Learning effect judgment method based on artificial intelligence

Abstract

The invention provides a learning effect judgment method based on artificial intelligence, and the method comprises the steps: obtaining the actual average learning time length of a student for different types of learning contents through employing an artificial intelligence method, such as multi-task cascaded convolutional network face detection, adjusting the actual average learning time length, and determining the learning effect of the student according to the actual average learning time length. Obtaining the adjusted actual learning average time length; judging whether the adjusted actual learning average duration is within a preset learning duration range of a learning duration prediction model or not; wherein the upper limit threshold value and the lower limit threshold value of the preset learning duration range are adjusted according to different types of learning contents; and if the judgment result is that the adjusted actual learning average duration is within the preset learning duration range, considering that the student meets the learning requirement. According to the invention, the technical problem that the learning effect of the student cannot be objectively evaluated in the prior art is solved, and the method and the device can be widely applied to situations such as online teaching and online learning.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a learning effect judgment method based on artificial intelligence. Background Art

[0002] Artificial intelligence technology has been applied to all walks of life, and has also been widely used in teaching, knowledge learning, etc. When applied to different industries to solve different technical problems, artificial intelligence technology must be adaptively adjusted and changed.

[0003] Currently, there are many online learning platforms that can realize online learning, practical tests, exam evaluation and review feedback for students based on the Internet. It can integrate learning resources, improve learning and exam efficiency, realize closed-loop management, and meet the differentiated and personalized teaching and learning needs of faculty and students. Online learning (such as online education, online teaching, online training, etc.) is a relatively important way to acquire knowledge, but online learning, without the supervision of teachers, basically relies on the learners' self-consciousness, and the effect is far less than offline learning. To this end, many existing technologies use artificial intelligence technology, for example, interactive teaching of artificial intelligence.

[0004] So far, there is no method that can objectively evaluate the learning effect of online learning. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a learning effect judgment method based on artificial intelligence, the method comprising:

[0006] A multi-task cascade convolutional network face detection method is used to obtain the actual average learning time of students for different types of learning content, and the actual average learning time is adjusted to obtain the adjusted actual average learning time;

[0007] Determine whether the adjusted actual average learning time is within the predetermined learning time range of the learning time prediction model; wherein the learning time prediction model is constructed based on the LSTM deep learning algorithm, and the upper and lower limit thresholds of the predetermined learning time range of the learning time prediction model are adjusted according to different types of learning content;

[0008] If the judgment result is that the adjusted actual average learning time is within the predetermined learning time range, it is considered that the student meets the learning requirements.

[0009] Optionally, the actual average learning duration is adjusted, including:

[0010] For learning content marked as type one, if the actual average learning duration for type one is within the first predetermined duration, the actual average learning duration is adjusted downward; if the actual average learning duration is greater than the first predetermined duration, the actual average learning duration is adjusted upward;

[0011] For learning content marked as type 2 and type 3, if the actual average learning duration for type 2 or type 3 is within the second predetermined duration, the actual average learning duration is not adjusted; if the actual average learning duration is greater than the second predetermined duration, the actual average learning duration is adjusted upward;

[0012] For the learning content marked as type four, the actual average learning duration is adjusted upward.

[0013] Optionally, for the learning content marked as type 1, the adjusted actual average learning time T1'=T1*(1.5+(T1 2 -A*T1)), where T1 is the actual learning time of type one and A is the first scheduled time;

[0014] For the learning content marked as type 2, the adjusted actual average learning time T2' is:

[0015]

[0016] For the learning content marked as type three, the adjusted actual average learning time T3' is:

[0017]

[0018] Wherein, T2 is the actual learning time for the learning content of type 2, T3 is the actual learning time for the learning content of type 3, and B is the second predetermined time;

[0019] For the learning content marked as type 4, the adjusted actual average learning time T4' is:

[0020] T4'=T4*(1+T4 2 *e T4 ), where T4 is the actual learning time for type 4 learning content.

[0021] Optionally, the upper and lower thresholds of the predetermined learning time range are adjusted according to different types of learning content, specifically including:

[0022] Calculate the difference between the maximum and minimum values ​​of the adjusted actual average learning time T1', T2', T3' and T4';

[0023] Calculate the sum of the adjusted actual average learning time T1', T2', T3' and T4';

[0024] Determining a lower limit adjustment coefficient according to the difference and the sum;

[0025] The lower limit threshold of the predetermined learning time range is adjusted according to the lower limit adjustment coefficient to obtain an adjusted lower limit threshold of the predetermined learning time range;

[0026] determining a cap adjustment factor based on the sum;

[0027] The upper limit threshold of the predetermined learning time range is adjusted according to the upper limit adjustment coefficient to obtain the adjusted upper limit threshold of the predetermined learning time range.

[0028] Optionally, the adjusted scheduled learning time range

[0029] The adjusted scheduled learning time range

[0030] in,

[0031] N=4, i∈{1,2,3,4}, x1 represents the actual average learning time of the type marked as type one; x2 represents the actual average learning time of the type marked as type two; x3 represents the actual average learning time of the type marked as type three; x4 represents the actual average learning time of the type marked as type four; arctan(x1) represents the adjusted actual average learning time of the type marked as type one, arctan(x2) represents the adjusted actual average learning time of the type marked as type two, arctan(x3) represents the adjusted actual average learning time of the type marked as type three The adjusted actual average learning time, arctan(x4) represents the adjusted actual average learning time of type marked as type four; Max{arctan(xi)} represents the maximum value of the four adjusted values ​​of arctan(x1), arctan(x2), arctan(x3), and arctan(x4); min{arctan(xi)} represents the minimum value of the four adjusted values ​​of arctan(x1), arctan(x2), arctan(x3), and arctan(x4).

[0032] Optionally, the initial upper threshold and the initial lower threshold are set by counting the study time of a large number of students, or are considered to be set.

[0033] Optionally, the actual average learning time x1 for a type marked as type one> the actual average learning time x2 for a type marked as type two> the actual average learning time x3 for a type marked as type three> the actual average learning time x4 for a type marked as type four.

[0034] Optionally, x1, x2, x3 and x4 are obtained through statistics of big data, or are set by teachers for different classes based on teaching experience.

[0035] Optionally, a multi-task cascade convolutional network face detection method is used to obtain the actual average learning time of students for different types of learning content, including:

[0036] A multi-task cascade convolutional network face detection method is used to calculate the Euler angle of the face relative to the three-dimensional space;

[0037] The facial orientation is determined based on the Euler angles to judge the actual average learning time spent by students browsing different types of learning content.

[0038] Optionally, type 1 is the learning content of derivation, calculation or recitation type;

[0039] Type 2 is theoretical explanation-based learning content;

[0040] Type three is reading-aloud learning content;

[0041] Type 4 is entertainment-oriented learning content.

[0042] The present invention can be widely used in online learning, online training, online teaching, etc. Compared with the prior art, the present invention has the following advantages:

[0043] The present invention uses the artificial intelligence method of multi-task cascade convolutional network face detection, and by setting different types of learning content that can be flexibly changed, it can more objectively judge the learning effect. Since the upper and lower limit thresholds of the predetermined learning time range will be adjusted according to different types of learning content, it takes into account learning knowledge of different difficulty levels, and the judgment of whether the learning effect is good is more objective;

[0044] After obtaining the actual average learning time, the present invention will make adaptive adjustments according to different types of learning content to obtain the adjusted actual average learning time. For difficult content, the actual learning time of students will be increased, while for easy content, the actual learning time of students will be reduced, so that the judgment of learning efficiency is closer to the actual learning situation;

[0045] The present invention determines the initial values ​​of the upper and lower limit thresholds of the predetermined learning time range of the learning time prediction model more accurately by using artificial intelligence learning calculations on big data.

[0046] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart of a method for determining learning effects based on artificial intelligence according to an embodiment of the present disclosure is shown;

[0049] Figure 2 A display diagram of face detection based on artificial intelligence according to an embodiment of the present disclosure is shown;

[0050] Figure 3 A schematic diagram of an adjustment amplitude curve used for time adjustment in a macroscopic situation according to an embodiment of the present disclosure is shown;

[0051] Figure 4 A schematic diagram of an adjustment value 1 curve for type 1 in a microscopic perspective according to an embodiment of the present disclosure is shown;

[0052] Figure 5 A schematic diagram of a curve of adjustment value 4 for type 4 in a microscopic perspective according to an embodiment of the present disclosure is shown;

[0053] Figure 6 A schematic diagram of an adjustment value curve for type two or type three in a microscopic perspective according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The present invention first determines the general time range that students or trainees spend on a certain learning content by studying big data. That is, under normal circumstances, students or trainees with high learning efficiency spend less learning time (i.e., the initial lower limit threshold) and students or trainees with slightly lower learning efficiency spend more learning time (i.e., the initial upper limit threshold).

[0056] Figure 1 FIG. 1 shows a flow chart of a method for determining learning effects based on artificial intelligence according to an embodiment of the present disclosure. Figure 1 As shown, a method for judging learning effects based on artificial intelligence includes: using a multi-task cascade convolutional network face detection method to obtain the actual average learning time of students for different types of learning content, adjusting the actual average learning time to obtain the adjusted actual average learning time; judging whether the adjusted actual average learning time is within the predetermined learning time range; wherein the upper and lower limit thresholds of the predetermined learning time range are adjusted according to different types of learning content; if the judgment result is that the adjusted actual average learning time is within the predetermined learning time range, it is considered that the student meets the learning requirements. Figure 2 A display diagram of face detection based on artificial intelligence is shown. Artificial intelligence detection methods such as multi-task cascade convolutional network face detection can be used to obtain the learning time.

[0057] In the scenarios of online education, online teaching, online training, etc., which realize the purpose of online learning through the Internet, in the actual learning process, it takes a certain amount of time to learn different contents, and different students spend different time due to their learning foundation, learning habits, etc., so the present invention first sets a general learning time range for a certain learning topic, that is, a predetermined learning time range. Generally speaking, within the learning time range, it is believed that the learning effect is relatively good; and beyond the learning time range, for example, below the minimum value of the learning time range, there may be a cursory browsing, and above the maximum value of the learning time range, there may be distraction, etc., and the learning effect is considered to be average.

[0058] In an embodiment of the present invention, for the upper and lower thresholds of the predetermined learning time of the learning time prediction model, the historical Internet data of the student and the historical location information of the student are obtained, and the historical Internet data of the student is compared with the preset database to obtain the historical daily behavior data of the student. The historical daily behavior data includes: the number of times the student accessed learning videos of different contents, the type of Internet device, and the number of times the student accessed third-party applications, wherein the preset database includes at least one of the following: data information of learning videos of different contents, data information of Internet device types, and data information of third-party applications.

[0059] The stay time of the student in the learning videos with different contents is determined according to the historical position information, and the stay time of the student in the learning videos with different contents is determined as the historical learning time of the student.

[0060] The LSTM deep learning algorithm is used to construct a learning time prediction model based on the student's historical daily behavior data and the student's historical learning time. When the LSTM deep learning algorithm is used to construct a learning time prediction model, the historical learning time can be used as indicator information when training the LSTM deep learning algorithm, and the historical daily behavior data of other dimensions can be used as feature dimension information when training the LSTM deep learning algorithm. The LSTM deep learning algorithm is trained based on the student's historical daily behavior data and the student's historical learning time to obtain a learning time prediction model.

[0061] As we all know, LSTM, or Long Short Term, is a type of recurrent neural network RNN ​​that can learn long-term dependency information. All RNNs have a chained form of repeated neural network modules. In standard RNNs, this repeated module has only a very simple structure, such as a tanh layer. The repeated module in standard RNNs contains a single layer, and LSTM has the same structure, but the repeated module has a different structure. Unlike a single neural network layer, LSTM has four. By adding complex logic gates to control backward transmission or discarding, the long-term dependency problem of ordinary RNN algorithms is avoided, and the real learning ability of the algorithm is improved. Through improvements, the LSTM time prediction algorithm has the following characteristics compared with ordinary classification or regression algorithms: 1. Like RNN, it solves the dependency problem of traditional neural network prediction algorithms, adds historical dependency information in the time dimension, and enables the algorithm to have time series prediction capabilities; 2. It enhances the shortcomings of the RNN algorithm in solving dependency problems, and improves the learning ability of the algorithm by adding complex logic control gates; 3. Due to the characteristics of deep learning itself, it is different from classic linear regression and classification algorithms, and the complexity of the algorithm is improved. More regular information can be obtained from the data, and the accuracy of the algorithm prediction can be improved. By using the LSTM time series model method, we can fully utilize the multi-dimensionality of the data and add the time series information to make the prediction effect of the learning duration prediction model more accurate.

[0062] The learning time prediction model is used to predict the learning time of a large number of students (eg, 30, 100 or more).

[0063] The minimum value among the predicted learning time of a large number of students is used as the lower limit threshold of the scheduled learning time of the learning time prediction model, and the maximum value among the predicted learning time of a large number of students is used as the upper limit threshold of the scheduled learning time of the learning time prediction model. Through this embodiment, the upper limit threshold and the lower limit threshold of the scheduled learning time are made more reasonable.

[0064] In some embodiments, a preliminary predetermined learning time range (i.e., an upper threshold of the predetermined learning time and a lower threshold of the predetermined learning time) is set by counting the learning time of a large number of students (e.g., 30, 100 or more) from teachers, learning content designers, etc. Of course, the preliminary learning time range (i.e., the upper threshold of the predetermined learning time and the lower threshold of the predetermined learning time) can also be directly set by the teacher or learning content designer. The learning time can be set as follows:

[0065] Initial lower threshold 1 < learning time < initial upper threshold 2.

[0066] The above-mentioned study time settings are fixed and do not take into account the actual conditions of different learning groups, such as whether they are in a new learning stage or a review stage, and whether they have a better foundation or a weaker foundation.

[0067] The present invention fully considers the different actual situations of different groups and different types of learning content, and adjusts the scheduled learning time range, that is, adjusts the initial lower threshold 1 and the initial upper threshold 2. By adjusting the upper threshold and the lower threshold, the scheduled learning time range for the subject is adjusted.

[0068] In the embodiment of the present invention, different types of learning content are set under the learning theme, and the types of learning content include:

[0069] Type 1: Derivation, calculation or recitation type. For this type of learning content, students need to spend time thinking and taking practical actions during the learning process, so it often takes a long time;

[0070] Type 2: Theoretical explanation type. For this type of learning content, it does not take a long time to deduce and calculate, but it also requires time to understand and think, so it usually takes less time than type 1, but it also takes a little longer;

[0071] Type 3: Reading aloud. For this type of learning content, students need to read aloud according to the text. Unlike Type 1 and Type 2, it does not require a long time of deduction, understanding and thinking. It only needs to read aloud, so the time spent is generally less than Type 2, so it takes a little less time;

[0072] Type 4: Entertainment type. This type of learning content usually adjusts learning pressure through short videos. It does not take a long time, and the actual learning process cannot take up too much time for students. It takes less time than type three.

[0073] From the above-mentioned types of learning content, the time spent by students on learning each type of learning content is as follows: the actual average learning time spent on type 1 x1> the actual average learning time spent on type 2 x2> the actual average learning time spent on type 3 x3> the actual average learning time spent on type 4 x4. Therefore, different types of learning content take different amounts of time.

[0074] Therefore, the present invention considers the existence of the above-mentioned different learning types of content in a learning topic, preliminarily sets the initial upper threshold and the initial lower threshold corresponding to the learning topic, and adjusts the initial upper threshold and the initial lower threshold to achieve adjustment of the preliminarily set learning time range.

[0075] Since the lower limit threshold of the scheduled learning time reflects that less time is spent on the content, this is generally the time spent by students with a good learning foundation. In actual situations, the number of students with a good foundation is small, so the adjustment of this lower limit threshold should not be too large. If the adjustment is too large, some students with a good foundation may be excluded. Therefore, for the adjustment coefficient of the lower limit threshold 1, the present invention determines the adjustment coefficient by the following formula:

[0076]

[0077] Wherein, N=4, i∈{1,2,3,4}, x1 represents the actual average learning time of the type marked as type one; x2 represents the actual average learning time of the type marked as type two; x3 represents the actual average learning time of the type marked as type three; x4 represents the actual average learning time of the type marked as type four; arctan(x1) represents the adjusted actual average learning time of the type marked as type one, arctan(x2) represents the adjusted actual average learning time of the type marked as type two, arctan(x3) represents the adjusted actual average learning time of the type marked as type four. The adjusted actual average learning time of type three, arctan(x4) represents the adjusted actual average learning time of type marked as type four; Max{arctan(xi)} represents the maximum value of the four adjustment values ​​of arctan(x1), arctan(x2), arctan(x3), and arctan(x4); min{arctan(xi)} represents the minimum value of the four adjustment values ​​of arctan(x1), arctan(x2), arctan(x3), and arctan(x4). The actual average learning time of each type of learning content can be counted through big data, can be set by teachers based on teaching experience, and can be personalized for different classes, all of which are covered within the scope of the present invention. It should be noted that the present invention is not limited to the above four types, but can be more than the above types according to actual conditions.

[0078] In the embodiment of the present invention, the initial lower limit threshold 1 is adjusted by the lower limit threshold adjustment coefficient based on the initial lower limit threshold 1:

[0079] Adjusted scheduled study time range

[0080] Through the lower threshold adjustment coefficient, it can be seen that the initial lower threshold 1 has changed, but since the adjustment is not large, it can cover students with a good foundation. The advantage of this lower threshold adjustment is that it takes into account different learning contents.

[0081] The upper limit threshold of the predetermined learning time range is mainly for students with weak learning foundation. Some students, especially those in primary and secondary schools, have relatively poor self-discipline and generally procrastinate and extend their learning time. Such students should not be absolutely excluded from learning. Therefore, the initial upper limit threshold 2 is adjusted significantly in the embodiment of the present invention. Therefore, the adjustment coefficient of the initial upper limit threshold 2 is determined by the following formula:

[0082]

[0083] like Figure 3The schematic diagram of the adjustment amplitude curve used for time adjustment in the macroscopic case according to the embodiment of the present disclosure is shown. The arctan(x) function is a monotonically increasing function, and its maximum value is π / 2 (since xi is not a negative value, its negative part is not considered in the present invention); therefore, for each type of learning content (especially type one), even if the average learning time is large, it will not lead to an infinite upper threshold adjustment coefficient. This design of the present invention not only takes into account the increase of the upper threshold with the average learning time, but also takes into account the situation of avoiding an infinite increase in the upper threshold.

[0084] At this time, the adjusted scheduled learning time range

[0085] In the embodiment of the present invention, the adjustment of the upper and lower thresholds is actually implemented to adjust the learning time. Since the adjustment of the upper and lower thresholds in the present invention takes into account different learning contents, that is, the different time spent in different learning rooms, it is more flexible and more meets the actual learning requirements compared to a fixed learning time.

[0086] For the actual average learning time of different types of learning content, the MTCNN (multi-task cascade convolutional network) face detection method can be used to realize the classification of faces and non-faces in the images taken by students, face bounding box regression and related coordinate acquisition, and face feature point acquisition. According to the face feature point data, the Euler angles of the face relative to the three-dimensional space are calculated, including pitch angle, yaw angle, and roll angle. The angle with the screen is calculated to determine the face direction. The time spent facing the screen is determined as the actual learning time. By obtaining multiple actual learning times, the actual average learning time spent by students browsing related content can be determined.

[0087] In another embodiment, the server detects faces within the video duration corresponding to different types of learning content, and calculates the center coordinates of the eyes and mouth, and then intercepts the three areas of the eyes and mouth, and inputs the face area image into the convolutional neural network in the eye attention recognition model (for example, VGG16), and inputs the eye and mouth area images into another convolutional neural network to obtain four one-dimensional feature vectors, obtains the weight coefficients corresponding to the global face image and the three local area images, and multiplies the feature vectors with the weight coefficients for feature fusion, and finally inputs the fully connected layer for feature dimensionality reduction, and activates through the Softmax activation function to obtain the actual average learning duration of different types of learning content. Among them, the eye attention recognition model is trained based on a pre-set neural network model.

[0088] In some embodiments, the learning time of multiple students in a class or at a certain basic level is collected to calculate the actual average learning time of different types of learning content, which also falls within the scope of the present invention. This design of the present invention also takes into account different learning groups and is more suitable for the actual situation of different learning groups (such as students and classes with different foundations).

[0089] The above adjustment of the actual average learning time is made from a macro perspective, that is, an adjustment for the student group. This adjustment method takes into account the overall situation of the group. On the basis of the above adjustment, the present invention can further make adjustments from a micro perspective, that is, adjust the specific learning time for each specific student.

[0090] Each student will take a certain amount of time to learn different types of content in a topic, and different situations will occur for different contents due to different requirements. In the present invention, taking into account the actual situation of learning, students will have interactive requirements according to different learning contents during the learning process, such as requiring students to answer questions, students actively mark learning contents with the mouse, click and select with the mouse, etc. Therefore, the present invention does not use the actual learning browsing time, but adjusts the actual learning browsing time according to different types of learning contents (such as theoretical, derivation, entertainment, calculation, and reading).

[0091] The present invention takes into account the content of different types of learning and makes a virtual adjustment to the actual average learning time occupied. That is, after obtaining the actual average learning time of specific students, the actual average learning time of the students is not used as the final result. Instead, the actual average learning time obtained is adjusted according to the specific content of the learning, and the adjusted actual average learning time is used as the basis for judging the learning effect.

[0092] For learning content marked as type one, it involves derivation, calculation or recitation, so it generally takes more time to learn. In order to give students of this type more time margin, the present invention considers judging whether it is within the first predetermined time, within the first predetermined time A (A>0), based on its actual average learning time T1, and then adjusting the actual average learning time T1 downward (i.e. reducing the learning time) to give it more sufficient time; and if the actual average learning time T1 exceeds the first predetermined time A, distraction and the like may occur. At this time, based on the actual average learning time T1, it is adjusted upward (i.e. increasing the learning time), that is, the learning of this content cannot spend unlimited time. Based on this, the present invention adjusts the actual time spent on learning type one in the following ways:

[0093] The adjusted actual average learning time for the type marked as type 1 is T1'=T1*(1.5+adjustment value 1)=T1*(1.5+(T1 2 -A*T1)).

[0094] Figure 4 A schematic diagram of the adjustment value 1 curve for type one in the microscopic perspective according to an embodiment of the present disclosure is shown. As shown in the figure, the adjustment is downward within the first predetermined duration A, and the adjustment is upward beyond the duration A. In an embodiment of the present invention, for the first predetermined duration A, it can be set according to the difficulty of the type one learning content. For example, if it is set to 0.5, when the actual average learning duration is less than 0.5, the actual average learning duration needs to be adjusted downward, and when the actual average learning duration T1 = 0.25 hours, the downward adjustment time is the largest, and when it exceeds 0.5, it needs to be adjusted upward.

[0095] For the learning content marked as type 4, it is entertainment-type. Generally speaking, it is necessary to avoid spending too much time on this content. Therefore, the present invention uses an upward adjustment method for the actual average learning time of this type from the beginning, and the adjustment range is large. Based on this, the present invention adjusts the actual time T4 spent on learning type 4 in the following way:

[0096] The adjusted actual average learning time of type 4 is T4'=T4*(1+adjustment value 4)=T4*(1+T4 2 *e T4 ).

[0097] Figure 5 FIG. 4 shows a schematic diagram of a curve of adjustment value 4 for type 4 in a microscopic perspective according to an embodiment of the present disclosure. Figure 5 It can be seen that the upward adjustment of the actual average learning time spent on type 4 varies greatly. This design of the present invention can effectively reduce the time spent by students on the content.

[0098] For the theoretical explanation and reading types of type 2 and type 3, no adjustment is required within the second predetermined time length B (B>0), and when it exceeds the second predetermined time length B, a slow upward adjustment is performed. In the embodiment of the present invention, the second predetermined time length B can be flexibly set according to the difficulty of the learning content. Based on this, the present invention adjusts the actual time length T2 and T3 spent on learning type 2 and type 3 in the following manner:

[0099] The adjusted actual average learning time for the type marked as type 2:

[0100]

[0101] The adjusted actual average learning time for the type marked as type 3:

[0102]

[0103] Figure 6 FIG. 2 shows a schematic diagram of an adjustment value curve for type 2 or type 3 in a microscopic perspective according to an embodiment of the present disclosure. Figure 6 It can be seen that when the actual average learning time is less than the second predetermined time B, the actual average learning time is not adjusted, and when the actual average learning time is greater than the second predetermined time B, the actual average learning time is adjusted upward.

[0104] It should be noted that, in the embodiment of the present invention, the first predetermined duration A and the second predetermined duration B are not related, and may be equal or unequal, and the present invention is not limited to this.

[0105] In an embodiment of the present invention, the actual average learning time of students for different types of learning content is obtained, and the MTCNN (multi-task cascaded convolutional network) face detection method is adopted to realize the classification of faces and non-faces in the images taken by students, face frame regression and related coordinate acquisition, and face feature point acquisition. According to the face feature point data, the Euler angles relatively mapped in the three-dimensional space are calculated, including the pitch angle, yaw angle, and roll angle, thereby calculating the angle with the screen, determining the facial orientation, and judging the actual average learning time spent by students on browsing the relevant content.

[0106] In the present invention, corresponding type tags can be set for different contents in the learning theme. For example, the first part of the content in Topic 1 is set to the derivation, calculation or recitation type and marked as Type 1, the second part is set to the theoretical explanation type and marked as Type 2, the third part is set to the reading type and marked as Type 3, and the fourth part is set to the entertainment type and marked as Type 4. The corresponding learning content can be set to different types according to the needs of the theme setting or the teaching needs. The learning software detects the actual average learning time spent by the students on different learning types, and detects different types of tags. According to the retrieved type tags and the actual average learning time, according to the adjustment formulas corresponding to the above different types, the adjusted actual average learning time is calculated.

[0107] Finally, it is determined whether the adjusted actual average learning time is within the range of the scheduled learning time, and the learning effect of the students is determined based on the judgment result. That is, if the adjusted actual average learning time satisfies the following formula, it is considered that the students meet the learning requirements, that is, the learning effect is good; if the adjusted actual average learning time is not within the range of the following formula, it is considered that the learning effect of the students is poor.

[0108] The lower limit threshold 1 of the adjusted scheduled learning time range ≤ the adjusted actual average learning time ≤ the upper limit threshold 2 of the adjusted scheduled learning time range.

[0109] In the prior art, some online learning, in the process of students' learning, in order to prevent students from not looking at the screen or doing things unrelated to learning, regularly sends learning reminders to students on the learning screen, and students must click on the reminders. However, in actual use, these reminders may affect students' concentration, such as causing tension, interrupting learning to deal with reminders. Therefore, in the present invention, no reminders are sent to students.

[0110] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining learning effects based on artificial intelligence, the method comprising: A multi-task cascade convolutional network face detection method is used to obtain the actual average learning time of students for different types of learning content, and the actual average learning time is adjusted to obtain the adjusted actual average learning time; Determine whether the adjusted actual average learning time is within the range of the predetermined learning time of the learning time prediction model; wherein, The learning time prediction model is constructed based on the LSTM deep learning algorithm, and the upper and lower limit thresholds of the predetermined learning time range of the learning time prediction model are adjusted according to different types of learning content; If the judgment result is that the adjusted actual average learning time is within the predetermined learning time range, it is considered that the student meets the learning requirements.

2. The method according to claim 1, wherein: The actual average learning time is adjusted to include: For learning content marked as type one, if the actual average learning duration for type one is within the first predetermined duration, the actual average learning duration is adjusted downward; if the actual average learning duration is greater than the first predetermined duration, the actual average learning duration is adjusted upward; For learning contents marked as type 2 and type 3, if the actual average learning duration for type 2 or type 3 is within the second predetermined duration, the actual average learning duration is not adjusted; If the actual average learning duration is greater than the second predetermined duration, the actual average learning duration is adjusted upward; For the learning content marked as type four, the actual average learning duration is adjusted upward.

3. The method according to claim 2, wherein: For the learning content marked as type 1, the adjusted actual average learning time T1' = T1*(1.5+(T1 2 -A*T1)), where T1 is the actual learning time of type one and A is the first scheduled time; For the learning content marked as type 2, the adjusted actual average learning time T2' is: For the learning content marked as type three, the adjusted actual average learning time T3' is: Wherein, T2 is the actual learning time for the learning content of type 2, T3 is the actual learning time for the learning content of type 3, and B is the second predetermined time; For the learning content marked as type 4, the adjusted actual average learning time T4' is: T4'=T4*(1+T4 2 *e T4 ), where T4 is the actual learning time for type 4 learning content.

4. The method according to claim 3, wherein: The upper and lower thresholds of the predetermined learning time range are adjusted according to different types of learning content, specifically including: Calculate the difference between the maximum and minimum values ​​of the adjusted actual average learning time T1', T2', T3' and T4'; Calculate the sum of the adjusted actual average learning time T1', T2', T3' and T4'; Determining a lower limit adjustment coefficient according to the difference and the sum; The lower limit threshold of the predetermined learning time range is adjusted according to the lower limit adjustment coefficient to obtain an adjusted lower limit threshold of the predetermined learning time range; determining a cap adjustment factor based on the sum; The upper limit threshold of the predetermined learning time range is adjusted according to the upper limit adjustment coefficient to obtain the adjusted upper limit threshold of the predetermined learning time range.

5. The method according to claim 4, wherein: The lower limit threshold of the adjusted scheduled learning time range = initial lower limit threshold × The upper limit threshold of the adjusted scheduled learning time range = initial upper limit threshold × in, N=4, i∈{1,2,3,4}, x1 represents the actual average learning time of the type marked as type one; x2 represents the actual average learning time of the type marked as type two; x3 represents the actual average learning time of the type marked as type three; x4 represents the actual average learning time of the type marked as type four; arctan(x1) represents the adjusted actual average learning time of the type marked as type one, arctan(x2) represents the adjusted actual average learning time of the type marked as type two, arctan(x3) represents the adjusted actual average learning time of the type marked as type three The adjusted actual average learning time, arctan(x4) represents the adjusted actual average learning time of type marked as type four; Max{arctan(xi)} represents the maximum value of the four adjusted values ​​of arctan(x1), arctan(x2), arctan(x3), and arctan(x4); min{arctan(xi)} represents the minimum value of the four adjusted values ​​of arctan(x1), arctan(x2), arctan(x3), and arctan(x4).

6. The method according to claim 5, wherein: The initial upper threshold and the initial lower threshold are set by counting the study time of a large number of students, or are considered to be set.

7. The method according to claim 5, wherein: The actual average learning time x1 for types marked as type one> the actual average learning time x2 for types marked as type two> the actual average learning time x3 for types marked as type three> the actual average learning time x4 for types marked as type four.

8. The method according to claim 5 or 7, wherein: x1, x2, x3 and x4 are obtained through statistics of big data, or are set by teachers for different classes based on teaching experience.

9. The method according to claim 1, wherein: The multi-task cascade convolutional network face detection method is used to obtain the actual average learning time of students for different types of learning content, including: A multi-task cascade convolutional network face detection method is used to calculate the Euler angle of the face relative to the three-dimensional space; The facial orientation is determined based on the Euler angles to judge the actual average learning time spent by students browsing different types of learning content.

10. The method according to claim 2 or 5, wherein: Type 1 is the learning content of derivation, calculation or recitation type; Type 2 is theoretical explanation-based learning content; Type three is reading-aloud learning content; Type 4 is entertainment-oriented learning content.

Images