the acceleration magnitude. The preset basic credibility is improved by the positive gain factor as the second credibility index, and the improvement result is controlled not to exceed the preset upper limit value.

[0082] The positive gain factor is calculated based on the slope change of the response time within a preset period. If three consecutive response time intervals show a decreasing trend, the gain factor is generated according to the square root of the decreasing rate.

[0083] For example: the positive gain factor based on the acceleration magnitude Set to: ,in To accelerate the magnitude;

[0084] Second Credibility Index The calculation formula is:

[0085] In this formula, the acceleration magnitude If it is a positive value, then The value of is greater than 1, that is The second credibility index is greater than 1. Greater than the preset basic credibility This mechanism, by dynamically adjusting the credibility index, avoids the excessive impact of occasional delays on the assessment results while capturing positive trends in student participation, thereby improving the accuracy of sentiment value calculation.

[0086] When the current response time is detected to exceed the historical average response time, a negative decay mechanism is automatically triggered.

[0087] For example, if the historical average response time is 5 seconds and the current response time is 7 seconds, the relative difference is 40%.

[0088] Then negative attenuation factor Values The second credibility index is coming soon. The value is reduced to a preset base confidence level. 71.4%, while controlling the reduction result to be no less than the preset lower limit value of 60%.

[0089] When the response time is shorter than the historical average and there is a continuous acceleration trend, the positive decay mechanism is automatically triggered:

[0090] For example: if the acceleration amplitude is set to 0.2, then the positive gain factor... The value is 1.2, which is the second credibility index. The value is increased to the preset base confidence level. The result is 1.2 times the original value, while controlling the improvement result to not exceed the preset upper limit of 1.2.

[0091] The final output of the second credibility index is the result of the above calculation.

[0092] Through the above technical solution, this application achieves dynamic evaluation of student response time. By comparing with historical data and combining it with current response trends, the credibility index is adjusted in real time. This method can more accurately reflect students' actual participation and attention status, improving the accuracy and timeliness of teaching feedback. At the same time, by setting upper and lower limits, excessive fluctuations in the credibility index in extreme cases are avoided, ensuring the stability and reliability of the evaluation results.

[0093] This application further proposes the following coupling process for the biological weight coefficient:

[0094] The first credibility index and the second credibility index are multiplied to generate a comprehensive credibility factor; the multiplication operation transforms the correlation between the first credibility index and the second credibility index into a single quantitative indicator.

[0095] When the overall credibility factor is lower than the preset credibility threshold, the real-time interaction frequency of the current course stage is detected. If the interaction frequency exceeds the preset activity threshold, the overall credibility factor is increased proportionally.

[0096] The initial biological weight is generated by multiplying the basic biological weight by the comprehensive credibility factor; the product operation of the basic biological weight and the comprehensive credibility factor realizes the initial integration of course type and real-time credibility.

[0097] The system obtains the real-time emotional value sequence of the target students in history class, and calculates the weight correction coefficient based on the fluctuation characteristics of the real-time emotional value sequence; the standard deviation of the fluctuation of the historical emotional value sequence is calculated by using the sliding window method to extract data within a preset time period; the weight correction coefficient maps the degree of fluctuation to a correction factor through the reciprocal function, and the larger the standard deviation of fluctuation, the smaller the correction coefficient.

[0098] Multiply the initial biological weight by the weight correction factor to output the final biological weight coefficient, and ensure that the result is within the preset biological weight range.

[0099] For example, when the overall credibility factor is 0.6 and the preset credibility threshold is 0.7, if the interaction frequency within the current five minutes is detected as 3 times per minute, which exceeds the preset activity threshold of 2 times per minute, the overall credibility factor will be increased to 0.9.

[0100] The initial biological weight of 0.45 is obtained by multiplying the base biological weight of 0.5 by the improved comprehensive credibility factor. The standard deviation of fluctuation is calculated based on the historical classroom emotional value sequence. When it exceeds a preset stability threshold, a weight correction coefficient is used to adjust the initial biological weight, ultimately outputting the biological weight coefficient, which is constrained within a preset biological weight range. This process, through a dynamic adjustment mechanism, introduces real-time interactive compensation when the comprehensive credibility is insufficient, and optimizes weight allocation based on historical emotional stability, achieving a multi-dimensional assessment of the credibility of biological signals.

[0101] Through the above technical solution, this application achieves dynamic adjustment of biological weight coefficients. Therefore, the system can adaptively adjust the weight of biological signals in emotion assessment based on network conditions, student responses, and historical emotional fluctuation characteristics. This dynamic weight allocation mechanism improves the accuracy and reliability of emotion assessment, enabling the system to better adapt to different teaching scenarios and individual student differences. Furthermore, by introducing factors such as interaction frequency and historical emotional fluctuations, the system can more comprehensively consider student participation and emotional stability, thereby generating more reasonable weight coefficients. This method effectively solves the limitations of fixed-weight strategies in dealing with complex teaching environments, providing strong support for achieving high-precision real-time emotion perception.

[0102] This application further proposes the following calculation process for the weight correction coefficient:

[0103] The standard deviation of volatility is calculated based on the real-time sentiment value sequence. ;

[0104] If the standard deviation of the fluctuation is lower than the preset stability threshold, maintain the initial biological weights;

[0105] If the standard deviation of the fluctuation exceeds the preset stability threshold, the correction factor is calculated using the following formula. :

[0106] .

[0107] Fluctuation Standard Deviation The calculation is achieved by statistically analyzing the dispersion of real-time sentiment value sequences, with a preset stability threshold used to distinguish whether sentiment fluctuations are significant. Correction coefficient. The standard deviation of fluctuation is calculated using the reciprocal function form. The larger the correction factor, the higher the correction factor. The smaller the value, the lower the biological weight coefficient, and it remains within the (0,1) range to avoid excessive weight adjustment. The preset stable threshold is determined based on the distribution of historical classroom emotional data, for example, by obtaining it through statistics from multiple rounds of teaching experiments.

[0108] For example, when hour, The initial biological weights were adjusted to 66.6% of their original values; when hour, The initial biological weight was adjusted to 33.3% of its original value. By dynamically adjusting the biological weight coefficient, the weight of the biological signal is reduced when there are large emotional fluctuations, thus reducing noise interference and improving the reliability of real-time emotional values.

[0109] Through the above technical solution, this application achieves a quantitative assessment of students' emotional state fluctuations and dynamically adjusts the biological weight coefficient accordingly. When students experience significant emotional fluctuations, the biological weight is reduced to mitigate the impact of unstable biological signals, thereby improving the accuracy and reliability of blended learning assessment. Furthermore, this solution can adapt to individual differences among students, generating personalized weight correction strategies for each student, further enhancing the relevance of the teaching assessment.

[0110] This application further proposes deriving text weight coefficients based on the aforementioned biological weight coefficients, including:

[0111] Calculate the basic text weights :

[0112] in Biological weighting coefficient;

[0113] The number of bullet comments per unit time is extracted based on the bullet comment text data;

[0114] The baseline value of the balance factor is retrieved from the preset mapping table based on the number of bullet comments per unit time. ;

[0115] Based on the balance factor benchmark value Basic text weights Make corrections to obtain text weight coefficients. :

[0116] And ensure the text weight coefficient Within the preset text weight range.

[0117] Among them, basic text weights Through biological weight coefficient The complementary operations are used to ensure that the sum of the biological and text weights remains constant. The extraction of the number of bullet comments employs a time window statistical mechanism, with a balancing factor baseline value. The mapping table is generated based on historical data and contains correction coefficients corresponding to different bullet screen density ranges. During the correction process, the baseline value of the balance factor is... Basic text weights Linear scaling is applied, and upper and lower threshold constraints are used to ensure the stability of weight allocation.

[0118] For example, during peak course interaction periods, if the system detects 20 bullet comments per minute, it will read from the mapping table. Weight the basic text Revised to The correction process responds in real time to changes in student engagement, automatically reducing text weights when bullet comments are sparse to avoid low-quality text data interfering with the fusion results. A hard constraint of a preset text weight range [0.3, 0.7] ensures that the weight allocation between biological and text data remains within an effective range, preventing weight imbalances in extreme cases.

[0119] Through the above technical solution, this application realizes the dynamic derivation of text weight coefficients based on biological weight coefficients. By introducing a mapping relationship between the number of bullet comments and a balancing factor, the text weight coefficients can be adaptively adjusted according to the actual bullet comment activity. This dynamic weight allocation mechanism improves the flexibility and accuracy of multimodal data fusion and effectively solves the problem that fixed weight strategies are difficult to cope with dynamic changes in the teaching environment.

[0120] This application further proposes that the calculation of biological emotion value includes:

[0121] Based on the student's facial image data, the electrophysiological signal intensity values ​​of the brow tip muscles and the corner of the mouth muscles are extracted. The brow tip muscle electrophysiological signal intensity value is obtained by detecting the muscle contraction frequency and amplitude in the brow area through an image recognition algorithm and converting it into a normalized value of 0 to 1. The corner of the mouth muscle electrophysiological signal intensity value is obtained by analyzing the displacement of the orbicularis oris muscle and combining it with the upward angle of the corner of the mouth.

[0122] The arithmetic mean of the electrical signal intensity values ​​of the brow muscles and the corner of the mouth muscles is used to generate the bio-emotion value.

[0123] The arithmetic average operation is achieved by directly adding the strength values ​​of the two signals and then dividing by 2. For example, when the signal strength at the tip of the eyebrow is 0.8 and the signal strength at the corner of the mouth is 0.4, the biological emotional value is (0.8 + 0.4) / 2 = 0.6. The acquisition time window for the two signals is constrained to within 500 milliseconds to ensure time synchronization.

[0124] Specifically, facial image data, after preprocessing, is input into a dual-channel feature extraction model. The first channel locates the area between the eyebrows and calculates the grayscale change rate of consecutive frames, mapping the change rate to the electromyographic signal intensity value of the brow apex muscle. The second channel tracks the coordinates of the corner of the mouth and measures its offset relative to a reference position, generating the electromyographic signal intensity value of the corner of the mouth muscle based on the ratio of the offset distance to a preset threshold. The two signals are time-aligned and then fed into an arithmetic averaging unit to output the final bio-emotion value. This design effectively suppresses signal noise caused by transient muscle twitching or changes in lighting by fusing bio-features from different facial regions. For example, when a student experiences abnormal twitching of the corner of their mouth due to a brief cough, the stability of the brow apex signal can offset the instantaneous fluctuation of the corner of the mouth signal, maintaining the continuity of the bio-emotion value and avoiding interference from single signal distortion in emotion assessment.

[0125] Through the above technical solution, this application effectively solves the problem that single features of biosignals are easily interfered with by local muscle movements. By fusing the electromuscular activity features of different facial regions, the robustness of emotional representation is significantly improved. This technical solution achieves noise-resistant capture of real emotional states, avoiding misjudgments caused by brief twitches of individual facial muscles or poor device contact, and provides stable and reliable biosignature input for subsequent multimodal emotion fusion.

[0126] The generation of further text sentiment values ​​in this application includes:

[0127] The bullet screen text data is segmented based on a preset sentiment dictionary;

[0128] Extract the sentiment weight value corresponding to each sentiment word, and generate the text sentiment value by averaging the word frequency of the bullet screen text data within a preset time window.

[0129] As a preferred embodiment, the solution of this application is implemented as follows: In the process of processing bullet screen text data, a preset sentiment dictionary is first divided into a positive sentiment lexicon, a negative sentiment lexicon, and a neutral lexicon. Words in each lexicon are assigned different sentiment weight values, with positive sentiment words corresponding to positive weights, negative sentiment words to negative weights, and neutral words having a weight of zero. After the bullet screen text data is processed through word segmentation, a set of sentiment words belonging to the above three lexicons is selected. For each sentiment word, its frequency is calculated by counting the number of times it appears within a preset time window. Subsequently, the sentiment weight value of each sentiment word is multiplied by its frequency to obtain the sentiment contribution value of that word. Finally, all sentiment contribution values ​​are accumulated within the time window and divided by the total word frequency to generate the text sentiment value.

[0130] Through the aforementioned technical solution, this application effectively captures the emotional fluctuation trends expressed by students in the classroom through bullet comments by dynamically statistically analyzing the real-time distribution frequency of different emotional words in the bullet comment text and performing weighted average calculations based on preset emotional weights. This method avoids the problem of insufficient sensitivity to high-frequency emotional words in traditional fixed-weight allocation methods. Especially in large-scale bullet comment interaction scenarios, the word frequency weighting mechanism improves the accuracy of identifying group emotional tendencies, thereby providing teachers with more accurate real-time emotional feedback data support.

[0131] like Figure 3 The diagram shown is a system module diagram of this application. This application further proposes a hybrid teaching evaluation optimization system based on real-time communication, including a data acquisition module, an emotion value calculation module, a credibility index calculation module, a weight coefficient calculation module, a weighted fusion module, and an early warning judgment module.

[0132] The data acquisition module obtains bullet screen text data, facial images, and network latency values ​​through an interface;

[0133] The sentiment value calculation module uses a word segmentation algorithm to process the bullet screen text to generate text sentiment values, and uses an image recognition algorithm to extract the electrical signals of the muscles at the tip of the eyebrows and the corners of the mouth to generate biological sentiment values.

[0134] The credibility index calculation module executes a nonlinear decay algorithm to process network latency values ​​to generate a first credibility index, and combines it with the historical response database to calculate time intervals to generate a second credibility index.

[0135] The weight coefficient calculation module calls the course type database to determine the basic biological weight, generates the biological weight coefficient by coupling the credibility index through multiplication operation, and generates the text weight coefficient by correcting the basic text weight based on the bullet screen number mapping table.

[0136] The weighted fusion module performs linear weighted calculations and outputs real-time sentiment values;

[0137] The early warning judgment module detects the continuous state of real-time emotion value through a threshold comparator and triggers an early warning signal to be transmitted to the teacher's terminal.

[0138] Specifically, the data acquisition module captures student-side bullet screen text and facial images in real time, while simultaneously monitoring network latency values ​​on both the teacher and student sides. The sentiment value calculation module segments the bullet screen text, matches weight values ​​from the sentiment dictionary to generate text sentiment values, and simultaneously collects facial muscle electromyography signal intensity using a biosensor, calculating the arithmetic mean to generate a biosensory value. The credibility index calculation module calculates the ratio of the standard deviation to the mean of the network latency value sequence as the network volatility, selects an exponential decay function based on the volatility threshold to generate a first credibility index, and generates a second credibility index by combining the difference between the current response time and the historical average response time. The weight coefficient calculation module multiplies the basic biosensory weights and credibility indices to generate initial biosensory weights, calculates correction coefficients based on the standard deviation of the real-time sentiment value sequence volatility, and finally outputs biosensory weight coefficients. The text weight coefficients are adjusted by querying the balance factor based on the number of bullet screens. The weighted fusion module fuses the biosensory values ​​and text sentiment values ​​according to a dynamic weight ratio to generate a real-time sentiment value. The early warning judgment module continuously monitors the time series of real-time sentiment values. When it detects values ​​continuously falling below a negative threshold and exceeding a timeout, it activates the early warning command transmission link.

[0139] Through the above technical solutions, this application addresses the problem of dynamic changes in the credibility of multimodal data caused by network latency, response latency, and fluctuations in engagement. It achieves dynamic optimization of the weights of biosignals and text signals, avoiding interference from low-credibility data in sentiment assessment under a fixed-weight strategy and improving the accuracy of real-time sentiment values. Simultaneously, through an automated triggering mechanism for early warning signals, it ensures that teachers can promptly perceive the persistence of students' negative emotions and optimize teaching interaction strategies.

[0140] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for optimizing blended learning evaluation based on real-time communication, characterized by: Includes the following steps: Acquire the target student's bullet screen text data, student's facial image, and network latency value between the teacher's and student's ends; The bullet screen text data is subjected to sentiment scoring to generate a text sentiment value; Biometric features are extracted from the student's facial image data to calculate bio-emotional values; A non-linear attenuation process is performed based on the network latency value to generate a first confidence index; The time interval between the teacher's question and the student's response in the live chat is obtained, and a second credibility index is generated by combining historical response data. The basic biological weight is determined according to the course type. The basic biological weight, the first credibility index, and the second credibility index are coupled to generate the biological weight coefficient. The text weight coefficient is then derived based on the biological weight coefficient. The biological sentiment value and the text sentiment value are weighted and calculated according to the biological weight coefficient and the text weight coefficient to generate a real-time sentiment value; Determine whether the real-time sentiment value is continuously lower than a preset negative sentiment threshold and whether the duration exceeds a preset time threshold; If the judgment result is yes, an early warning signal is triggered, and an early warning instruction is sent to the teacher's end.

2. The method for optimizing blended learning evaluation based on real-time communication according to claim 1, characterized in that: Based on the network latency value, a non-linear attenuation process is performed to generate a first confidence index, including: Obtain the network latency value sequence within a preset time window, and use the ratio of the standard deviation to the mean of the network latency value sequence as the network volatility. Determine whether the network volatility is less than a preset volatility threshold; If the judgment result is yes, perform the baseline attenuation process: If the network latency value does not exceed the preset latency threshold, set the first confidence index to 1; If the network latency value exceeds the preset latency threshold, exponential decay will be performed according to the preset basic decay rate. If the judgment result is negative, a volatility amplification factor is generated based on the magnitude of the network volatility exceeding the preset volatility threshold. The actual decay rate is obtained by superimposing the preset base decay rate with the volatility amplification factor. Exponential decay is performed according to the actual decay rate, and the first confidence index is guaranteed to be no lower than the preset lower limit.

3. The method for optimizing blended learning evaluation based on real-time communication according to claim 2, characterized in that: Exponential decay based on a preset base decay rate includes: The attenuation formula is: ,in As the first credibility index, To preset the basic attenuation coefficient, This is the difference between the network latency value and the preset latency threshold. The formula for calculating the fluctuation amplification factor is as follows: ,in This is the difference between the network volatility and the preset volatility threshold.

4. The method for optimizing blended learning evaluation based on real-time communication according to claim 1, characterized in that: The process of generating the second credibility index is as follows: Obtain the historical average response time of the target student and calculate the relative difference between the current response time interval and the historical average response time. Determine if the current response time exceeds the historical average response time; If the judgment result is yes, then a negative attenuation factor is set based on the relative difference value. The preset basic confidence level is reduced by the negative attenuation factor as the second confidence index, and the reduction result is controlled to be no lower than the preset lower limit value. If the judgment result is negative, the response time change trend within the preset period is analyzed. When there is a continuous acceleration trend, a positive gain factor is set based on the acceleration magnitude. The preset basic credibility is improved by the positive gain factor as the second credibility index, and the improvement result is controlled not to exceed the preset upper limit value.

5. The method for optimizing blended learning evaluation based on real-time communication according to claim 1, characterized in that: The coupling process of the biological weight coefficient is as follows: Perform a product operation on the first credibility index and the second credibility index to generate a comprehensive credibility factor; When the overall credibility factor is lower than the preset credibility threshold, the real-time interaction frequency of the current course stage is detected. If the interaction frequency exceeds the preset activity threshold, the overall credibility factor is increased proportionally. The initial biological weight is generated by multiplying the basic biological weight by the comprehensive credibility factor. Obtain the real-time sentiment value sequence of the target students in history class, and calculate the weight correction coefficient based on the fluctuation characteristics of the real-time sentiment value sequence; Multiply the initial biological weight by the weight correction factor to output the final biological weight coefficient, and ensure that the result is within the preset biological weight range.

6. The method for optimizing blended learning evaluation based on real-time communication according to claim 5, characterized in that: The calculation process for the weight correction coefficient is as follows: The standard deviation of volatility is calculated based on the real-time sentiment value sequence. ; If the standard deviation of the fluctuation is lower than the preset stability threshold, maintain the initial biological weights; If the standard deviation of the fluctuation exceeds the preset stability threshold, the correction factor is calculated using the following formula. : 。 7. The method for optimizing blended learning evaluation based on real-time communication according to claim 1, characterized in that: The text weight coefficients derived based on the aforementioned biological weight coefficients include: Calculate the basic text weights : in Biological weighting coefficient; The number of bullet comments per unit time is extracted based on the bullet comment text data; The baseline value of the balance factor is retrieved from the preset mapping table based on the number of bullet comments per unit time. ; Based on the balance factor benchmark value Basic text weights Make corrections to obtain text weight coefficients. : And ensure the text weight coefficient Within the preset text weight range.

8. The method for optimizing blended learning evaluation based on real-time communication according to claim 1, characterized in that: The calculation of the biological emotion value includes: Based on the student's facial image data, extract the electrical signal intensity values ​​of the brow muscle and the corner of the mouth muscle. The arithmetic mean of the electrical signal intensity values ​​of the brow muscles and the corner of the mouth muscles is used to generate the bio-emotion value.

9. The method for optimizing blended learning evaluation based on real-time communication according to claim 1, characterized in that: The generation of the text sentiment value includes: The bullet screen text data is segmented based on a preset sentiment dictionary; Extract the sentiment weight value corresponding to each sentiment word, and generate the text sentiment value by averaging the word frequency of the bullet screen text data within a preset time window.

10. A blended learning evaluation optimization system based on real-time communication, characterized in that: include: The data acquisition module is used to acquire the target student's bullet screen text data, student facial images, and network latency values ​​between the teacher's and student's ends; The sentiment value calculation module is used to score the sentiment of the bullet screen text data and generate a text sentiment value; and to extract biometric features based on the student facial image data and calculate a biometric sentiment value. The credibility index calculation module is used to perform non-linear attenuation processing based on the network latency value to generate a first credibility index; and to obtain the time interval from the time of the teacher's question to the time of the student's bullet screen response, and to generate a second credibility index by combining historical response data. The weight coefficient calculation module is used to determine the basic biological weight according to the course type, couple the basic biological weight, the first credibility index, and the second credibility index to generate the biological weight coefficient, and derive the text weight coefficient based on the biological weight coefficient. The weighted fusion module is used to calculate the biological sentiment value and the text sentiment value by weighting them according to the biological weight coefficient and the text weight coefficient, and generate a real-time sentiment value. The early warning judgment module is used to determine whether the real-time emotion value is continuously lower than the preset negative emotion threshold and whether the duration exceeds the preset time threshold. If the judgment result is yes, an early warning signal is triggered, and an early warning instruction is sent to the teacher's end.

Images