A network live teaching management system and method based on artificial intelligence

Abstract

The present invention relates to the field of live teaching management, and specifically to an artificial intelligence-based network live teaching management system and method, comprising: an intelligent classification module, an impact analysis module, a channel planning module, a picture adjustment module and a live evaluation module, wherein the intelligent classification module is used to classify live broadcasts according to influencing factors, the impact analysis module is used to generate a joint impact function of click volume with each live broadcast factor, the channel planning module is used to format live broadcast channels and pre-allocate communication resources, the picture adjustment module is used to adjust the quality of live broadcast pictures and the insertion ratio of video recordings, and the live broadcast evaluation module is used to output live broadcast effect evaluation results. The present invention can improve the quality of live broadcasts, realize more accurate resource allocation for live broadcast content, and at the same time help anchors and operation teams understand the performance effects of live broadcasts in many aspects, thereby better developing and developing live broadcast services.

Description

Technical Field

[0001] The present invention relates to the field of live teaching management, and in particular to an artificial intelligence-based network live teaching management system and method. Background Art

[0002] Live streaming refers to the real-time transmission of audio and video content over the Internet, allowing users to watch live content such as sports events, concerts, lectures, games, etc. anywhere there is an Internet connection. Live streaming usually requires the use of webcams, microphones and other equipment to capture live content, which is then encoded, transmitted and played through special live streaming software or platforms. In addition, it is necessary to train experienced anchors to enrich the live content in order to achieve a popular live streaming effect.

[0003] Since the live broadcast is broadcast through the platform, and the communication resources between the platform and the live broadcast site are limited, different communication resources often need to be allocated for various types of live broadcasts on the platform to meet the real-time and picture quality requirements of the live broadcast. The platform's allocation of resources is usually based on artificial rules, which is not conducive to maximizing the live broadcast effect of the platform.

[0004] In addition, each anchor's style and live broadcast focus are different. For some new anchors, teaching and training are often required to find the most suitable live broadcast style. Commonly used teaching software cannot evaluate the anchor's specific live broadcast situation, resulting in low teaching reference value. Summary of the invention

[0005] The purpose of the present invention is to provide an artificial intelligence-based online live teaching management system and method to solve the problems raised in the above-mentioned background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an artificial intelligence-based online live teaching management system, comprising: an intelligent classification module, an impact analysis module, a channel planning module, a picture adjustment module and a live evaluation module;

[0007] The intelligent classification module is composed of a live broadcast database, a training library and a neural network model. The live broadcast database is used to record the historical live broadcast data of the platform and the influencing factors corresponding to the live broadcast. The training library is used to classify the influencing factors corresponding to the live broadcast and store them in a separate classification database to form an artificial intelligence network training data set. The neural network model is used to learn the input and output data in the training library and classify the live broadcast according to the influencing factors.

[0008] The influence analysis module is used to separate the training libraries of each type of live broadcast according to the live broadcast classification results, and for each record in the live broadcast database, the live broadcast click volume is used as the output variable and each influencing factor is used as the input variable, and a correlation model between the live broadcast factor and the click volume is constructed by using a neural network. After the live broadcast factor is changed, a joint influence function of the click volume with each factor is generated according to the live broadcast click drop rate;

[0009] The channel planning module is used to pre-record live video, upload the live video to the local server, format the live channel, calculate the second function between the quality of the live picture and the channel capacity, predict the proportion of communication resources used by each live broadcast according to the average picture quality of each type of live broadcast and the future live broadcast arrangement, and pre-allocate network resources;

[0010] The picture adjustment module is used to calculate the maximum convergence value of the click volume according to the joint influence function under the condition that the live broadcast bandwidth is less than the total channel bandwidth, adjust the quality of the live broadcast picture and the insertion ratio of the live broadcast video according to the influence factor corresponding to the maximum convergence value of the click volume, and count the click volume before and after the adjustment;

[0011] The live broadcast evaluation module is used to obtain the live broadcast data of the current time after the live broadcast ends, generate an influence function of the click volume of the current live broadcast as the factors change during the live broadcast process, compare the current influence function with the joint influence function, obtain the deviation value of each live broadcast factor, output the live broadcast effect evaluation record according to the deviation value, and give suggestions for adjusting the live broadcast factors.

[0012] Furthermore, the intelligent classification module includes: a network training unit and a type output unit;

[0013] The network training unit is used to record influencing factors in the platform's historical live broadcast process in the database, and the influencing factors include: live broadcast tags, number of interactions, picture quality, video stability and the proportion of real people appearing on the screen;

[0014] The type output unit is used to separately store the influencing factors of each live broadcast, generate a training data set, use a neural network model to learn the training library, and use the learning results to classify the live broadcast.

[0015] Further, the influence analysis module includes: a click association unit, a factor analysis unit and a joint influence unit;

[0016] The click association unit is used to construct a database of various types of live broadcasts according to the classification results, with the click volume as the output variable and the influencing factors as the input variables;

[0017] The factor analysis unit is used to construct a correlation model between each live broadcast element and the click volume, and after the live broadcast element is changed, the output of the evaluated click volume appears;

[0018] The joint influence unit is used to output the joint influence function of the click volume with each element according to the influence of each live broadcast element on the click volume.

[0019] Further, the channel planning module includes: a pre-upload unit and a bandwidth allocation unit;

[0020] The pre-upload unit is used to record live video, upload the live video to the platform server in advance, and realize screen replacement locally;

[0021] The bandwidth allocation unit is used to format the platform transmission channel and pre-allocate network communication resources for various live broadcasts.

[0022] Furthermore, the picture adjustment module includes: an effect convergence unit and a video recording supplement unit;

[0023] The effect convergence unit is used to calculate the size of each live broadcast influencing factor when the click volume is the highest according to the joint influence function under the constraint that the transmission bandwidth is less than or equal to the total bandwidth;

[0024] The video recording supplement unit adjusts the live broadcast image quality in real time according to the calculated live broadcast influencing factors, and replaces the live broadcast image with the video recording in the server.

[0025] Further, the live broadcast evaluation module includes: a digital fitting unit, a deviation recording unit and a report output unit;

[0026] The digital fitting unit is used to generate an influence function of the live broadcast after the live broadcast ends, which is recorded as the third function;

[0027] The deviation recording unit is used to calculate the deviation between the third function and the joint influence function on different influence factors;

[0028] The report output unit is used to generate a live broadcast evaluation report according to the deviation of each influencing factor, output the report and give live broadcast suggestions for the live broadcast at that time.

[0029] A network live teaching management method based on artificial intelligence includes the following steps:

[0030] Step S1. Record the platform's historical live broadcast data in the database, use the neural network model to learn the historical live broadcast data, and classify the live broadcast according to various influencing factors during the live broadcast process;

[0031] Step S2. For each type of live broadcast, the live broadcast click volume is used as the output variable, and each influencing factor is used as the input variable. The association model between each live broadcast factor and the click volume is fitted using the model building method to obtain the joint influence function of the click volume changing with the influencing factor;

[0032] Step S3. Upload the recorded live video to the platform local server in advance, calculate the average channel capacity occupied by various types of live broadcasts, and pre-allocate network communication resources for each live broadcast according to the live broadcast arrangement of the next period;

[0033] Step S4. Under the constraint that the transmission bandwidth is less than or equal to the total bandwidth, the size of each live broadcast influencing factor when the click volume is the highest is calculated by the joint influence function according to the conditional constraint method, the live broadcast image quality is adjusted according to the obtained influencing factors, and the live broadcast image is replaced with the video in the server;

[0034] Step S5. After the live broadcast ends, obtain the live broadcast data of that time, generate the influence function of that time in which the number of clicks of the live broadcast changes with each influencing factor, calculate the partial derivative difference between the influence function of that time and the joint influence function under each influencing factor, and output all the obtained partial derivative differences as an evaluation report.

[0035] Further, step S1 includes:

[0036] Step S11. Obtain the historical live broadcast data of the platform, wherein the historical live broadcast data consists of the live broadcast name, live broadcast click volume and influencing factors, wherein the influencing factors include: live broadcast tag, number of interactions, picture quality, video stability and proportion of real people appearing on screen;

[0037] Step S12. Taking the influencing factors of each piece of historical live broadcast data as a training library, using the neural network model to learn the data in the training library, and the neural network model after learning is able to classify the live broadcast type according to the proportional relationship of each influencing factor;

[0038] Step S13. Store the historical live broadcast data into the database according to the classification results to form a set {W1, W2, ..., Wn}, where n represents the total number of types, and Wn represents the historical live broadcast data corresponding to the nth live broadcast type.

[0039] Further, step S2 includes:

[0040] Step S21. Based on the classification results, the click rate of the live broadcast is used as the output variable, and each influencing factor is used as the input variable, and a training library is constructed for each type of live broadcast;

[0041] Step S22. Use a model building method to fit the association model between each live broadcast element and the click volume, and output a joint influence function of the click volume changing with the influencing factors. The model building method is as follows:

[0042] S22-1. For each input and output variable of a live broadcast record in the training library, a sequence [y, x1, x2, ..., xm] is generated, where y represents the output variable corresponding to the live broadcast record, and x1, x2 ... xm represent the 1st, 2nd ... mth input variables respectively;

[0043] S22-2. Use elastic network regression to fit the loss function:

[0044]

[0045] Where LOSS is the loss function, MSE is the root mean square error of the model, a is the regularization strength of the model, n is the number of live broadcast records in the training library, wi represents the preset weight of the i-th live broadcast record in the training library, p is the proportion of the regularization term, Fi(x1,x2,…,xm) represents the functional relationship between the output variable and the input variable in the i-th live broadcast record, Represents the linear regression function of the function Fi(x1,x2,…,xm);

[0046] S22-3. Calculate the regularization strength a and the regularization term ratio p when the loss function LOSS is minimized, generate a positive definite matrix X based on the regularization strength a and the regularization term ratio p, and obtain the joint influence function F(x1,x2,…,xm), where y=F(x1,x2,…,xm)=X·[x1,x2,…,xm] T ;

[0047] Step S23. For each live broadcast type, output the joint influence function corresponding to the type.

[0048] Further, step S3 includes:

[0049] Step S31. Pre-record a video for replacing the live broadcast image, and upload the video to the local server of the platform;

[0050] Step S32. Format the server channel, input the average value of each type of live picture quality into the data generator, obtain the total communication data volume Q of the live broadcast, divide it by the live broadcast duration, and obtain the average transmission speed v of the live broadcast type, where v=Q / t0, where t0 is the average live broadcast duration;

[0051] Step S33. Get the live broadcast schedule for the next period, and allocate channel capacity for each live broadcast:

[0052]

[0053] Among them, Bc represents the pre-allocated channel capacity for the cth live broadcast in the next cycle, B0 represents the total channel capacity of the server, and t c and t j Respectively represent the duration of the cth and jth live broadcasts in the next cycle, v c and v j represents the average transmission speed of the corresponding live broadcast type between the cth and jth live broadcasts in the next cycle, and k represents the number of live broadcasts scheduled in the next cycle.

[0054] Further, step S4 includes:

[0055] Step S41. Taking the transmission bandwidth less than or equal to the total bandwidth as a constraint condition, and using the joint influence function as a linear function to perform linear programming;

[0056]

[0057] Among them, Max represents the maximum value of the objective function, St represents the constraint condition, b represents the conversion coefficient between channel capacity and picture quality, xb∈{x1,x2,…,xm}, represents the influencing factor corresponding to the picture quality, and xb z Represents the image quality of the zth live broadcast in the next cycle;

[0058] Step S42. According to the linear programming results, the values ​​of each influencing factor when the objective function takes the maximum value are obtained, and the picture quality of the live broadcast picture is adjusted according to the values ​​of the influencing factors corresponding to the picture quality. According to the values ​​of the influencing factors corresponding to the proportion of real people appearing on the screen, the live broadcast picture is replaced with the recording of the corresponding time ratio in the server.

[0059] Further, step S5 includes:

[0060] Step S51. After the live broadcast ends, the influence parameters and click volume of the live broadcast are obtained, and the influence function G (x1, x2, ..., xm) of the live broadcast data is output according to the model building method;

[0061] Step S52. Calculate the partial derivative difference of each influencing factor of the live broadcast:

[0062]

[0063] Among them, Rg represents the partial derivative of the g-th influencing factor, xg represents the g-th influencing factor, and xg0 represents the maximum value of the g-th influencing factor;

[0064] Step S53. Output an evaluation report based on the partial derivatives of each influencing factor, wherein the evaluation report can reflect the deviation of the live broadcast from the historical live broadcast of the platform under the dimensions of each influencing factor.

[0065] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0066] The present invention can utilize an artificial intelligence network to classify historical live broadcast recordings according to live broadcast tags, number of interactions, picture quality, picture stability, and proportion of real-person appearances, and obtain a correlation model between live broadcast elements and click volumes of various types of live broadcast recordings. This can help understand the audience's preferences and reactions to the content, thereby better grasping audience needs, adjusting content strategies and program arrangements, and improving live broadcast quality and audience experience.

[0067] The present invention can upload live broadcast recordings to the local server of the platform in advance, and adjust the corresponding real-person appearance ratio and picture quality according to the maximum convergence value of the calculated click volume under the condition of meeting the total bandwidth of the live broadcast channel. It can achieve better live broadcast effect under limited bandwidth, improve the live broadcast quality and live broadcast process, and realize more accurate resource allocation for live broadcast content.

[0068] The present invention can obtain live broadcast data after the live broadcast ends, generate a click volume influence function of the live broadcast, evaluate the live broadcast effect of the anchor according to the deviation value of the two influence functions, and give the anchor's factor adjustment suggestions, which can help the anchor and the operation team understand the performance of the live broadcast in terms of stability, picture clarity, sound quality, etc., so as to better develop and develop the live broadcast business. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0070] Figure 1 It is a structural schematic diagram of an artificial intelligence-based online live teaching management system of the present invention;

[0071] Figure 2 It is a schematic diagram of the steps of an artificial intelligence-based online live teaching management method of the present invention. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0073] See also Figure 1 , the present invention provides a technical solution: an artificial intelligence-based online live teaching management system, including: an intelligent classification module, an impact analysis module, a channel planning module, a picture adjustment module and a live evaluation module;

[0074] The intelligent classification module is composed of a live broadcast database, a training library and a neural network model. The live broadcast database is used to record the historical live broadcast data of the platform and the influencing factors corresponding to the live broadcast. The training library is used to classify the influencing factors corresponding to the live broadcast and store them in a separate classification database to form an artificial intelligence network training data set. The neural network model is used to learn the input and output data in the training library and classify the live broadcast according to the influencing factors.

[0075] The intelligent classification module includes: a network training unit and a type output unit;

[0076] The network training unit is used to record influencing factors in the platform's historical live broadcast process in the database, and the influencing factors include: live broadcast tags, number of interactions, picture quality, video stability and the proportion of real people appearing on the screen;

[0077] The type output unit is used to separately store the influencing factors of each live broadcast, generate a training data set, use a neural network model to learn the training library, and use the learning results to classify the live broadcast.

[0078] The influence analysis module is used to separate the training libraries of each type of live broadcast according to the live broadcast classification results, and for each record in the live broadcast database, the live broadcast click volume is used as the output variable and each influencing factor is used as the input variable, and a correlation model between the live broadcast factor and the click volume is constructed by using a neural network. After the live broadcast factor is changed, a joint influence function of the click volume with each factor is generated according to the live broadcast click drop rate;

[0079] The influence analysis module includes: a click association unit, a factor analysis unit and a joint influence unit;

[0080] The click association unit is used to construct a database of various types of live broadcasts according to the classification results, with the click volume as the output variable and the influencing factors as the input variables;

[0081] The factor analysis unit is used to construct a correlation model between each live broadcast element and the click volume, and after the live broadcast element is changed, the output of the evaluated click volume appears;

[0082] The joint influence unit is used to output the joint influence function of the click volume with each element according to the influence of each live broadcast element on the click volume.

[0083] The channel planning module is used to pre-record live video, upload the live video to the local server, format the live channel, calculate the second function between the quality of the live picture and the channel capacity, predict the proportion of communication resources used by each live broadcast according to the average picture quality of each type of live broadcast and the future live broadcast arrangement, and pre-allocate network resources;

[0084] The channel planning module includes: a pre-upload unit and a bandwidth allocation unit;

[0085] The pre-upload unit is used to record live video, upload the live video to the platform server in advance, and realize screen replacement locally;

[0086] The bandwidth allocation unit is used to format the platform transmission channel and pre-allocate network communication resources for various live broadcasts.

[0087] The picture adjustment module is used to calculate the maximum convergence value of the click volume according to the joint influence function under the condition that the live broadcast bandwidth is less than the total channel bandwidth, adjust the quality of the live broadcast picture and the insertion ratio of the live broadcast video according to the influence factor corresponding to the maximum convergence value of the click volume, and count the click volume before and after the adjustment;

[0088] The picture adjustment module includes: an effect convergence unit and a video supplement unit;

[0089] The effect convergence unit is used to calculate the size of each live broadcast influencing factor when the click volume is the highest according to the joint influence function under the constraint that the transmission bandwidth is less than or equal to the total bandwidth;

[0090] The video recording supplement unit adjusts the live broadcast image quality in real time according to the calculated live broadcast influencing factors, and replaces the live broadcast image with the video recording in the server.

[0091] The live broadcast evaluation module is used to obtain the live broadcast data of the current time after the live broadcast ends, generate an influence function of the click volume of the current live broadcast as the factors change during the live broadcast process, compare the current influence function with the joint influence function, obtain the deviation value of each live broadcast factor, output the live broadcast effect evaluation record according to the deviation value, and give suggestions for adjusting the live broadcast factors.

[0092] The live broadcast evaluation module includes: a digital fitting unit, a deviation recording unit and a report output unit;

[0093] The digital fitting unit is used to generate an influence function of the live broadcast after the live broadcast ends, which is recorded as the third function;

[0094] The deviation recording unit is used to calculate the deviation between the third function and the joint influence function on different influence factors;

[0095] The report output unit is used to generate a live broadcast evaluation report according to the deviation of each influencing factor, output the report and give live broadcast suggestions for the live broadcast at that time.

[0096] like Figure 2 As shown, a network live teaching management method based on artificial intelligence includes the following steps:

[0097] Step S1. Record the platform's historical live broadcast data in the database, use the neural network model to learn the historical live broadcast data, and classify the live broadcast according to various influencing factors during the live broadcast process;

[0098] Step S1 includes:

[0099] Step S11. Obtain the historical live broadcast data of the platform, wherein the historical live broadcast data consists of the live broadcast name, live broadcast click volume and influencing factors, wherein the influencing factors include: live broadcast tag, number of interactions, picture quality, video stability and proportion of real people appearing on screen;

[0100] Step S12. Taking the influencing factors of each piece of historical live broadcast data as a training library, using the neural network model to learn the data in the training library, and the neural network model after learning is able to classify the live broadcast type according to the proportional relationship of each influencing factor;

[0101] Step S13. Store the historical live broadcast data into the database according to the classification results to form a set {W1, W2, ..., Wn}, where n represents the total number of types, and Wn represents the historical live broadcast data corresponding to the nth live broadcast type.

[0102] Step S2. For each type of live broadcast, the live broadcast click volume is used as the output variable, and each influencing factor is used as the input variable. The association model between each live broadcast factor and the click volume is fitted using the model building method to obtain the joint influence function of the click volume changing with the influencing factor;

[0103] Step S2 includes:

[0104] Step S21. Based on the classification results, the click rate of the live broadcast is used as the output variable, and each influencing factor is used as the input variable, and a training library is constructed for each type of live broadcast;

[0105] Step S22. Use a model building method to fit the association model between each live broadcast element and the click volume, and output a joint influence function of the click volume changing with the influencing factors. The model building method is as follows:

[0106] S22-1. For each input and output variable of a live broadcast record in the training library, a sequence [y, x1, x2, ..., xm] is generated, where y represents the output variable corresponding to the live broadcast record, and x1, x2 ... xm represent the 1st, 2nd ... mth input variables respectively;

[0107] S22-2. Use elastic network regression to fit the loss function:

[0108]

[0109] Where LOSS is the loss function, MSE is the root mean square error of the model, a is the regularization strength of the model, n is the number of live broadcast records in the training library, wi represents the preset weight of the i-th live broadcast record in the training library, p is the proportion of the regularization term, Fi(x1,x2,…,xm) represents the functional relationship between the output variable and the input variable in the i-th live broadcast record, Represents the linear regression function of the function Fi(x1,x2,…,xm);

[0110] S22-3. Calculate the regularization strength a and the regularization term ratio p when the loss function LOSS is minimized, generate a positive definite matrix X based on the regularization strength a and the regularization term ratio p, and obtain the joint influence function F(x1,x2,…,xm), where y=F(x1,x2,…,xm)=X·[x1,x2,…,xm] T ;

[0111] Step S23. For each live broadcast type, output the joint influence function corresponding to the type.

[0112] Step S3. Upload the recorded live video to the platform local server in advance, calculate the average channel capacity occupied by various types of live broadcasts, and pre-allocate network communication resources for various types of live broadcasts according to the type of live broadcast in the next period;

[0113] Step S3 includes:

[0114] Step S31. Pre-record a video for replacing the live broadcast image, and upload the video to the local server of the platform;

[0115] Step S32. Format the server channel, input the average value of each type of live picture quality into the data generator, obtain the total communication data volume Q of the live broadcast, divide it by the live broadcast duration, and obtain the average transmission speed v of the live broadcast type, where v=Q / t0, where t0 is the average live broadcast duration;

[0116] Step S33. Get the live broadcast schedule for the next period, and allocate channel capacity for each live broadcast:

[0117]

[0118] Among them, Bc represents the pre-allocated channel capacity for the cth live broadcast in the next cycle, B0 represents the total channel capacity of the server, and t c and t j Respectively represent the duration of the cth and jth live broadcasts in the next cycle, v c and v j represents the average transmission speed of the corresponding live broadcast type between the cth and jth live broadcasts in the next cycle, and k represents the number of live broadcasts scheduled in the next cycle.

[0119] Step S4. Under the constraint that the transmission bandwidth is less than or equal to the total bandwidth, the size of each live broadcast influencing factor when the click volume is the highest is calculated by the joint influence function according to the conditional constraint method, the live broadcast image quality is adjusted according to the obtained influencing factors, and the live broadcast image is replaced with the video in the server;

[0120] Step S4 includes:

[0121] Step S41. Taking the transmission bandwidth less than or equal to the total bandwidth as a constraint condition, and using the joint influence function as a linear function to perform linear programming;

[0122]

[0123] Among them, Max represents the maximum value of the objective function, St represents the constraint condition, b represents the conversion coefficient between channel capacity and picture quality, xb∈{x1,x2,…,xm}, represents the influencing factor corresponding to the picture quality, and xb z Represents the image quality of the zth live broadcast in the next cycle;

[0124] Step S42. According to the linear programming results, the values ​​of each influencing factor when the objective function takes the maximum value are obtained, and the picture quality of the live broadcast picture is adjusted according to the values ​​of the influencing factors corresponding to the picture quality. According to the values ​​of the influencing factors corresponding to the proportion of real people appearing on the screen, the live broadcast picture is replaced with the recording of the corresponding time ratio in the server.

[0125] Step S5. After the live broadcast ends, obtain the live broadcast data of that time, generate the influence function of that time in which the number of clicks of the live broadcast changes with each influencing factor, calculate the partial derivative difference between the influence function of that time and the joint influence function under each influencing factor, and output all the obtained partial derivative differences as an evaluation report.

[0126] Step S5 includes:

[0127] Step S51. After the live broadcast ends, the influence parameters and click volume of the live broadcast are obtained, and the influence function G (x1, x2, ..., xm) of the live broadcast data is output according to the model building method;

[0128] Step S52. Calculate the partial derivative difference of each influencing factor of the live broadcast:

[0129]

[0130] Among them, Rg represents the partial derivative of the g-th influencing factor, xg represents the g-th influencing factor, and xg0 represents the maximum value of the g-th influencing factor;

[0131] Step S53. Output an evaluation report based on the partial derivatives of each influencing factor, wherein the evaluation report can reflect the deviation of the live broadcast from the historical live broadcast of the platform under the dimensions of each influencing factor.

[0132] Example: In the next day, the platform arranges 5 live broadcasts, and the live broadcast types are type 1, type 2, type 3, type 1 and type 3 respectively. The average transmission speeds of types 1, 2 and 3 are 10Mb / s, 20Mb / s and 30Mb / s, and the total transmission bandwidth of the platform is 90Mb. Then the pre-allocated bandwidths of the 5 live broadcasts are 9Mb, 18Mb, 27Mb, 9Mb and 27Mb respectively;

[0133] During the live broadcast, linear planning was performed based on the joint influence function of each type of live broadcast and the total bandwidth. The result showed that the picture quality and the proportion of real people appearing on screen of Live Broadcast 1 when the number of clicks was the largest were 0.84 and 80% respectively. The picture quality of Live Broadcast 1 was adjusted to 0.84, and 20% of the live broadcast process was replaced by local recording.

[0134] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0135] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. 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 network live teaching management method based on artificial intelligence, characterized in that: The method comprises the following steps: Step S1. Record the platform's historical live broadcast data in the database, use the neural network model to learn the historical live broadcast data, and classify the live broadcast according to various influencing factors during the live broadcast process; Step S2. For each type of live broadcast, the live broadcast click volume is used as the output variable, and each influencing factor is used as the input variable. The association model between each live broadcast factor and the click volume is fitted using the model building method to obtain the joint influence function of the click volume changing with the influencing factor; Step S3. Upload the recorded live video to the platform local server in advance, calculate the average channel capacity occupied by various types of live broadcasts, and pre-allocate network communication resources for each live broadcast according to the live broadcast arrangement of the next period; Step S4. Under the constraint that the transmission bandwidth is less than or equal to the total bandwidth, the size of each live broadcast influencing factor when the click volume is the highest is calculated by the joint influence function according to the conditional constraint method, the live broadcast image quality is adjusted according to the obtained influencing factors, and the live broadcast image is replaced with the video in the server; Step S5. After the live broadcast ends, obtain the live broadcast data, generate the influence function of the live broadcast click volume as the influence factors change, calculate the partial derivative difference between the influence function and the joint influence function under each influence factor, and output all the obtained partial derivative differences as an evaluation report; Step S3 includes: Step S31. Pre-record a video for replacing the live broadcast image, and upload the video to the local server of the platform; Step S32. Format the server channel, input the average value of each type of live picture quality into the data generator, obtain the total communication data volume Q of the live broadcast, divide it by the live broadcast duration, and obtain the average transmission speed v of the live broadcast type, where v=Q / t0, where t0 is the average live broadcast duration; Step S33. Get the live broadcast schedule for the next period, and allocate channel capacity for each live broadcast: Among them, Bc represents the pre-allocated channel capacity for the cth live broadcast in the next cycle, B0 represents the total channel capacity of the server, and t c and t j Respectively represent the duration of the cth and jth live broadcasts in the next cycle, v c and v j represents the average transmission speed of the cth and jth live broadcast types in the next cycle, and k represents the number of live broadcasts scheduled in the next cycle; Step S4 includes: Step S41. Taking the transmission bandwidth less than or equal to the total bandwidth as a constraint condition, and using the joint influence function as a linear function to perform linear programming; Max F(x1,x2,…,xm) Among them, Max represents the maximum value of the objective function, St represents the constraint condition, b represents the conversion coefficient between channel capacity and picture quality, xb∈{x1,x2,…,xm}, represents the influencing factor corresponding to the picture quality, and xb z Represents the image quality of the zth live broadcast in the next cycle; Step S42. According to the linear programming results, the values ​​of each influencing factor when the objective function takes the maximum value are obtained, and the picture quality of the live broadcast picture is adjusted according to the values ​​of the influencing factors corresponding to the picture quality. According to the values ​​of the influencing factors corresponding to the proportion of real people appearing on the screen, the live broadcast picture is replaced with the recording of the corresponding time ratio in the server.

2. According to the artificial intelligence-based online live teaching management method of claim 1, it is characterized by: Step S1 includes: Step S11. Obtain the historical live broadcast data of the platform, wherein the historical live broadcast data consists of the live broadcast name, live broadcast click volume and influencing factors, wherein the influencing factors include: live broadcast tag, number of interactions, picture quality, video stability and proportion of real people appearing on screen; Step S12. Taking the influencing factors of each piece of historical live broadcast data as a training library, using the neural network model to learn the data in the training library, and the neural network model after learning is able to classify the live broadcast type according to the proportional relationship of each influencing factor; Step S13. Store the historical live broadcast data into the database according to the classification results to form a set {W1, W2, ..., Wn}, where n represents the total number of types, and Wn represents the historical live broadcast data corresponding to the nth live broadcast type; Step S2 includes: Step S21. Based on the classification results, the click rate of the live broadcast is used as the output variable, and each influencing factor is used as the input variable, and a training library is constructed for each type of live broadcast; Step S22. Use a model building method to fit the association model between each live broadcast element and the click volume, and output a joint influence function of the click volume changing with the influencing factors. The model building method is as follows: S22-1. For each input and output variable of a live broadcast record in the training library, a sequence [y, x1, x2, ..., xm] is generated, where y represents the output variable corresponding to the live broadcast record, and x1, x2 ... xm represent the 1st, 2nd ... mth input variables respectively; S22-2. Use elastic network regression to fit the loss function: Where LOSS is the loss function, MSE is the root mean square error of the model, a is the regularization strength of the model, n is the number of live broadcast records in the training library, wi represents the preset weight of the i-th live broadcast record in the training library, p is the proportion of the regularization term, Fi(x1,x2,…,xm) represents the functional relationship between the output variable and the input variable in the i-th live broadcast record, Represents the linear regression function of the function Fi(x1,x2,…,xm); S22-3. Calculate the regularization strength a and the regularization term ratio p when the loss function LOSS is minimized, generate a positive definite matrix X based on the regularization strength a and the regularization term ratio p, and obtain the joint influence function F(x1,x2,…,xm), where y=F(x1,x2,…,xm)=X·[x1,x2,…,xm] T ; Step S23. For each live broadcast type, output the joint influence function corresponding to the type.

3. The method for managing online live teaching based on artificial intelligence according to claim 2 is characterized by: Step S5 includes: Step S51. After the live broadcast ends, the influence parameters and click volume of the live broadcast are obtained, and the influence function G (x1, x2, ..., xm) of the live broadcast data is output according to the model building method; Step S52. Calculate the partial derivative difference of each influencing factor of the live broadcast: Among them, Rg represents the partial derivative of the g-th influencing factor, xg represents the g-th influencing factor, and xg0 represents the maximum value of the g-th influencing factor; Step S53. Output an evaluation report based on the partial derivatives of each influencing factor, wherein the evaluation report can reflect the deviation of the live broadcast from the historical live broadcast of the platform under the dimensions of each influencing factor.

4. An artificial intelligence-based online live teaching management system, characterized in that: The system includes the following modules: intelligent classification module, impact analysis module, channel planning module, picture adjustment module and live broadcast evaluation module; The intelligent classification module is composed of a live broadcast database, a training library and a neural network model. The live broadcast database is used to record the historical live broadcast data of the platform and the influencing factors corresponding to the live broadcast. The training library is used to classify the influencing factors corresponding to the live broadcast and store them in a separate classification database to form an artificial intelligence network training data set. The neural network model is used to learn the input and output data in the training library and classify the live broadcast according to the influencing factors. The influence analysis module is used to separate the training libraries of each type of live broadcast according to the live broadcast classification results, and for each record in the live broadcast database, the live broadcast click volume is used as the output variable and each influencing factor is used as the input variable, and a correlation model between the live broadcast factor and the click volume is constructed by using a neural network. After the live broadcast factor is changed, a joint influence function of the click volume with each factor is generated according to the live broadcast click drop rate; The channel planning module is used to pre-record live video, upload the live video to the local server, format the live channel, calculate the second function between the quality of the live picture and the channel capacity, predict the proportion of communication resources used by each live broadcast according to the average picture quality of each type of live broadcast and the future live broadcast arrangement, and pre-allocate network resources; The picture adjustment module is used to calculate the maximum convergence value of the click volume according to the joint influence function under the condition that the live broadcast bandwidth is less than the total channel bandwidth, adjust the quality of the live broadcast picture and the insertion ratio of the live broadcast video according to the influence factor corresponding to the maximum convergence value of the click volume, and count the click volume before and after the adjustment; The live broadcast evaluation module is used to obtain the live broadcast data of the current live broadcast after the live broadcast ends, generate an influence function of the number of clicks of the current live broadcast with the changes of factors during the live broadcast, compare the current influence function with the joint influence function, obtain the deviation value of each live broadcast factor, output the live broadcast effect evaluation record according to the deviation value, and give the adjustment suggestions of the live broadcast factors; The channel planning module includes: a pre-upload unit and a bandwidth allocation unit; The pre-upload unit is used to record live video, upload the live video to the platform server in advance, and realize screen replacement locally; The bandwidth allocation unit is used to format the platform transmission channel and pre-allocate network communication resources for various types of live broadcasts; The picture adjustment module includes: an effect convergence unit and a video supplement unit; The effect convergence unit is used to calculate the size of each live broadcast influencing factor when the click volume is the highest according to the joint influence function under the constraint that the transmission bandwidth is less than or equal to the total bandwidth; The video recording supplement unit adjusts the live broadcast image quality in real time according to the calculated live broadcast influencing factors, and replaces the live broadcast image with the video recording in the server.

5. According to the artificial intelligence-based online live teaching management system of claim 4, it is characterized by: The intelligent classification module includes: a network training unit and a type output unit; The network training unit is used to record influencing factors in the platform's historical live broadcast process in the database, and the influencing factors include: live broadcast tags, number of interactions, picture quality, video stability and the proportion of real people appearing on the screen; The type output unit is used to separately store the influencing factors of each live broadcast, generate a training data set, use a neural network model to learn the training library, and use the learning results to classify the live broadcast.

6. The network live teaching management system based on artificial intelligence according to claim 5 is characterized by: The influence analysis module includes: a click association unit, a factor analysis unit and a joint influence unit; The click association unit is used to construct a database of various types of live broadcasts according to the classification results, with the click volume as the output variable and the influencing factors as the input variables; The factor analysis unit is used to construct a correlation model between each live broadcast element and the click volume, and after the live broadcast element is changed, the output of the evaluated click volume appears; The joint influence unit is used to output the joint influence function of the click volume with each element according to the influence of each live broadcast element on the click volume.

7. The network live teaching management system based on artificial intelligence according to claim 6 is characterized by: The live broadcast evaluation module includes: a digital fitting unit, a deviation recording unit and a report output unit; The digital fitting unit is used to generate an influence function of the live broadcast after the live broadcast ends, which is recorded as the third function; The deviation recording unit is used to calculate the deviation between the third function and the joint influence function on different influence factors; The report output unit is used to generate a live broadcast evaluation report according to the deviation of each influencing factor, output the report and give live broadcast suggestions for the live broadcast at that time.

Images