A learning effect evaluation system and method based on Logistic regression model

Abstract

The present invention relates to the field of data processing technology, and in particular to a learning effect evaluation system and method based on a Logistic regression model. The present invention determines whether a user has suspected behavior of catering to evaluation through the time difference between each question time of the user and a preset question time period of the current course, draws a histogram through the single viewing time of the user, determines the user's learning habit category according to the number of data corresponding to each data group in the histogram, determines the corresponding habit feature parameters according to different learning habit categories, and calculates the dominant feature representation value, and finally adds the dominant feature representation value and the preset general feature parameters to the Logistic regression model to calculate the probability of the user's qualified learning performance, thereby realizing the analysis of the user's suspicious behavior according to the user's learning process information, and extracting data with strong representation for suspicious users to add to the model calculation analysis to avoid misjudgment of the user's suspicious behavior, thereby improving the accuracy of model evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a learning effect evaluation system and method based on a Logistic regression model. Background Art

[0002] Logistic regression is mainly used to analyze the impact of one or more independent variables on binary or multi-classification dependent variables. It is widely used in practical applications and is often used in data mining, disease diagnosis and analysis, economic forecasting and other fields that require comprehensive evaluation of large amounts of data. Learning effect evaluation is an important part of educational evaluation and an indispensable means of teaching support in the increasingly widespread online teaching methods. It aims to ensure that online teaching activities can achieve the expected educational goals. By evaluating students' learning outcomes, teachers can better understand and grasp students' learning situations and adjust teaching strategies in a timely manner to improve students' learning effects. For the data collected on students' learning situations, the same data may correspond to different learning effects. Accurately judging students' behavior in meeting evaluation rules is an urgent problem that needs to be solved.

[0003] For example, Chinese Patent Publication No.: CN116664013A, the invention discloses a method for evaluating the effectiveness of a collaborative learning model, a ubiquitous intelligent learning system and a medium, the method comprising: obtaining target learning trajectory data and / or target scale data recorded by a target learner in a collaborative learning mode, wherein the target learning trajectory data is associated with learning trajectory data between at least two other learners, the reliability of the target scale data is greater than a preset reliability threshold, and determining a collaborative learning evaluation score corresponding to the target learner according to the target learning trajectory data and / or target scale data; determining an evaluation result according to the collaborative learning evaluation scores corresponding to at least two learners in the collaborative learning mode.

[0004] The prior art still has the following problems:

[0005] The existing technology does not take into account the phenomenon that users obtain good evaluation results by brushing data. The existing technology cannot analyze the suspicious behavior of users based on the information of the user's learning process. Moreover, it cannot extract data with strong representation for suspicious users and add it to the model calculation analysis to avoid misjudgment of the user's suspicious behavior, resulting in inaccurate model evaluation due to the lack of data representation of the data involved in the model calculation. Summary of the invention

[0006] To this end, the present invention provides a learning effect evaluation system and method based on a Logistic regression model to overcome the problem that the prior art cannot analyze the suspicious behavior of users based on the user's learning process information, and cannot extract data with strong representation for suspicious users and add it to the model calculation and analysis to avoid misjudgment of the user's suspicious behavior.

[0007] To achieve the above object, the present invention provides a learning effect evaluation method based on a Logistic regression model, comprising:

[0008] In response to the user completing the current course, obtaining a number of questioning moments of the user, and determining whether the user has suspected behavior of catering to the evaluation based on the time difference between each questioning moment and a preset questioning time period of the current course;

[0009] In response to the user having suspected behavior of catering to the evaluation, obtaining the single viewing time of the user, drawing a histogram of the single viewing time, and determining the learning habit category of the user according to the number of data corresponding to each data group in the histogram;

[0010] Determine corresponding habit characteristic parameters based on different learning habit categories to calculate advantage characteristic representation values;

[0011] The habit characteristic parameters include the total course viewing time, the number of course playback replays, the number of course playback pauses, and the single viewing time;

[0012] Add the advantageous characteristic representation value and the preset general characteristic parameters into the Logistic regression model to calculate the probability of the user's passing the learning performance;

[0013] The general characteristic parameters include the number of interactions the user has completed in the current course and the user's historical scores.

[0014] Furthermore, the process of obtaining the user's multiple questioning moments includes:

[0015] A number of questions asked by the user during the current course playback are obtained, and the time corresponding to each question is determined as the question time.

[0016] Furthermore, the step of determining the time difference between each questioning moment and the preset questioning time period of the current course includes:

[0017] Based on the historical data of questions asked in the current course, several question time periods are preset in the current course based on the playback progress time;

[0018] Filtering moments that are not within the questioning time period based on the questioning moment;

[0019] Determine the minimum time difference between the time and the time of the endpoint of the time period corresponding to the question time period;

[0020] A minimum time difference average value is calculated based on the minimum time difference corresponding to each moment, and the minimum time difference average value is determined as the time difference.

[0021] Furthermore, the process of determining whether the user has suspected behavior of catering to the evaluation includes:

[0022] Comparing the time difference with a preset time difference threshold;

[0023] If the time difference is greater than the time difference threshold, it is determined that the user has suspected behavior of catering to the evaluation.

[0024] Furthermore, the process of drawing a histogram of a single viewing duration includes:

[0025] The single viewing durations of the user are divided into a number of data groups based on a preset group distance, the duration of each data group is distributed as the horizontal axis, and the histogram is drawn with the number of data corresponding to each data group as the vertical axis.

[0026] Furthermore, the process of determining the user's learning habit category includes:

[0027] Obtaining the data quantity corresponding to each data group in the histogram, determining the maximum data quantity and the minimum data quantity, determining the difference between the maximum data quantity and the minimum data quantity as the frequency difference, and comparing the frequency difference with a preset frequency difference threshold;

[0028] If the frequency difference is less than or equal to the frequency difference threshold, determining that the learning habit category of the user is a first habit category;

[0029] If the frequency difference is greater than the frequency difference threshold, it is determined that the learning habit category of the user is the second habit category.

[0030] Furthermore, the process of determining corresponding habit characteristic parameters based on different learning habit categories includes:

[0031] If the user's learning habit category is the first habit category, determining the corresponding habit characteristic parameters includes the total course viewing time, the number of course playback playbacks, and the number of course playback pauses;

[0032] If the user's learning habit category is the second habit category, the corresponding habit characteristic parameters are determined to include the number of times the course playback is replayed, the number of times the course playback is paused, and the single viewing duration.

[0033] Furthermore, the process of calculating the dominant feature representation value based on different habit feature parameters includes:

[0034] Calculate the dominant characteristic representation value corresponding to the first habit category according to formula (1);

[0035] (1);

[0036] Among them, Xa is the dominant characteristic representation value corresponding to the first habit category, t a is the total viewing time of the course, t a0 is the corresponding duration of the current course progress bar, n1 is the number of times the course is played back, n2 is the number of times the course is paused, n0 is the reference total number of preset course playback and course playback pause times, λ is the weight coefficient of the total course viewing time, η is the weight coefficient of the number of times the course is played back and the number of times the course is paused, and e is a constant.

[0037] Calculate the dominant characteristic representation value corresponding to the second habit category according to formula (2);

[0038] (2);

[0039] Among them, X b is the dominant characteristic representation value corresponding to the second habit category, t b is the single viewing duration, t b0 is the preset reference value of single viewing time, n1 is the number of times the course is played back, n2 is the number of times the course is paused, n0 is the reference total number of preset times of course playback playback and course playback pause, γ is the weight coefficient of single viewing time, η is the weight coefficient of the number of times the course is played back and the number of times the course is paused, and e is a constant.

[0040] Furthermore, the process of calculating the probability that the user's learning performance is qualified includes:

[0041] Adding the dominant characteristic characterization value and the preset general characteristic parameters into the Logistic regression model (3);

[0042] ;

[0043] ;

[0044] ; (3)

[0045] Where P is the probability that the user's learning performance is qualified, X1 is the dominant feature representation value, and when the user's learning habit category is the first habit category, X1=X a , when the user's learning habit category is the second habit category, X1=X b , X2 is the number of interactions the user has completed in the current course, X2' is the preset reference value for the number of interactions, X3 is the user's historical score, X3' is the preset reference value for the user's historical score, β0, β1, β3 are logistic regression coefficients, and e is a constant.

[0046] Furthermore, the present invention also provides a learning effect evaluation system based on a Logistic regression model, which comprises:

[0047] An event recording unit, used to obtain the user's questioning moments and determine whether the user has suspected behavior of catering to the evaluation;

[0048] A drawing unit, connected to the event recording unit, for drawing a histogram of a single viewing time;

[0049] a category determination unit connected to the drawing unit, for determining the category of the user's learning habits according to the number of data corresponding to each data group in the histogram;

[0050] A feature characterization unit is connected to the drawing unit and is used to calculate the dominant feature characterization value and the probability of the user's learning performance being qualified.

[0051] Compared with the prior art, the beneficial effect of the present invention lies in that the present invention determines whether the user has suspected behavior of catering to the evaluation through the time difference between each question moment of the user and the preset question time period of the current course, draws a histogram through the single viewing time of the user, determines the user's learning habit category according to the number of data corresponding to each data group in the histogram, determines the corresponding habit feature parameters according to different learning habit categories, and calculates the dominant feature characterization value, and finally adds the dominant feature characterization value and the preset general feature parameters to the Logistic regression model to calculate the probability of the user's passing learning performance, thereby realizing the analysis of the user's suspicious behavior according to the user's learning process information, and extracting data with strong representation for suspicious users and adding them to the model calculation and analysis to avoid misjudgment of the user's suspicious behavior and improve the accuracy of model evaluation.

[0052] In particular, the present invention determines whether the user has suspected behavior of catering to the evaluation through the time difference between each question moment and the preset question time period of the current course. Those skilled in the art can understand that in the actual online learning process, users' concentrated questions will appear at the locations where course knowledge points and difficult points appear. If the question moment corresponding to the user's question is quite different from the time period corresponding to the concentrated questioning, and the above phenomenon exists in several of the user's questions, it can be determined that the user has not learned the course content, but has only interacted to achieve the interactive task. The present invention determines users whose time difference is greater than a preset value as having suspected behavior of catering to the evaluation through the time difference between each question moment and the preset question time period of the current course, and multiple data avoid data randomness, thereby realizing the analysis of users' suspicious behavior based on user learning process information.

[0053] In particular, the present invention obtains the single viewing time of the user and draws a histogram of the single viewing time. Those skilled in the art can understand that the learning habits of users are diverse. By drawing the single viewing time of the user for course learning as a histogram, a clear analysis of the user's learning habits can be achieved.

[0054] In particular, the present invention determines the learning habit category of the user according to the data quantity corresponding to each data group in the histogram. In the actual user learning habit analysis process, some users prefer a stable learning duration each time, and some users prefer irregular learning in fragmented time. The present invention calculates the difference between the maximum data quantity and the minimum data quantity in the data quantity corresponding to each data group. The smaller the difference, the more it indicates that the user's learning preference is a learning method with irregular duration, and the larger the difference, the more it indicates that the user's learning preference is a learning method with regular duration. By dividing the users' learning habit categories, it is convenient to determine targeted feature parameters for people with different learning habits.

[0055] In particular, the present invention determines user characteristics for users whose learning preference is irregular in duration based on the total course viewing duration, the number of course playback replays, and the number of course playback pauses. Those skilled in the art can understand that due to the fragmented use of time in the learning method with irregular duration, it is difficult for the learned knowledge points to establish mutual associations in incoherent learning. The longer the total course viewing duration, the deeper the memory of the knowledge points of the fragmented learning duration, and the better the learning effect pursued by the user. Therefore, the total course viewing duration can reduce the suspicion that users with irregular duration have suspected behaviors of catering to evaluations. Similarly, the more the number of course playback replays and the number of course playback pauses, the more frequent the phenomenon of users learning and understanding the key points and difficult points of the course. Therefore, reducing the number of course playback replays and the number of course playback pauses can reduce the suspicion that users with irregular duration have suspected behaviors of catering to evaluations. Furthermore, it is achieved to extract data with strong representation for suspicious users and add them to model calculation and analysis to avoid misjudgment of users' suspicious behaviors and improve the accuracy of model evaluation.

[0056] In particular, the present invention determines user characteristics based on the number of course playback replays, the number of course playback pauses, and the single viewing time for users whose learning preference is a stable learning time. Those skilled in the art can understand that for a learning method with a stable learning time, the longer the single viewing time, the easier it is for the learned knowledge points to establish mutual associations within a coherent time, which represents the better learning effect pursued by the user. Therefore, the single viewing time can reduce the suspicion that users with a stable learning time have suspected behaviors that cater to the evaluation. Similarly, the more the number of course playback replays and the number of course playback pauses, the more frequent the phenomenon of users learning and understanding the key points and difficult points of the course. Therefore, reducing the number of course playback replays and the number of course playback pauses can reduce the suspicion that users with a stable learning time have suspected behaviors that cater to the evaluation. Furthermore, it is achieved to extract data with strong representation for suspicious users and add them to model calculation and analysis to avoid misjudgment of users' suspicious behaviors and improve the accuracy of model evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A step diagram of a learning effect evaluation method based on a Logistic regression model according to an embodiment of the present invention;

[0058] Figure 2 A diagram showing the steps of determining the time difference between each questioning moment and a preset questioning time period of the current course according to an embodiment of the present invention;

[0059] Figure 3 A logic flow chart for determining whether a user has suspected behavior of catering to the evaluation according to an embodiment of the present invention;

[0060] Figure 4 A histogram corresponding to the first habit category of the learning habit category in the embodiment of the present invention;

[0061] Figure 5 This is a histogram corresponding to the second habit category of the learning habit category in the embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0063] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0064] It should be noted that in the description of the present invention, terms such as "upper" and "inner" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0065] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the term "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0066] See also Figure 1 As shown, it is a step diagram of a learning effect evaluation method based on a Logistic regression model according to an embodiment of the present invention. A learning effect evaluation method based on a Logistic regression model according to the present invention comprises:

[0067] Step S100, in response to the user completing the current course, obtaining a number of questioning moments of the user, and determining whether the user has suspected behavior of catering to the evaluation based on the time difference between each questioning moment and a preset questioning time period of the current course;

[0068] Step S200, in response to the user having suspected behavior of catering to the evaluation, obtaining the single viewing time of the user, drawing a histogram of the single viewing time, and determining the learning habit category of the user according to the number of data corresponding to each data group in the histogram;

[0069] Step S300, determining corresponding habit characteristic parameters based on different learning habit categories to calculate the dominant characteristic representation value;

[0070] The habit characteristic parameters include the total course viewing time, the number of course playback replays, the number of course playback pauses, and the single viewing time;

[0071] Step S400, adding the advantageous characteristic representation value and the preset general characteristic parameters into a Logistic regression model to calculate the probability that the user's learning performance is qualified;

[0072] The general characteristic parameters include the number of interactions the user has completed in the current course and the user's historical scores.

[0073] Specifically, the process of obtaining the user's multiple questioning moments includes:

[0074] A number of questions asked by the user during the current course playback are obtained, and the time corresponding to each question is determined as the question time.

[0075] Specifically, see Figure 2 As shown, it is a step diagram of determining the time difference between each questioning moment and the preset questioning time period of the current course according to an embodiment of the present invention. The step of determining the time difference between each questioning moment and the preset questioning time period of the current course includes:

[0076] Step S101, based on the historical data of questions asked in the current course, a number of question time periods are preset in the current course based on the playback progress time;

[0077] Step S102, filtering the time that is not within the question time period based on the question time;

[0078] Step S103, determining the minimum time difference between the time and the time point endpoint of the time period corresponding to the question time period;

[0079] Step S104: calculating a minimum time difference average value based on the minimum time difference corresponding to each moment, and determining the minimum time difference average value as the time difference.

[0080] Preferably, in an embodiment of the present invention, the time points corresponding to several questions in the historical data are obtained in advance based on the historical data of questions asked in the current course, and the time period in which the number of questions asked in a unit time exceeds a preset number threshold is determined as the question time period, wherein the unit time t c The value is based on the corresponding duration t of the current course progress bar a0 OK, c =δ×t a0 , δ is the unit time value factor, the value range of δ is [0.05, 0.1], and the time period endpoint time corresponding to the question time period is the start time and the end time of the question time period.

[0081] Specifically, see Figure 3 As shown, it is a logic flow chart of determining whether a user has suspected behavior of catering to evaluation according to an embodiment of the present invention. The process of determining whether the user has suspected behavior of catering to evaluation includes:

[0082] Comparing the time difference with a preset time difference threshold;

[0083] If the time difference is less than or equal to the time difference threshold, it is determined that the user does not have suspected behavior of catering to the evaluation;

[0084] If the time difference is greater than the time difference threshold, it is determined that the user has suspected behavior of catering to the evaluation.

[0085] Preferably, in the embodiment of the present invention, the value range of the time difference threshold is [0.02t a0 ,0.03t a0 ].

[0086] Specifically, the present invention determines whether the user has suspected behavior of catering to the evaluation through the time difference between each question moment and the preset question time period of the current course. Those skilled in the art can understand that in the actual online learning process, users' concentrated questions will appear at the locations where course knowledge points appear and where difficult points appear. If the question moment corresponding to the user's question is quite different from the time period corresponding to the concentrated questioning, and the above phenomenon exists in several of the user's questions, it can be determined that the user has not learned the course content, but is only interacting to achieve the interactive task. The present invention determines the user whose time difference is greater than the preset value as having suspected behavior of catering to the evaluation through the time difference between each question moment and the preset question time period of the current course, and multiple data avoid data randomness, thereby realizing the analysis of the user's suspicious behavior based on the user's learning process information.

[0087] Specifically, the process of drawing a histogram of single viewing duration includes:

[0088] The single viewing durations of the user are divided into a number of data groups based on a preset group distance, the duration of each data group is distributed as the horizontal axis, and the histogram is drawn with the number of data corresponding to each data group as the vertical axis.

[0089] Preferably, in an embodiment of the present invention, a maximum single viewing time and a minimum single viewing time are determined according to several single viewing times of the user, and the difference between the maximum single viewing time and the minimum single viewing time is determined, and the difference is divided into m equal parts, where m has a value range of [10,20], and the difference between the maximum value and the minimum value of each equal part is determined as the group distance.

[0090] See also Figure 4 as well as Figure 5 As shown, Figure 4 The histogram corresponding to the first habit category of the learning habit category in the embodiment of the present invention is: Figure 5 The histogram corresponding to the second habit category of the learning habit category in the embodiment of the present invention, the group interval of the histogram corresponding to the first habit category and the histogram corresponding to the second habit category of the learning habit category is 2, and the unit is min.

[0091] Specifically, the present invention obtains the single viewing time of the user and draws a histogram of the single viewing time. Those skilled in the art can understand that the learning habits of users are diverse. By drawing the single viewing time of the user for course learning as a histogram, a clear analysis of the user's learning habits can be achieved.

[0092] For more details, please refer to Figure 4 as well as Figure 5 As shown, the process of determining the user's learning habit category includes:

[0093] Obtain the data quantity corresponding to each data group in the histogram, determine the maximum value and the minimum value of the data quantity, and set the maximum value k max The minimum number of data points k min The difference is determined as the frequency difference k c , k c =k max -k min , the frequency difference k c The frequency difference threshold k c0 Make a comparison;

[0094] If the frequency difference k c Less than or equal to the frequency difference threshold k c0 , it is determined that the learning habit category of the user is the first habit category;

[0095] If the frequency difference k c Greater than the frequency difference threshold k c0 , it is determined that the user's learning habit category is the second habit category.

[0096] Preferably, in the embodiment of the present invention, the frequency difference threshold k c0 Based on the total number of data in the histogram k all OK, k c0 =μ×k all , μ is the frequency difference threshold factor, and the value range of μ is [0.2, 0.25].

[0097] Specifically, the present invention determines the learning habit category of the user according to the data quantity corresponding to each data group in the histogram. In the actual user learning habit analysis process, some users prefer a stable learning duration for each study, while some users prefer irregular learning in fragmented time. The present invention calculates the difference between the maximum data quantity and the minimum data quantity in the data quantity corresponding to each data group. The smaller the difference, the more it indicates that the user's learning preference is a learning method with irregular duration, and the larger the difference, the more it indicates that the user's learning preference is a learning method with regular duration. By dividing the users' learning habit categories, it is convenient to determine targeted feature parameters for people with different learning habits.

[0098] Specifically, the process of determining corresponding habit characteristic parameters based on different learning habit categories includes:

[0099] If the user's learning habit category is the first habit category, determining the corresponding habit characteristic parameters includes the total course viewing time, the number of course playback playbacks, and the number of course playback pauses;

[0100] If the user's learning habit category is the second habit category, the corresponding habit characteristic parameters are determined to include the number of times the course playback is replayed, the number of times the course playback is paused, and the single viewing duration.

[0101] Specifically, the process of calculating the dominant feature representation value based on different habit feature parameters includes:

[0102] Calculate the dominant characteristic representation value corresponding to the first habit category according to formula (1);

[0103] (1);

[0104] Among them, X a is the dominant characteristic representation value corresponding to the first habit category, t a is the total viewing time of the course, t a0 is the corresponding duration of the current course progress bar, n1 is the number of times the course is played back, n2 is the number of times the course is paused, n0 is the reference total number of preset course playback and course pause times, λ is the weight coefficient of the total course viewing time, η is the weight coefficient of the number of times the course is played back and the number of times the course is paused, λ+η=1, and e is a constant.

[0105] Preferably, in an embodiment of the invention, the reference total number n0 of course playback playback times and course playback pause times is based on the average of the total number of course playback playback times and course playback pause times of each user in the historical data of the current course, and the average is determined as the reference total number n0 of course playback playback times and course playback pause times.

[0106] The weight coefficient λ of the total course viewing time and the weight coefficient η of the number of course playback replays and the number of course playback pauses are selected by technical personnel in this field according to the degree of influence of the total course viewing time, the number of course playback replays and the number of course playback pauses in historical data. Preferably, the weight coefficient λ of the total course viewing time can be set to 0.6, and the weight coefficient η of the number of course playback replays and the number of course playback pauses can be set to 0.4.

[0107] Specifically, the present invention does not limit the method for obtaining the number of times the course is played back and the number of times the course is paused. Preferably, in an embodiment of the present invention, an event listener can be used in the background of the course playback to capture the number of times the user plays back and the number of times the playback is paused during the course playback. Different programming languages ​​can be selected to monitor the player's progress bar stop and move events, and the progress bar stop and move events can be counted. In this way, the number of times the user's course is played back and the number of times the course is paused can be known. For example, Javascript can be used to determine the monitoring processing of events such as course playback, pause, and progress bar movement. This technology is widely used in related fields of obtaining video content and viewing behavior, and will not be repeated here.

[0108] Calculate the dominant characteristic representation value corresponding to the second habit category according to formula (2);

[0109] (2);

[0110] Among them, X b is the dominant characteristic representation value corresponding to the second habit category, t b is the single viewing duration, t b0 is the preset reference value of single viewing time, n1 is the number of times the course is played back, n2 is the number of times the course is paused, n0 is the reference total number of preset times of course playback playback and course playback pause, γ is the weight coefficient of single viewing time, η is the weight coefficient of the number of times the course is played back and the number of times the course is paused, γ+η=1, and e is a constant.

[0111] Preferably, in the embodiment of the invention, the single viewing time reference value t b0 Based on the corresponding duration t of the current course progress bar a0 OK, b0 =ε×t a0 , ε is the reference value factor of single viewing duration, and the value range of ε is [0.5,1].

[0112] The weight coefficient γ of a single viewing time and the weight coefficient η of the number of times the course is played back and the number of times the course is paused are selected by technical personnel in this field according to the degree of influence of a single viewing time, the number of times the course is played back and the number of times the course is paused in historical data. Preferably, the weight coefficient γ of the total course viewing time can be set to 0.6, and the weight coefficient η of the number of times the course is played back and the number of times the course is paused can be set to 0.4.

[0113] Specifically, the present invention determines user characteristics for users whose learning preference is irregular in duration based on the total course viewing duration, the number of course playback replays, and the number of course playback pauses. Those skilled in the art can understand that due to the fragmented time utilization of the learning method with irregular duration, it is difficult for the learned knowledge points to establish mutual associations in incoherent learning. The longer the total course viewing duration, the deeper the memory of the knowledge points of the fragmented learning duration, and the better the learning effect pursued by the user. Therefore, the total course viewing duration can reduce the suspicion that users with irregular duration have suspected behaviors of catering to evaluations. Similarly, the more the number of course playback replays and the number of course playback pauses, the more frequent the phenomenon of users learning and understanding the key points and difficult points of the course. Therefore, reducing the number of course playback replays and the number of course playback pauses can reduce the suspicion that users with irregular duration have suspected behaviors of catering to evaluations. Furthermore, it is achieved to extract data with strong representation for suspicious users and add them to model calculation and analysis to avoid misjudgment of users' suspicious behaviors and improve the accuracy of model evaluation.

[0114] Specifically, the present invention determines user characteristics based on the number of course playback replays, the number of course playback pauses, and the single viewing time for users whose learning preference is a stable learning time. Those skilled in the art can understand that for a learning method with a stable learning time, the longer the single viewing time, the easier it is for the learned knowledge points to establish mutual associations within a coherent time, which represents the better learning effect pursued by the user. Therefore, the single viewing time can reduce the suspicion that users with a stable learning time have suspected behaviors that cater to the evaluation. Similarly, the more the number of course playback replays and the number of course playback pauses, the more frequent the phenomenon of users learning and understanding the key points and difficult points of the course. Therefore, reducing the number of course playback replays and the number of course playback pauses can reduce the suspicion that users with a stable learning time have suspected behaviors that cater to the evaluation. Furthermore, it is achieved that the data with strong representation is extracted for suspicious users and added to the model calculation and analysis to avoid misjudgment of users' suspicious behaviors and improve the accuracy of model evaluation.

[0115] Specifically, see Figure 5 As shown, it is a system block diagram of a learning effect evaluation system based on a Logistic regression model according to an embodiment of the present invention. The process of calculating the probability of a user's learning performance being qualified includes:

[0116] Adding the dominant characteristic characterization value and the preset general characteristic parameters into the Logistic regression model (3);

[0117] ;

[0118] ;

[0119] ; (3)

[0120] Where P is the probability that the user's learning performance is qualified, X1 is the dominant feature representation value, and when the user's learning habit category is the first habit category, X1=X a , when the user's learning habit category is the second habit category, X1=X b , X2 is the number of interactions the user has completed in the current course, X2' is the preset reference value for the number of interactions, X3 is the user's historical score, X3' is the preset reference value for the user's historical score, β0, β1, β3 are logistic regression coefficients, and e is a constant.

[0121] In an embodiment of the present invention, the interaction times reference value X2' is determined based on the interaction times of each user completing the current course in the historical data of the current course, and the average interaction times of users completing the current course is determined as the interaction times reference value X2'. The user historical score reference value X3' is determined based on the score of the current course in the historical data of the current course, and the average score of users of the current course is determined as the user historical score reference value X3'.

[0122] Preferably, by determining the logistic regression coefficients of X1, X2 / X2', and X3 / X3', the probability P that the user has passed the learning test can be obtained through the above model. The larger P is, the greater the probability that the user has passed the learning test, that is, the user does not have any suspected behavior of catering to the evaluation. Example

[0123] This paper selects the online teaching of the public elective course "Excel Advanced Application Office Technology" as the data source. This course involves 300 students from different colleges, majors and grades in the whole school, with a total of 36 hours, all of which are taught online;

[0124] First, we obtain the questioning moments of 300 students who have completed the current course, and count the number of students who have suspected behaviors of catering to the evaluation, a total of 90. We randomly select 70% (63) of them to generate the model, and the remaining 30% (27) are used for model verification. Then, we randomly select 90 students from the 210 students who do not have suspected behaviors of catering to the evaluation, and calculate X together with the students who have suspected behaviors of catering to the evaluation. 1d The average value of X1, the average value of X2 / X2' and the average value of X3 / X3' are used for regression analysis to obtain the logistic regression coefficients of X1, X2 / X2' and X3 / X3'. The probability that the user's learning performance is qualified is calculated based on the values ​​of the obtained logistic regression coefficients β0, β1, and β3.

[0125] Specifically, in practice, the logistic regression coefficient is usually determined using an iterative optimization algorithm. The gradient of the loss function can be calculated and the value of the coefficient can be updated according to the direction of the gradient until the convergence condition is reached. Common loss functions include the log-likelihood loss function or the cross-entropy loss function. Through the iterative optimization algorithm, the output of the logistic regression model is compared with the actual observation value, and the logistic regression coefficient is updated by maximum likelihood estimation, so that the prediction result of the model is closer to the actual observation value, thereby determining the logistic regression coefficient in the Logistic Regression Model. This is a commonly used technical means in determining the regression model and will not be repeated here.

[0126] Specifically, the present invention also provides a learning effect evaluation system based on a Logistic regression model, which includes:

[0127] An event recording unit, used to obtain the user's questioning moments and determine whether the user has suspected behavior of catering to the evaluation;

[0128] A drawing unit, connected to the event recording unit, for drawing a histogram of a single viewing time;

[0129] a category determination unit connected to the drawing unit, for determining the category of the user's learning habits according to the number of data corresponding to each data group in the histogram;

[0130] A feature characterization unit is connected to the drawing unit and is used to calculate the dominant feature characterization value and the probability of the user's learning performance being qualified.

[0131] Specifically, the present invention does not limit the specific structure of the feature representation unit, which can be constructed using a computing processor and will not be described in detail here.

[0132] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A learning effect evaluation method based on a Logistic regression model, characterized in that: include: In response to the user completing the current course, obtaining a number of questioning moments of the user, and determining whether the user has suspected behavior of catering to the evaluation based on the time difference between each questioning moment and a preset questioning time period of the current course; In response to the user having suspected behavior of catering to the evaluation, obtaining the single viewing time of the user, drawing a histogram of the single viewing time, and determining the learning habit category of the user according to the number of data corresponding to each data group in the histogram; Determine corresponding habit characteristic parameters based on different learning habit categories to calculate advantage characteristic representation values; The habit characteristic parameters include the total course viewing time, the number of course playback replays, the number of course playback pauses, and the single viewing time; Add the advantageous characteristic representation value and the preset general characteristic parameters into the Logistic regression model to calculate the probability of the user's passing the learning performance; The general characteristic parameters include the number of interactions that the user has completed in the current course and the user's historical scores; The process of determining the user's learning habit category includes: Obtaining the data quantity corresponding to each data group in the histogram, determining the maximum data quantity and the minimum data quantity, determining the difference between the maximum data quantity and the minimum data quantity as the frequency difference, and comparing the frequency difference with a preset frequency difference threshold; If the frequency difference is less than or equal to the frequency difference threshold, determining that the learning habit category of the user is a first habit category; If the frequency difference is greater than the frequency difference threshold, determining that the learning habit category of the user is a second habit category; The process of determining corresponding habit characteristic parameters based on different learning habit categories includes: If the user's learning habit category is the first habit category, determining the corresponding habit characteristic parameters includes the total course viewing time, the number of course playback playbacks, and the number of course playback pauses; If the user's learning habit category is the second habit category, determining the corresponding habit characteristic parameters includes the number of course playback replays, the number of course playback pauses, and the single viewing duration; The process of calculating the dominant feature representation value based on different habit feature parameters includes: Calculate the dominant characteristic representation value corresponding to the first habit category according to formula (1); （1）； Among them, X a is the dominant characteristic representation value corresponding to the first habit category, t a is the total viewing time of the course, t a0 is the corresponding duration of the current course progress bar, n1 is the number of times the course is played back, n2 is the number of times the course is paused, n0 is the reference total number of preset times of course playback and pauses, λ is the weight coefficient of the total course viewing duration, η is the weight coefficient of the number of times the course is played back and paused, and e is a constant; Calculate the dominant characteristic representation value corresponding to the second habit category according to formula (2); （2）； Among them, X b is the dominant characteristic representation value corresponding to the second habit category, t b is the single viewing duration, t b0 is the preset reference value of single viewing time, n1 is the number of times the course is played back, n2 is the number of times the course is paused, n0 is the reference total number of preset times of course playback playback and course playback pause, γ is the weight coefficient of single viewing time, η is the weight coefficient of the number of times the course is played back and the number of times the course is paused, and e is a constant.

2. The learning effect evaluation method based on the Logistic regression model according to claim 1, characterized in that: The process of obtaining the user's multiple questioning moments includes: A number of questions asked by the user during the current course playback are obtained, and the time corresponding to each question is determined as the question time.

3. The learning effect evaluation method based on the Logistic regression model according to claim 2 is characterized in that: The step of determining the time difference between each question time and the preset question time period of the current course includes: Based on the historical data of questions asked in the current course, several question time periods are preset in the current course based on the playback progress time; Filtering the moments that are not within the questioning time period based on the questioning moment; Determine the minimum time difference between the time and the time of the endpoint of the time period corresponding to the question time period; A minimum time difference average value is calculated based on the minimum time difference corresponding to each moment, and the minimum time difference average value is determined as the time difference.

4. The learning effect evaluation method based on the Logistic regression model according to claim 3 is characterized in that: The process of determining whether the user has suspected behavior of catering to the evaluation includes: Comparing the time difference with a preset time difference threshold; If the time difference is greater than the time difference threshold, it is determined that the user has suspected behavior of catering to the evaluation.

5. The learning effect evaluation method based on the Logistic regression model according to claim 4 is characterized in that: The process of drawing a histogram of single viewing duration includes: The single viewing durations of the user are divided into a number of data groups based on a preset group distance, the duration of each data group is distributed as the horizontal axis, and the histogram is drawn with the number of data corresponding to each data group as the vertical axis.

6. The learning effect evaluation method based on the Logistic regression model according to claim 5 is characterized in that: The process of calculating the probability that a user has passed the course includes: Adding the dominant characteristic characterization value and the preset general characteristic parameters into the Logistic regression model (3); ； ； ； （3） Where P is the probability that the user's learning performance is qualified, X1 is the dominant feature representation value, and when the user's learning habit category is the first habit category, X1=X a , when the user's learning habit category is the second habit category, X1=X b , X2 is the number of interactions the user has completed in the current course, X2' is the preset reference value for the number of interactions, X3 is the user's historical score, X3' is the preset reference value for the user's historical score, β0, β1, β3 are logistic regression coefficients, and e is a constant.

7. A learning effect evaluation system based on a Logistic regression model, applied to the learning effect evaluation method based on a Logistic regression model according to any one of claims 1 to 6, characterized in that: include: An event recording unit, used to obtain the user's questioning moments and determine whether the user has suspected behavior of catering to the evaluation; A drawing unit, connected to the event recording unit, for drawing a histogram of a single viewing time; a category determination unit connected to the drawing unit, for determining the category of the user's learning habits according to the number of data corresponding to each data group in the histogram; A feature characterization unit is connected to the drawing unit and is used to calculate the dominant feature characterization value and the probability of the user's learning performance being qualified.

Images