CN120238671A - Cinema live broadcast optimization method and system based on big data analysis - Google Patents

Abstract

The invention provides a cinema live broadcast optimizing method and system based on big data analysis, which comprises the steps of firstly acquiring historical live broadcast big data of a target cinema, wherein the historical live broadcast big data comprises a plurality of live broadcast data streams consisting of user film watching behavior records and live broadcast service state records, then carrying out multi-dimensional film watching feature extraction on the historical live broadcast big data to obtain real-time film watching features and service stability features, then carrying out dynamic strategy matching on the real-time film watching features and the service stability features based on a pre-trained dynamic optimizing model to generate an optimizing labeling result, obtaining a live broadcast service adjusting strategy according to the optimizing labeling result, synchronizing the live broadcast service adjusting strategy to a live broadcast service node in real time, adjusting current live broadcast parameter configuration according to the live broadcast service adjusting strategy, and finally updating parameters of the dynamic optimizing model according to real-time live broadcast quality indexes acquired by the adjusted live broadcast parameter configuration, thereby realizing live broadcast dynamic optimization of the cinema and improving live broadcast service quality.

Description

Cinema live broadcast optimization method and system based on big data analysis

Technical Field

The invention relates to the technical field of big data, in particular to a cinema live broadcast optimization method and system based on big data analysis.

Background

At the moment of rapid development of cinema live broadcasting technology, with the increasing demands of audience for high-quality film viewing experience, cinema live broadcasting services face a plurality of technical challenges to be solved urgently. In the operation process of the traditional cinema live broadcast system, live broadcast parameter configuration and service adjustment are mainly carried out by relying on manual experience or preset fixed rules. However, there are significant limitations to this approach.

On the one hand, the manual experience often has subjectivity and unilateral performance, and is difficult to comprehensively and accurately grasp the film watching behavior mode and the demand characteristics of different user groups in different scenes. The preference, habit and sensitivity to live broadcast quality of different users are huge in viewing, and manual experience is difficult to analyze and process the complex and changeable factors finely, so that the live broadcast service cannot accurately meet the personalized requirements of the users, and the viewing experience of the users is affected.

On the other hand, the preset fixed rules lack flexibility and adaptability. The cinema live broadcast environment is dynamically changed, and the factors such as network conditions, equipment performance, the number of users and the like are changed at all times. The fixed rules cannot be dynamically adjusted according to the real-time change conditions, and when sudden conditions such as network fluctuation, user flow sudden increase and the like are met, problems such as blocking and delay are easy to occur in live broadcast service, and the stability and fluency of live broadcast are seriously affected.

Therefore, when dealing with complex and changeable film watching demands and dynamically changing live broadcast environments, the existing cinema live broadcast technology is worry about not providing high-quality, stable and personalized live broadcast services.

Disclosure of Invention

In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a cinema live broadcast optimization method based on big data analysis, the method including:

acquiring historical live broadcast big data of a target cinema, wherein the historical live broadcast big data comprise a plurality of live broadcast data streams, and each live broadcast data stream consists of at least one user film watching behavior record and a corresponding live broadcast service state record;

performing multidimensional video watching feature extraction processing on the historical live broadcast big data to obtain real-time video watching features and service stability features of each live broadcast data stream;

Based on a pre-trained dynamic optimization model, carrying out dynamic policy matching processing on the real-time video watching characteristics and the service stability characteristics to generate an optimization labeling result of the live broadcast data stream, and generating a live broadcast service adjustment policy according to the optimization labeling result;

synchronizing the live broadcast service adjustment strategy to a live broadcast service node in real time, and triggering the live broadcast service node to adjust the current live broadcast parameter configuration according to the live broadcast service adjustment strategy;

and acquiring a real-time live broadcast quality index based on the adjusted live broadcast parameter configuration, and updating parameters of the dynamic optimization model according to the real-time live broadcast quality index.

In yet another aspect, an embodiment of the present invention further provides a cinema live broadcast optimization system based on big data analysis, which includes a processor, a machine-readable storage medium, where the machine-readable storage medium is connected to the processor, and the machine-readable storage medium is used to store a program, an instruction, or a code, and the processor is used to execute the program, the instruction, or the code in the machine-readable storage medium, so as to implement the method described above.

Based on the above aspects, the embodiment of the application realizes global performance improvement of the live cinema service in a complex dynamic environment, takes the historical live broadcast big data as an analysis basis, and enables the description of the live cinema service state to have space-time relevance and behavior perception capability by integrating the multidimensional film watching characteristics and service stability characteristics. The pre-trained dynamic optimization model is used as a core decision engine, so that intelligent mapping of real-time video watching characteristics and service stability characteristics is realized, and a prospective optimization labeling result is generated through a dynamic strategy matching mechanism, so that live broadcast service adjustment strategies can accurately respond to video watching preference and network fluctuation characteristics of different user groups. By synchronizing the adjustment strategy to the live broadcast service node in real time and dynamically reconstructing live broadcast parameter configuration, the problem of insufficient adaptability caused by parameter solidification in the traditional live broadcast service is effectively solved, and full-link real-time closed-loop control from data acquisition to strategy execution is realized. Particularly, the real-time live broadcast quality index collected based on the adjusted live broadcast parameter configuration can reversely drive the parameter update of the dynamic optimization model, and an intelligent optimization mechanism with self-evolution capability is formed, so that the intelligent optimization mechanism can be continuously optimized along with the change of the video watching behavior mode and the evolution of the network environment, and the collaborative gain is realized in multiple dimensions of improving the video watching experience of a user, enhancing the service stability, reducing the operation and maintenance cost and the like.

Drawings

Fig. 1 is a schematic diagram of an execution flow of a cinema live broadcast optimization method based on big data analysis according to an embodiment of the present invention.

Fig. 2 is a schematic diagram of exemplary hardware and software components of a theatre live optimization system based on big data analysis provided by an embodiment of the present invention.

Detailed Description

The invention is specifically described below with reference to the accompanying drawings, and fig. 1 is a schematic flow diagram of a cinema live broadcast optimization method based on big data analysis according to an embodiment of the invention, and the cinema live broadcast optimization method based on big data analysis is described in detail below.

Step S110, acquiring historical live broadcast big data of a target cinema, wherein the historical live broadcast big data comprise a plurality of live broadcast data streams, and each live broadcast data stream consists of at least one user film watching behavior record and a corresponding live broadcast service state record.

In this embodiment, it is assumed that the target theater often performs live showing activities of movies. In detail, taking a live broadcast process of a science fiction movie as an example, historical live broadcast big data can be obtained. Illustratively, for each user viewing behavior record, e.g., user a's record, may contain the time he entered the live, pause operations during the live, fast forward operations, volume adjustment operations, etc. The corresponding live service status record may include a network connection condition in the live process, a load condition of the server, and the like. If 100 users watch the live broadcast, there will be live broadcast data streams composed of 100 user movie watching behavior records and corresponding live broadcast service state records, and then consider other live broadcast of movies performed before the cinema, such as action movie live broadcast, comedy movie live broadcast, etc., where relevant data of the different movie live broadcast together form historical live broadcast big data comprising a plurality of live broadcast data streams.

And step S120, performing multi-dimensional film watching feature extraction processing on the historical live broadcast big data to obtain real-time film watching features and service stability features of each live broadcast data stream.

In this embodiment, taking the case of a live broadcast of a science fiction movie, first, a time window division process is performed on a user movie viewing record in a live broadcast data stream, and it is assumed that the whole live broadcast duration is divided into movie viewing segments of a continuous time period of one every 10 minutes. For the movie fragment of user a, a pre-trained real-time feature encoder may be invoked to process it.

Further, in generating the user interaction frequency feature, the number of interaction operations and the operation interval duration of the user a within a single 10 minute time window are analyzed. For example, within a certain 10 minutes, the user a performs 3 times of pause operations, each of which has a duration of 2 minutes, 3 minutes and 4 minutes, respectively, and generates the user interaction frequency characteristic from these data.

Further, in generating the picture switching preference feature, a timing relationship between a picture switching request time point in the user a operation log and the live content key frame is identified. Assume that the set of screen switching request time points extracted from the operation log of the user a is [15 minutes, 25 minutes, 32 minutes ]. And analyzing the live content key frame time point set comprising the scene switching key frame, the special effect starting key frame and the scenario turning key frame from the live content metadata to be [10 minutes, 20 minutes, 30 minutes and 40 minutes ]. Processing each point in the picture switching request time point set, for example, for 15 minutes, locating the closest two candidate key frame time points in the ordered key frame time point list to be 10 minutes and 20 minutes through a binary search algorithm, and finding that the difference between the two-way time difference and the 10 minutes is minimum after calculating the two-way time difference, so that the 10 minutes is the target key frame time point. The absolute time difference between 15 minutes and 10 minutes is calculated to be 5 minutes, which is the original time deviation amount, since 15 minutes are later than 10 minutes, and are marked as positive values according to the marking rule. Such processing is performed for all the screen switching request time points, and a time deviation amount sequence with a time sequence direction identification is obtained. And then counting a first frequency in a preset positive deviation interval, a second frequency in a preset negative deviation interval and a third frequency in a zero deviation tolerance interval. Assume that the preset positive deviation interval is [3 minutes, 8 minutes ], the negative deviation interval is [ -8 minutes, -3 minutes ], and the zero deviation tolerance interval is [ -1 minute, 1 minute ]. After statistics, 2 times are found in the positive deviation interval, 1 time is found in the negative deviation interval, and 0 time is found in the zero deviation tolerance interval. According to the frequency proportion relation, an active switching tendency index of a user for a scene switching key frame, a delay following index for a special effect starting key frame and a pre-judging switching density index for a scenario turning key frame are calculated. Assuming that the total number of scene switching key frames is 5, the total number of special effect initial key frames is 3, the total number of scenario turning key frames is 4, the active switching tendency index is calculated to be 2/5=0.4, the pre-judgment switching density index is 1/3=0.33 (a simple weighted summation example is performed here), and the delay following index is 0/4=0. And configuring a weight distribution strategy for the science fiction film (multi-scene switching type) according to the live content type, improving the weight coefficient of the active switching trend index, if the weight coefficient is improved to 0.6, performing a series of operations such as product operation on the adjusted weight coefficient and the corresponding index, and then generating a picture switching preference feature set.

For the content response delay feature, the difference between the content loading request and the server response timestamp in the user a operation log is monitored, and the number of abnormal events that the content loading delay exceeds a preset threshold (e.g., 3 seconds) is counted. The content response delay feature is generated assuming that the number of abnormal events that delay the content loading by more than 3 seconds is 2 in the whole live broadcast process.

And meanwhile, carrying out abnormal event detection processing on the live broadcast service state record. Assuming that during live broadcast, the timestamp of a live broadcast service interruption event is identified as 30 minutes after the start of live broadcast, the trigger condition of the service degradation event is that the network bandwidth is lower than 10Mbps for 5 minutes. And (3) carrying out stability quantitative analysis on the live service state record, extracting a network bandwidth sampling value sequence associated with a time stamp of a live service interruption event from the live service state record, wherein the bandwidth sampling data from 5 minutes before the interruption event to 5 minutes after the interruption event is [8Mbps,9Mbps,7Mbps,6Mbps,5Mbps,4Mbps,5Mbps,6Mbps,7Mbps,8Mbps ], calculating the standard deviation and average value ratio of the data, and generating the bandwidth fluctuation characteristic. And analyzing video decoding log fragments matched with the triggering condition of the service degradation event in the live broadcast service state record, identifying video frame data transmitted during the effective period of the triggering condition of the service degradation event, counting the percentage of the number of frames carrying decoding error identification in the video frame data to the total transmission frame, and generating decoding error rate characteristics. Traversing the hardware resource usage record of the live broadcast service state record, positioning the central processing unit occupancy rate data points overlapped with the time stamp of the service interruption event, extracting a continuous time period exceeding a preset safety threshold (such as 80%), calculating the weighted difference between the maximum value and the average value of the central processing unit occupancy rate in the continuous time period, and generating the resource occupancy peak value characteristic. And finally, performing time sequence alignment processing on the features to generate a second feature set containing bandwidth fluctuation features, decoding error rate features and resource occupation peak value features. And carrying out feature fusion on the first feature set and the second feature set to obtain real-time video watching features and service stability features of each live broadcast data stream.

And step S130, carrying out dynamic policy matching processing on the real-time video watching characteristics and the service stability characteristics based on a pre-trained dynamic optimization model, generating an optimization labeling result of the live broadcast data stream, and generating a live broadcast service adjustment policy according to the optimization labeling result.

In this embodiment, taking the related real-time video feature and service stability feature of the live broadcast of the previous science fiction movie as an example, feature vector stitching is performed to generate a multidimensional feature input vector. The historical feature weight template stored in the strategy matching layer in the dynamic optimization model is assumed to contain reference weight values of feature dimensions under different live broadcast scenes, for example, the reference weight value of the user interaction frequency feature is 0.2, the reference weight value of the picture switching preference feature is 0.3, the reference weight value of the content response delay feature is 0.1, the reference weight value of the bandwidth fluctuation feature is 0.2, the reference weight value of the decoding error rate feature is 0.1, and the reference weight value of the resource occupation peak feature is 0.1. And calculating cosine similarity between each feature dimension in the multidimensional feature input vector and the historical feature weight template, and generating a scene matching degree index. The method comprises the steps of carrying out dynamic adjustment on a reference weight value according to a scene matching degree index after calculation, adjusting the weight of a user interaction frequency characteristic to be 0.15, adjusting the weight of a picture switching preference characteristic to be 0.35, adjusting the weight of a content response delay characteristic to be 0.12, adjusting the weight of a bandwidth fluctuation characteristic to be 0.18, adjusting the weight of a decoding error rate characteristic to be 0.08, adjusting the weight of a resource occupation peak value characteristic to be 0.12, and carrying out normalization processing to obtain characteristic importance distribution. And carrying out weighted fusion on the multidimensional feature input vector according to the feature importance distribution to generate an optimized decision vector. And inputting the optimization decision vector to a labeling output layer of the dynamic optimization model to generate an optimization labeling result comprising a picture quality optimization direction (such as improving picture definition), a bandwidth allocation optimization direction (such as increasing bandwidth allocation) and a decoding priority optimization direction (such as improving decoding priority of a special effect scene). And generating a live broadcast service adjustment strategy according to the optimization labeling results, such as adjusting parameters of a video encoder to improve picture quality, adjusting parameters of a bandwidth allocation module to increase bandwidth, adjusting parameters of a decoding module to improve special effect scene decoding priority, and the like.

And step S140, synchronizing the live broadcast service adjustment strategy to a live broadcast service node in real time, and triggering the live broadcast service node to adjust the current live broadcast parameter configuration according to the live broadcast service adjustment strategy.

In this embodiment, for the generated live service adjustment policy, the live scene of the science fiction movie is continuously analyzed by taking an example, and it is assumed that the live service adjustment policy includes a picture resolution adjustment parameter (for example, the adjustment from 720p to 1080 p), a code rate control parameter (for example, the maximum allowable code rate is 5Mbps, the minimum guaranteed code rate is 2 Mbps), and a buffer interval configuration parameter (for example, the initial buffer threshold is adjusted from 10 seconds to 15 seconds). And sending the picture resolution adjustment parameter to a video coding module, and triggering a video coder to dynamically adjust the resolution preset value from 720p to 1080p. And synchronizing the code rate control parameters to a bandwidth allocation module, and monitoring the real-time network throughput and the data packet loss rate of the live service node in the process. Assuming that the real-time network throughput is 3Mbps, the data packet loss rate is 1%, and calculating the recommended code rate interval under the current network condition as [2Mbps,3Mbps ] according to the maximum allowable code rate and the minimum guaranteed code rate in the code rate control parameters. And calling a code rate self-adaptive controller to dynamically adjust the video transmission code rate in the recommended code rate interval, so that the actual code rate and the network throughput keep a preset proportional relationship, and for example, the actual code rate is adjusted to 2.5Mbps. When the packet loss rate is detected to exceed a threshold (e.g., 3%), an emergency rate reduction operation is triggered and the forward error correction coding mechanism is enabled. The buffer interval configuration parameters are deployed to a data buffer module, and the initial buffer threshold value is triggered to be adjusted according to the performance of the user equipment, for example, the initial buffer threshold value is adjusted from 10 seconds to 15 seconds, so that the smoothness of playing is ensured.

And step S150, acquiring a real-time live broadcast quality index based on the adjusted live broadcast parameter configuration, and updating parameters of the dynamic optimization model according to the real-time live broadcast quality index.

In this embodiment, in the case of the science fiction movie live broadcast, the real-time live broadcast quality index may be collected based on the adjusted live broadcast parameter configuration. In detail, the user side playing fluency index, the picture quality score and the interactive response delay index are collected in a preset observation period (such as the whole live broadcast residual duration). And (3) carrying out the frequency statistics of the jamming events on the play fluency index, and assuming that the jamming events occur 2 times within the remaining 30 minutes of live broadcasting, generating a jamming frequency characteristic of 2/30 (concept of jamming times per minute). The picture quality scores are analyzed in time series, and the quality fluctuation characteristics are generated by analyzing the data, assuming that the scores of the audience on the picture quality are [8 points, 7 points, 8 points, 9 points ] in different time periods in the live broadcast process. And calculating the percentile of the interactive response delay index, and generating delay distribution characteristics by calculating the interactive response delay of 90% of the percentile as 2 seconds through false design. The degree of difference between the stuck frequency characteristic and the historical stuck baseline data (assuming that the historical stuck baseline is 0.1 stuck per minute) is calculated to generate a first loss component. Analyzing the degree of deviation of the quality fluctuation feature from the expected quality stability curve (assuming that the expected quality stability curve is within 1 of the score fluctuation), and generating a second loss component. The proportion of high delay samples (assuming high delay is defined as a delay greater than 3 seconds) in the delay profile is counted, generating a third loss component. The first loss component, the second loss component, and the third loss component are weighted and summed to generate a composite loss function. Assuming that the first loss component weight is 0.3, the second loss component weight is 0.4, and the third loss component weight is 0.3, a comprehensive loss function is calculated. And then calculating the partial derivative of the model weight of the comprehensive loss function through a back propagation algorithm, generating a model parameter adjustment gradient, and updating the weight parameter of the dynamic optimization model based on the model parameter adjustment gradient so as to perform optimization operation more accurately in subsequent live broadcasting.

Based on the steps, the embodiment of the application realizes the global performance improvement of the live cinema service in a complex dynamic environment, takes the historical live broadcast big data as an analysis basis, and enables the description of the live broadcasting service state to have space-time relevance and behavior perception capability by integrating the multidimensional video watching characteristics and the service stability characteristics. The pre-trained dynamic optimization model is used as a core decision engine, so that intelligent mapping of real-time video watching characteristics and service stability characteristics is realized, and a prospective optimization labeling result is generated through a dynamic strategy matching mechanism, so that live broadcast service adjustment strategies can accurately respond to video watching preference and network fluctuation characteristics of different user groups. By synchronizing the adjustment strategy to the live broadcast service node in real time and dynamically reconstructing live broadcast parameter configuration, the problem of insufficient adaptability caused by parameter solidification in the traditional live broadcast service is effectively solved, and full-link real-time closed-loop control from data acquisition to strategy execution is realized. Particularly, the real-time live broadcast quality index collected based on the adjusted live broadcast parameter configuration can reversely drive the parameter update of the dynamic optimization model, and an intelligent optimization mechanism with self-evolution capability is formed, so that the intelligent optimization mechanism can be continuously optimized along with the change of the video watching behavior mode and the evolution of the network environment, and the collaborative gain is realized in multiple dimensions of improving the video watching experience of a user, enhancing the service stability, reducing the operation and maintenance cost and the like.

In one possible implementation, step S120 includes:

Step S121, performing a time window division process on the user movie viewing behavior record in the live data stream, so as to obtain movie viewing behavior segments of a plurality of continuous time periods.

For example, the live duration of the whole science fiction movie is 120 minutes, and the live duration can be divided into continuous time periods of one every 15 minutes, so that 8 movie viewing behavior fragments can be generated, and each movie viewing behavior fragment covers various movie viewing operation information of users within the 15 minutes.

Step S122, invoking a pre-trained real-time feature encoder to extract real-time video watching features of the video watching behavior fragments, and generating a first feature set comprising user interaction frequency features, picture switching preference features and content response delay features.

And step S123, carrying out abnormal event detection processing on the live broadcast service state record, and identifying the time stamp of the live broadcast service interruption event and the triggering condition of the service degradation event.

For example, during a science fiction movie live, the timestamp identifying the live service interruption event is 45 minutes after the live is started, and the trigger condition for the service degradation event is that the network bandwidth is below 10Mbps for 5 minutes. Thus, a stability quantitative analysis can be performed on the live service status record based on the timestamp and the trigger condition. In particular, a sequence of network bandwidth sample values associated with a timestamp of a live service interruption event may be extracted from a live service status record, e.g., the bandwidth sample data of [9Mbps,8Mbps,7Mbps,6Mbps,5Mbps,4Mbps,5Mbps,6Mbps,7Mbps,8Mbps ] 5 minutes before the interruption event occurs to 5 minutes after the interruption event occurs. The standard deviation to mean ratio of these data is calculated to generate the bandwidth fluctuation signature. The average value is calculated first, the data are added to obtain 65Mbps total, and then divided by 10 data to obtain 6.5Mbps average value. Then, the square of the difference between each data and the mean is calculated, for example, the difference between the first data 9Mbps and the mean 6.5Mbps is 2.5Mbps, the square is 6.25Mbps ₂, all the data are calculated and summed to obtain 35Mbps ₂, the sum is divided by 10 data to obtain 3.5Mbps ₂, and the standard deviation is the square root of the variance of about 1.87Mbps. Finally, the standard deviation to mean ratio is calculated, and 1.87Mbps divided by 6.5Mbps is approximately equal to 0.29, which is the bandwidth fluctuation feature.

And step S124, performing stability quantitative analysis on the live broadcast service state record based on the time stamp and the triggering condition, and generating a second feature set containing bandwidth fluctuation features, decoding error rate features and resource occupation peak value features.

And step S125, carrying out feature fusion on the first feature set and the second feature set to obtain the real-time video watching feature and the service stability feature.

In this embodiment, a first feature set including a user interaction frequency feature, a picture switching preference feature and a content response delay feature and a second feature set including a bandwidth fluctuation feature, a decoding error rate feature and a resource occupation peak feature are feature-fused to obtain a real-time video watching feature and a service stability feature, for example, each feature in the first feature set and the second feature set may be combined together according to a set running rule, so that the finally obtained real-time video watching feature and service stability feature may comprehensively reflect the video watching experience of the user and the stability condition of the service.

In one possible implementation, step S122 includes:

Step S1221, invoking a pre-trained real-time feature encoder to analyze the behavior mode of the user operation log in the movie behavior segment, extracting the interaction operation times and operation interval duration of the user in a single time window, and generating the user interaction frequency feature.

For example, taking the aforementioned 15-minute movie fragment as an example, looking at the user operation log, the user can know that 3 times of pause operation, 1 time of fast forward operation and 2 times of volume adjustment operation are performed in the time period. The total number of these operations is calculated, 3 times of pause+1 times of fast forward+2 times of volume adjustment=6 times of operations, thereby constituting the number of interactive operations of the user within the 15-minute time window. Looking again at the operation interval duration, the first pause is at 3 minutes, the second pause is at 7 minutes, then the operation interval duration from the first pause to the second pause is 7-3=4 minutes, the operation interval duration from the second pause to the third pause (assuming the third pause is at 12 minutes) is 12-7=5 minutes, etc., these operation interval duration data together form the operation interval duration portion in the user interaction frequency feature. In this way, the user interaction frequency characteristic is completely generated, and the user interaction frequency characteristic can reflect the operation activity degree of the user in the film watching process.

Step S1222, identifying a time sequence relationship between the time point of the picture switching request in the user operation log and the key frame of the live broadcast content, calculating a time deviation amount between the picture switching request and the key frame of the content, and generating a picture switching preference feature.

In one possible implementation, step S1222 includes:

step S1222-1, extracting the time stamp data of the trigger time of the screen switching operation from the user operation log, and generating a screen switching request time point set.

For example, in the above 15-minute movie fragment, when the screen switching operation trigger time is found to be 5 th, 10 th and 13 th minutes from the user operation log, respectively, the screen switching request time point set is [5 minutes, 10 minutes, 13 minutes ].

Step S1222-2, analyzing the pre-marked key frame type and the corresponding time stamp sequence in the metadata of the live broadcast content, and generating a live broadcast content key frame time point set comprising a scene switching key frame, a special effect starting key frame and a scenario turning key frame.

In this embodiment, it is assumed that the time stamp of the scene switching key frame is analyzed from the live content metadata to be 4 th minute and 8 th minute, the time stamp of the special effect starting key frame is 6 th minute and 11 th minute, and the time stamp of the scenario turning key frame is 9 th minute and 14 th minute, and then the live content key frame time point set is [ 4min, 6min, 8min, 9 min, 11 min and 14 min ].

Step S1222-3, traversing each picture switching request time point in the picture switching request time point set, and searching a target key frame time point with the smallest time difference with the current picture switching request time point in the live broadcast content key frame time point set.

In this embodiment, taking the picture switching request time point as an example, the difference between the picture switching request time point and each time point in the key frame time point set of the live content is calculated, wherein the difference between 5 minutes and 4 minutes is 5-4=1 minute, the difference between 5 minutes and 6 minutes is 6-5=1 minute, the difference between 5 minutes and 8 minutes is 8-5=3 minutes, the difference between 5 minutes and 9 minutes is 9-5=4 minutes, the difference between 5 minutes and 11 minutes is 11-5=6 minutes, and the difference between 5 minutes and 14 minutes is 14-5=9 minutes. It can be seen that the difference from 4 minutes and 6 minutes is the smallest and equal, and the 6 minutes later in time order are preferentially selected as target key frame time points according to the rule. The same operation is also performed for the screen switching request time points of 10 minutes and 13 minutes.

Step S1222-4, calculating the absolute time difference between the picture switching request time point and the target key frame time point, and generating the original time deviation of the single picture switching request time point.

For example, for a 5-minute screen switching request time point and a 6-minute target key frame time point, the absolute time difference is 6-5=1 minute, whereby the original time deviation amount is output as the screen switching request time point.

And step S1222-5, performing sign marking processing on the original time deviation, marking as a positive value when the picture switching request time point is later than the target key frame time point, marking as a negative value when the picture switching request time point is earlier than the target key frame time point, and generating a time deviation amount sequence with a time sequence direction mark.

For example, since 5 minutes is earlier than 6 minutes, it is noted as-1 minute. The original time deviation amounts of all the picture switching request time points are marked in this way, and a time deviation amount sequence with a time sequence direction mark is obtained.

Step S1222-6, counting the first frequency in the preset positive deviation interval, the second frequency in the preset negative deviation interval and the third frequency in the zero deviation tolerance interval in the time deviation amount sequence.

In this embodiment, it is assumed that the preset positive deviation interval is [3 minutes, 8 minutes ], the negative deviation interval is [ -8 minutes, -3 minutes ], and the zero deviation tolerance interval is [ -1 minute, 1 minute ]. After statistics, 0 positive deviations were found, 1 negative deviations were found (i.e., the deviation was calculated as-1 minute earlier), and 2 zero deviations were found (i.e., a deviation was found to be within the interval).

Step S1222-7, according to the proportional relation of the first frequency, the second frequency and the third frequency, generating an active switching tendency index of the user for the scene switching key frame, a delay following index for the special effect initial key frame and a pre-judging switching density index for the scenario turning key frame.

In this embodiment, the total number of scene switching key frames is assumed to be 5, the total number of special effect starting key frames is assumed to be 3, and the total number of scenario turning key frames is assumed to be 4. And calculating an active switching tendency index of the user aiming at the scene switching key frames, and calculating the ratio of the frequency 0 of the forward deviation interval to the total number 5 of the scene switching key frames, wherein the ratio of 0 to 5 is equal to 0. The delay following index for the special effect start key frame is calculated, and the frequency 1 of the negative deviation interval is weighted and summed with the total number 3 of the special effect start key frames (here, the weighting coefficient is assumed to be 1), and 1 divided by 3 is equal to about 0.33. Calculating a pre-judging switching density index for the scenario turning key frames, and carrying out dynamic proportion mapping on the frequency 2 of the zero deviation tolerance interval and the total number 4 of the scenario turning key frames (assuming that the mapping relation is direct division), wherein the frequency 2 is divided by the total number 4, and the frequency is equal to 0.5.

Step S1222-8, the active switching tendency index, the delay following index and the pre-judging switching density index are subjected to weight distribution according to the key frame type, and a picture switching preference feature set containing time sensitivity weight and content association degree weight is generated.

For example, as the science fiction movie is live broadcast, the weight coefficient of the active switching tendency index is promoted according to the weight distribution strategy, for example, the weight coefficient is promoted to 0.6, which belongs to the multi-scene switching type. For the special effect initial key frame, the weight coefficient of the delay following index is adjusted to 0.2 according to the special effect condition assumption in the live broadcast content, and for the scenario turning key frame, the weight coefficient of the pre-judging switching density index is adjusted to 0.2. And then carrying out product operation on the adjusted weight coefficient and the corresponding index, wherein the active switching tendency index is multiplied by the weight coefficient of the active switching tendency index to be 0 and multiplied by 0.6 to be 0, the delay following index is multiplied by the weight coefficient of the delay following index to be 0.33 and multiplied by 0.2 to be 0.066, and the pre-judging switching density index is multiplied by the weight coefficient of the delay following index to be 0.5 and multiplied by 0.2 to be 0.1. Combining these results, a picture switching preference feature set is generated that includes a time sensitivity weight and a content relevance weight, which can reflect the user's preference characteristics for different types of key frame picture switching.

Step S1223, monitoring the difference between the content loading request and the server response time stamp in the user operation log, counting the number of abnormal events that the content loading delay exceeds the preset threshold, and generating the content response delay feature.

For example, in the 15-minute movie fragment, the time stamp of the content loading request and the server response in the user operation log is checked. Assuming that the content load request occurs at 2 minutes, the server response time stamp is 6 minutes, so the difference between them is 6-2=4 minutes. Assuming the preset threshold is 3 minutes again, this is an event in which the content loading delay exceeds the preset threshold. And continuously checking all content loading requests and server response conditions within the whole 15 minutes, counting the times of such abnormal events, and supposing that 2 times are counted in total, forming a content response delay characteristic by the 2 times, wherein the content response delay characteristic can reflect the delay condition of a user when acquiring the content.

In one possible implementation, step S1222-2 includes:

Step S1222-21, reading the embedded metadata mark block in the live content video stream, extracting the frame number, frame type label and offset relative to the live start time of each key frame.

For example, in live broadcast of a science fiction movie, it is assumed that a metadata tag block in a video stream contains a series of key frame information. For one of the key frames, the frame number is 100, the frame type label is "scene switch", and the offset relative to the live start time is 5 minutes. In this way, information extraction is performed on all key frames.

Step S1222-22, dividing the key frame into scene switching key frame, special effect starting key frame and scenario turning key frame according to the frame type label.

For example, among the extracted key frames, the key frame with the frame type label of "scene switching" is classified as a scene switching key frame category, the key frame with the frame type label of "special effect start" is classified as a special effect start key frame category, and the key frame with the frame type label of "scenario turning" is classified as a scenario turning key frame category.

Step S1222-23, converting the frame number of each key frame into absolute timestamp data based on the offset, and generating a time point set of key frames of the live broadcast content arranged in time sequence.

For example, assuming that the first scene change key frame is shifted by 5 minutes, its absolute timestamp is 5 minutes after the start of live broadcast, and the second scene change key frame is shifted by 12 minutes, its absolute timestamp is 12 minutes. Such conversion is performed on all types of key frames, and then the time stamps of the key frames are arranged in time sequence to form a live content key frame time point set. For example, the set of live content keyframes time points may be [5 minutes (scene-switch keyframe), 8 minutes (special effects start keyframe), 12 minutes (scene-switch keyframe), 15 minutes (scenario-break keyframe), 20 minutes (special effects start keyframe) ].

Step S1222-24, verifying the time stamp continuity of the adjacent key frames in the live content key frame time point set, if the time stamp jump exceeds the preset frame interval threshold, inserting the virtual key frame time point in the jump section and marking as unclassified key frame type.

For example, assuming that the preset frame interval threshold is 3 minutes, when the live content key frame time point set is checked, a timestamp jump is found to exist between the 12-minute scene-switching key frame and the 15-minute scenario turning key frame. 15-12 = 3 minutes, just reaching the threshold, if the jump exceeds 3 minutes, a virtual key frame time point is inserted in the interval and labeled as unclassified key frame type.

For example, in one possible implementation, step S1222-3 includes:

step S1222-31, performing time ascending sort processing on the live content key frame time point set to generate an ordered key frame time point list.

Continuing with the above-mentioned live content key frame time point set [5 minutes (scene switching key frame), 8 minutes (special effect starting key frame), 12 minutes (scene semana switching key frame), 15 minutes (scenario turning key frame), 20 minutes (special effect starting key frame) ] as an example, after the time-ascending sorting process, the ordered key frame time point list is still [5 minutes (scene switching key frame), 8 minutes (special effect starting key frame), 12 minutes (scene switching key frame), 15 minutes (scenario turning key frame), 20 minutes (special effect starting key frame) ].

Step S1222-32, locating two candidate key frame time points closest to the current picture switching request time point in the ordered key frame time point list by adopting a binary search algorithm.

In this embodiment, assuming that the screen switching request time point is 11 minutes, first, the element in the middle of the ordered key frame time point list, namely, 12 minutes (scene switching key frame), is taken, and since 11 minutes is smaller than 12 minutes, the intermediate element is taken in the first half of the list [5 minutes (scene switching key frame), 8 minutes (special effect starting key frame) ], here, 8 minutes (special effect starting key frame). This locates two candidate key frame time points closest to 11 minutes, 8 minutes (special effect start key frame) and 12 minutes (scene change key frame).

Step S1222-33, calculating the two-way time difference between the current picture switching request time point and the two candidate key frame time points, and selecting the candidate key frame time point with the smallest difference as the target key frame time point.

For example, for a screen switching request time point of 11 minutes, the difference from 8 minutes (special effect start key frame) is 11-8=3 minutes, and the difference from 12 minutes (scene switching key frame) is 12-11=1 minute. Less than 3 minutes 1 minute, 12 minutes (scene change key frame) is selected as the target key frame time point.

In step S1222-34, when the difference between the two candidate key frame time points and the current picture switching request time point is equal, the candidate key frame time point with the subsequent time sequence is preferentially selected as the target key frame time point.

For example, if the screen switching request time point is 9 minutes, the difference from 8 minutes (special effect start key frame) is 9-8=1 minute, and the difference from 12 minutes (scene switching key frame) is 12-9=3 minutes. However, if the difference is equal, for example, the screen switching request time point is 10 minutes, the difference from 8 minutes (special effect start key frame) is 10-8=2 minutes, and the difference from 12 minutes (scene switching key frame) is 12-10=2 minutes, then 12 minutes (scene switching key frame) is preferentially selected as the target key frame time point.

For example, in one possible implementation, step S1222-5 includes:

In step S1222-51, a time deviation amount annotation rule base is built defining positive values to indicate that the user operation is lagging behind the key frame event and negative values to indicate that the user operation is leading ahead of the key frame event.

For example, when the time deviation amount between the screen switching request time point and the target key frame time point is processed, the time deviation amount is marked according to the marking rule base.

Step S1222-52, adding symbol identification to each original time deviation according to the sequence of the target key frame time point and the picture switching request time point.

Assuming that the screen switching request time point is 11 minutes, the target key frame time point is 12 minutes, and since 11 minutes are earlier than 12 minutes, the original time deviation amount is 12-11=1 minute, labeled as-1 minute according to the rule, indicating that the user operation leads the key frame event.

And step S1222-53, calculating the absolute value of the marked time deviation amount, and generating a time deviation amount absolute value sequence and a symbol identification sequence.

The absolute value of the time deviation amount previously designated as-1 minute was 1 minute, and thus the time deviation amount absolute value sequence was [1 minute ], and the symbol identification sequence was [ - ].

And step S1222-54, storing the time deviation amount absolute value sequence and the symbol identification sequence in a time sequence association way, and generating a time deviation amount sequence with a time sequence direction identification.

For example, the absolute values of the time offsets corresponding to the plurality of screen switch request time points are associated with symbol identifications in time series, such as [ -1 minute, 2 minutes, -3 minutes ], where-1 minute indicates that the first screen switch request time point is advanced by 1 minute relative to the target key frame time point, 2 minutes indicates that the second screen switch request time point is retarded by 2 minutes relative to the target key frame time point, and-3 minutes indicates that the third screen switch request time point is advanced by 3 minutes relative to the target key frame time point.

For example, in one possible implementation, step S1222-6 includes:

Step S1222-61, obtaining the upper limit value and the lower limit value of the preset forward deviation interval, and counting the number of events with the absolute value of the time deviation value in the interval and the sign of positive value as the first frequency.

For example, assuming that the preset forward deviation interval is [3 minutes, 8 minutes ], only 7 minutes, the absolute value of which is [3 minutes, 8 minutes ] and the sign of which is positive, is an event in the time deviation amount sequence of [ -1 minute, 2 minutes, -3 minutes, 5 minutes, -2 minutes, 7 minutes ], the first frequency is 1.

Step S1222-62, obtaining the upper limit value and the lower limit value of the preset negative deviation interval, and counting the number of events with the absolute value of the time deviation in the interval and the sign of the negative value as the second frequency.

For example, it is assumed that the predetermined negative bias interval is [ -8 min, -3 min ], and in the time bias amount sequence, the absolute value is [ -8 min, -3 min ] and the sign is negative, and the second frequency is 1.

Step S1222-63, defining zero deviation tolerance interval as event with time deviation absolute value smaller than or equal to preset error threshold, counting the number of event meeting the condition as the third frequency.

For example, assuming that the preset error threshold is 1 minute, in the time deviation amount sequence, there is an event of-1 minute in which the absolute value is 1 minute or less, so the third frequency is 1.

Step S1222-64, excluding abnormal deviation events which do not fall into the three intervals in the time deviation amount sequence, and generating a frequency statistical result after cleaning.

In this example, 2 minutes and-2 minutes do not fall within the three intervals, so that the two events are not considered in generating the post-cleaning frequency statistics, the post-cleaning frequency statistics being 1 for the first frequency, 1 for the second frequency, and 1 for the third frequency.

For example, in one possible implementation, step S1222-7 includes:

Step S1222-71, calculating the ratio of the first frequency to the total number of the scene switching key frames to generate an active switching tendency index reflecting that the user actively initiates the switching operation after the scene switching.

For example, assuming that the total number of scene-cut key frames is 5 and the first frequency is 1, the active-cut tendency index is 1 divided by 5 and is equal to 0.2.

And step S1222-72, carrying out weighted summation on the second frequency and the total number of the special effect initial key frames to generate a pre-judging switching density index reflecting that a user switches in advance before the special effect starts.

For example, assuming that the total number of special effect start key frames is 3, the second frequency is 1, and the weighted sum (here, assuming that the weighting coefficient is 1), the pre-determined switching density index is 1 divided by 3 and is approximately equal to 0.33.

Step S1222-73, mapping the third frequency and the total number of the scenario turning key frames in a dynamic proportion mode, and generating a delay following index reflecting that the user immediately follows the scenario turning point to switch.

For example, assuming that the total number of scenario turning key frames is 4, the third frequency is 1, and here assuming that the dynamic scale map is a direct division, then the delay following index is 1 divided by 4 and is equal to 0.25.

Step S1222-74, normalize the active switching tendency index, the pre-judging switching density index and the delay following index, and eliminate the statistical deviation caused by the difference of the key frame types.

For example, assuming that the active switching tendency index is 0.2, the pre-judgment switching density index is 0.33, and the delay following index is 0.25. Then, first, the sum thereof is calculated to be 0.2+0.33+0.25=0.78. And dividing each index by the sum to obtain a normalized active switching tendency index of 0.2 divided by 0.78 approximately equal to 0.26, a normalized pre-judgment switching density index of 0.33 divided by 0.78 approximately equal to 0.42, and a normalized delay following index of 0.25 divided by 0.78 approximately equal to 0.32.

For example, in one possible implementation, step S1222-8 includes:

Step S1222-81, configuring a weight distribution strategy according to the live content type, and when the live content is detected to be of a multi-scene switching type, improving the weight coefficient of the active switching tendency index.

For example, when it is detected that the live content is of a multi-scene switching type (the science fiction movie belongs to the multi-scene switching type), the weight coefficient of the active switching tendency index is raised. Assume that the weight coefficient of the active handover tendency index is raised to 0.6.

And step S1222-82, when the live broadcast content is detected to be of a significant special effect density type, the weight coefficient of the pre-judgment switching density index is improved.

Assuming that the special effect is more remarkable in the science fiction film, the weight coefficient of the pre-judgment switching density index is improved to 0.25.

And step S1222-83, when the live broadcast content is detected to be of a strong scenario coherent type, the weight coefficient of the delay following index is improved.

And assuming strong scenario continuity of the science fiction film, the weight coefficient of the delay following index is improved to 0.15.

Step S1222-84, multiply the adjusted weight coefficient with the corresponding index to generate a time sensitivity weight vector.

The active switching tendency index is multiplied by 0.26 multiplied by 0.6=0.156, the pre-judgment switching density index is multiplied by 0.42 multiplied by 0.25=0.105, and the delay following index is multiplied by 0.32 multiplied by 0.15=0.048. The time sensitivity weight vector is [0.156,0.105,0.048].

Step S1222-85, calculating the content association weight vector according to the switching success rate of different key frame types in the user history film-viewing data.

The scene switching key frame switching success rate is 80%, the special effect starting key frame switching success rate is 70%, and the scenario turning key frame switching success rate is 60% which are obtained by analysis from the user history film watching data. These success rates were normalized to sum to 80% +70% +60% = 210%. The content association weight of the normalized scene change key frame is 80 percent divided by 210 percent and is approximately equal to 0.38, the content association weight of the special effect initial key frame is 70 percent divided by 210 percent and is approximately equal to 0.33, and the content association weight of the scenario turning key frame is 60 percent divided by 210 percent and is approximately equal to 0.29. The content association weight vector is [0.38,0.33,0.29].

Step S1222-86, performing dot product operation on the time sensitivity weight vector and the content association weight vector to generate a picture switching preference feature set.

For example, the calculation process is 0.156 times 0.38+0.105 times 0.33+0.048 times 0.29=0.05928+0.03465+0.01392= 0.10785. The picture switching preference feature set is composed of the calculation result, related weight, index and other information, and the picture switching preference feature set can comprehensively reflect the preference feature of the user in the aspect of picture switching.

In one possible implementation, step S124 includes:

Step S1241, extracting a network bandwidth sampling value sequence associated with the timestamp of the live service interruption event from the live service status record, screening out bandwidth sampling data in a set time range before and after each live service interruption event, calculating a standard deviation and mean value ratio of the bandwidth sampling data, and generating a bandwidth fluctuation feature.

For example, in a science fiction movie live, it is assumed that a live service interruption event occurs, with a timestamp of 45 minutes after the live is started. A time frame of 5 minutes before and after the occurrence of the live service interruption event is set to extract bandwidth sampling data, i.e., a time period of from 40 minutes to 50 minutes. Acquiring bandwidth sampling value in the time period from live service state record the sequences are [9Mbps,8Mbps,7Mbps,6Mbps,5Mbps,4Mbps,5Mbps,6Mbps,7Mbps,8Mbps ]. When the average value is calculated, the data are added to obtain 65Mbps total, and then divided by 10 data to obtain 6.5Mbps average value. Then the square of the difference between each data and the mean is calculated, for example, the difference between the first data 9Mbps and the mean 6.5Mbps is 2.5Mbps, the square is 6.25Mbps ₂, the difference between the second data 8Mbps and the mean 6.5Mbps is 1.5Mbps, the square is 2.25Mbps ₂, and so on, all the data are calculated and summed to obtain 35Mbps ₂, the variance is 3.5Mbps ₂ after dividing the data by 10, and the square root of the standard deviation is about 1.87Mbps. Finally, the standard deviation to mean ratio is calculated, and 1.87Mbps divided by 6.5Mbps is approximately equal to 0.29, so that the output is a bandwidth fluctuation characteristic which can reflect the fluctuation condition of network bandwidth near the live service interruption event.

Step S1242, analyzing the video decoding log segment in the live broadcast service state record, which is matched with the triggering condition of the service degradation event, identifying video frame data transmitted during the effective period of the triggering condition of the service degradation event, counting the percentage of the number of frames carrying decoding error identification in the video frame data to the total transmission frame, and generating decoding error rate features.

For example, assume that the triggering condition for a service degradation event is that the network bandwidth is below 10Mbps for 5 minutes, which condition is met for some period of time during the live broadcast. And analyzing the video decoding log segments matched with the service degradation event triggering condition, and identifying video frame data transmitted during the period that the service degradation event triggering condition is effective. Assuming that 100 frames of video frame data are transmitted in total during this period, the number of frames in which the decoding error flag is carried is counted as 5 frames by looking at the video decoding log segment. The number of frames carrying the decoding error identification in the video frame data is calculated as a percentage of the total transmission frame number, i.e. 5 frames divided by 100 frames equals 5%, whereby the output is a decoding error rate feature which can reflect the proportion of errors occurring in video decoding during service degradation.

Step S1243, traversing the hardware resource usage record of the live broadcast service state record, locating the central processing unit occupancy rate data point overlapped with the timestamp of the service interruption event, extracting a continuous time period exceeding a preset safety threshold in the data point, and calculating the weighted difference between the maximum value and the average value of the central processing unit occupancy rate in the continuous time period to generate the resource occupancy peak value feature.

For example, in the live service interruption event mentioned above, 45 minutes after the live start, the hardware resource usage record of the live service status record is traversed to find the central processor occupancy data point that overlaps the timestamp. Assuming these data points as [82%,85%,83%,81% ], the preset safety threshold is 80%, and the consecutive time period exceeding 80% is [82%,85%,83% ]. The average value over the continuous period was calculated, and the three data were added to give a total of 250%, and divided by 3 to give an average value of about 83.3% and a maximum value of 85%. Assuming that the weighting coefficient is 1, a weighted difference between the maximum value and the average value is calculated, and 85% -83.3% = 1.7%, so that the output is a resource occupation peak characteristic, which can reflect the peak condition of the central processor resource occupation at the time of the live service interruption event.

Step S1244, performing timing alignment processing on the bandwidth fluctuation feature, the decoding error rate feature and the resource occupation peak feature, so that a time interval corresponding to the bandwidth fluctuation feature, the decoding error rate feature and the resource occupation peak feature and a trigger condition of the service interruption event timestamp and the service degradation event maintain a synchronous mapping relationship, and generating a second feature set with a timing label.

For example, taking the live service interruption event of 45 minutes after the start of the live broadcast mentioned earlier, the bandwidth fluctuation feature (0.29) calculated around the time stamp, the decoding error rate feature (5%) obtained during the validation of the service degradation event trigger condition, and the resource occupancy peak feature (1.7%) overlapping the live service interruption event time stamp are time aligned. These features are mapped synchronously with the time stamp of 45 minutes and the triggering condition of service degradation event (network bandwidth below 10Mbps for 5 minutes), for example marked as 45 minutes after the start of live broadcast, bandwidth fluctuation feature of 0.29, decoding error rate feature of 5% and resource occupancy peak feature of 1.7% when network bandwidth is below 10Mbps for 5 minutes. By such a time alignment process, a second feature set with a time sequence tag is generated that can comprehensively reflect service stability related features under specific live service states (such as live service interruption and service degradation).

In one possible implementation, step S130 includes:

And S131, performing feature vector splicing on the real-time video watching features and the service stability features to generate a multidimensional feature input vector.

In this embodiment, the real-time video watching features include a user interaction frequency feature, a picture switching preference feature, a content response delay feature, and the like, and the service stability features include a bandwidth fluctuation feature, a decoding error rate feature, a resource occupation peak feature, and the like. For example, assume that the value of the user interaction frequency characteristic is [0.3] (here, a value under a certain quantization standard is represented) that the value of the picture switching preference characteristic is [0.5], the value of the content response delay characteristic is [0.2], the value of the bandwidth fluctuation characteristic is [0.15], the value of the decoding error rate characteristic is [0.05], and the value of the resource occupancy peak characteristic is [0.1]. The features are spliced together in sequence to obtain a multi-dimensional feature input vector [0.3,0.5,0.2,0.15,0.05,0.1].

Step S132, a historical feature weight template stored in a strategy matching layer in the dynamic optimization model is obtained, wherein the historical feature weight template comprises reference weight values of feature dimensions in different live broadcast scenes.

For example, for a science fiction movie live scene, assume that in the historical feature weight template, the reference weight value of the user interaction frequency feature is 0.2, the reference weight value of the picture switching preference feature is 0.3, the reference weight value of the content response delay feature is 0.1, the reference weight value of the bandwidth fluctuation feature is 0.2, the reference weight value of the decoding error rate feature is 0.1, and the reference weight value of the resource occupation peak feature is 0.1.

And step S133, calculating cosine similarity between each feature dimension in the multi-dimensional feature input vector and the historical feature weight template, generating a scene matching degree index, dynamically adjusting the reference weight value according to the scene matching degree index, generating a real-time feature weight value, and carrying out normalization processing on the real-time feature weight value to obtain feature importance distribution.

Taking the user interaction frequency characteristic as an example, the cosine similarity of the user interaction frequency characteristic and the corresponding characteristic dimension in the historical characteristic weight template is calculated. Assuming that the cosine similarity of the user interaction frequency characteristics is 0.8 after calculation (the specific calculation process involves vector calculation, and the result is described in text here). According to the same manner, the cosine similarity of other feature dimensions is calculated, the cosine similarity of the picture switching preference feature is 0.9, the cosine similarity of the content response delay feature is 0.7, the cosine similarity of the bandwidth fluctuation feature is 0.85, the cosine similarity of the decoding error rate feature is 0.75, and the cosine similarity of the resource occupation peak feature is 0.8. And generating a scene matching degree index according to the cosine similarities, and then dynamically adjusting the reference weight value. For example, for the user interaction frequency feature, the weight value adjusted according to the scene matching degree index may become 0.18 (here, the result obtained according to the scene matching degree index and the set adjustment rule). And adjusting the reference weight value of other characteristics in the same way, wherein the weight value after adjustment of the picture switching preference characteristic is 0.32, the weight value after adjustment of the content response delay characteristic is 0.09, the weight value after adjustment of the bandwidth fluctuation characteristic is 0.19, the weight value after adjustment of the decoding error rate characteristic is 0.08, and the weight value after adjustment of the resource occupation peak characteristic is 0.09. These adjusted weight values are normalized to calculate their sum as 0.18+0.32+0.09+0.19+0.08+0.09=0.95. And dividing the weight value of each adjusted weight value by the sum to obtain a normalized weight value of the user interaction frequency characteristic which is 0.18 divided by 0.95 and is approximately equal to 0.19, a weight value of the picture switching preference characteristic which is 0.32 divided by 0.95 and is approximately equal to 0.34, a weight value of the content response delay characteristic which is 0.09 divided by 0.95 and is approximately equal to 0.095, a weight value of the bandwidth fluctuation characteristic which is 0.19 divided by 0.95 and is equal to 0.2, a weight value of the decoding error rate characteristic which is 0.08 divided by 0.95 and is approximately equal to 0.084, and a weight value of the resource occupation peak value characteristic which is 0.09 divided by 0.95 and is approximately equal to 0.095, wherein the normalized weight values form the characteristic importance distribution.

And step S134, carrying out weighted fusion on the multidimensional feature input vector according to the feature importance distribution to generate an optimized decision vector.

For example, each feature in the multi-dimensional feature input vector is multiplied by a corresponding weight value according to the feature importance distribution obtained previously, and then added to obtain the optimal decision vector. I.e. (0.3×0.19)+(0.5×0.34)+(0.2×0.095)+(0.15×0.2)+(0.05×0.084)+(0.1×0.095)＝0.057+0.17+0.019+0.03+0.0042+0.0095＝0.2897,, 0.2897 is a quantized representation of the optimized decision vector (in practice, it may be a multidimensional vector, here abbreviated as a numerical value representing its calculation).

And S135, inputting the optimized decision vector to a labeling output layer of the dynamic optimization model, and generating an optimized labeling result comprising a picture quality optimization direction, a bandwidth allocation optimization direction and a decoding priority optimization direction.

For example, at the label output layer, the picture quality optimization direction is determined according to the value of the optimization decision vector. If the value of the optimization decision vector satisfies the set condition (here, according to a predefined model rule), it may be determined that the picture quality needs to be improved in sharpness, which is the picture quality optimization direction. For the bandwidth allocation optimization direction, it may be determined that bandwidth allocation needs to be increased according to the value of the optimization decision vector. For the decoding priority optimization direction, it may be determined that the decoding priority of the special effect scene needs to be improved, and the like. In this way, optimized labeling results for the science fiction movie live data stream are comprehensively generated, and the results are used for subsequent live service adjustment strategy formulation.

In one possible implementation, step S140 includes:

Step S141, analyzing the picture resolution adjustment parameter, the code rate control parameter and the buffer interval configuration parameter included in the live broadcast service adjustment policy.

For example, assume that in the live service adjustment policy, the picture resolution adjustment parameter is increased from 720p to 1080p, the rate control parameter specifies a maximum allowable rate of 5Mbps, a minimum guaranteed rate of 2Mbps, and the buffer interval configuration parameter is set to increase the initial buffer threshold from 10 seconds to 15 seconds.

Step S142, the picture resolution adjustment parameter is sent to the video encoding module, and the resolution preset value of the video encoder is triggered to be dynamically adjusted.

In this embodiment, the video encoding module, after receiving the parameter for increasing the resolution of the picture from 720p to 1080p, adjusts the preset resolution value of the video encoder according to the instruction. In the process of the live broadcast of the science fiction movie, the video encoder originally encodes the video stream according to the preset resolution value of 720p, and after receiving the new picture resolution adjustment parameter, the preset resolution value of the encoding process is changed to 1080p, so that the resolution of the output video stream is changed to adapt to the optimized live broadcast service requirement.

Step S143, synchronizing the code rate control parameters to a bandwidth allocation module, and triggering a dynamic code rate self-adaptive algorithm based on the current network condition.

Wherein, step S143 includes:

Step S1431, monitoring the real-time network throughput and the data packet loss rate of the live service node, and generating a network condition evaluation index.

For example, during a science fiction movie live, network conditions of the live service node are continuously monitored. The real-time network throughput is assumed to be 3Mbps by monitoring, and the data packet loss rate is assumed to be 1%. And according to a preset rule, integrating the two data to generate a network condition evaluation index. For example, a simple evaluation rule may be set, where the network condition evaluation index=network throughput- (packet loss rate×10) (this is just an example rule, which may be more complex in practice), and the network condition evaluation index is calculated according to the rule to be 3Mbps- (1% ×10) =3 Mbps-0.1 mbps=2.9 Mbps.

And step S1432, calculating a recommended code rate interval under the current network condition according to the maximum allowable code rate and the minimum guaranteed code rate in the code rate control parameters.

For example, the maximum allowable code rate in the known code rate control parameters is 5Mbps, the minimum guaranteed code rate is 2Mbps, and the recommended code rate interval is calculated by combining the network condition evaluation index 2.9Mbps obtained by the previous calculation. Because the network condition evaluation index 2.9Mbps is greater than the minimum guaranteed bit rate 2Mbps, the recommended bit rate interval is [2Mbps,2.9Mbps ].

And step S1433, calling a code rate self-adaptive controller to dynamically adjust the video transmission code rate in the recommended code rate interval so that the actual code rate and the network throughput keep a preset proportional relation.

Assuming that the preset proportional relationship is that the actual code rate is 80% of the network throughput (this is the proportional relationship set according to the requirement of the live broadcast service), and the current network throughput is 3Mbps, then the actual code rate is calculated according to the proportional relationship to be 3mbps×80% =2.4 Mbps, and the code rate adaptive controller adjusts the video transmission code rate to be 2.4Mbps, so that a proper video transmission code rate is provided on the premise of meeting the network condition.

In step S1434, when the packet loss rate is detected to exceed the threshold, an emergency rate reduction operation is triggered and the forward error correction coding mechanism is enabled.

For example, assuming that the set packet loss rate threshold is 3%, if the packet loss rate is detected to exceed 3%, for example, to reach 5%, during the live broadcast, an emergency rate-reducing operation is triggered. Assuming that the code rate needs to be reduced to 50% of the current code rate according to a preset urgent code rate reduction rule (this is an example rule, and is actually determined according to a specific policy), the current code rate is 2.4Mbps, and then the code rate after urgent code rate reduction is 2.4mbps×50% =1.2 Mbps, and meanwhile, a forward error correction coding mechanism is enabled to reduce the influence of data packet loss on video playing quality.

Step S144, the buffer interval configuration parameters are deployed to a data buffer module, and initial buffer threshold values are triggered to be adjusted according to the performance of the user equipment.

For example, in a science fiction movie live broadcast scenario, the data buffer module may adjust according to the performance of the user equipment after receiving the buffer configuration parameter that increases the initial buffer threshold from 10 seconds to 15 seconds. Assuming that the memory capacity of the ue is large and the processing power is high, the data buffer module may directly set the initial buffer threshold to 15 seconds. If the performance of the ue is poor, for example, the memory capacity is small, the data buffer module may set an initial buffer threshold as close as possible to 15 seconds, for example, 13 seconds, on the premise of ensuring the smoothness of video playing according to a set algorithm (for example, comprehensive calculation according to the remaining memory of the device and factors such as video code rate). In such a way, the initial buffer threshold is reasonably adjusted according to the performance of the user equipment so as to optimize the playing experience of the live broadcast service at the user side.

In one possible implementation, step S150 includes:

Step S151, collecting the user side play fluency index, the picture quality score and the interactive response delay index in a preset observation period.

In this embodiment, it is assumed that the preset observation period is a period from 30 minutes after the start of live broadcast to the end of live broadcast. For the user side playing fluency index, data is collected through a monitoring tool installed at the user equipment side, such as information of the katon condition and the like in the video playing process is recorded, the picture quality scoring is the scoring given by the user in the watching process according to the subjective feeling of the user, the value range can be 1-10 minutes, the scoring data is collected in real time in the live broadcasting process, the interactive response delay index refers to the time interval from the user sending operation (such as pause, fast forward and the like) to the server responding operation and the operation result is seen at the user side, and the interactive response delay index is continuously collected in the observing period.

Step S152, performing a click event frequency statistics on the play fluency index, and generating a click frequency feature.

For example, in the observation period from 30 minutes to the end of live broadcast, the smoothness index data is carefully checked, and the occurrence times of the katon event are counted. Assuming that a total of 5 stuck events are found to occur, and the total duration of the observation period is 30 minutes, the stuck frequency characteristic is the number of stuck events divided by the total duration, i.e., 5 times divided by 30 minutes, and about 1 stuck event per 6 minutes occurs (the calculation result here indicates the stuck frequency characteristic).

And step S153, performing time series analysis on the picture quality scores to generate quality fluctuation characteristics.

For example, the picture quality scoring data collected over the observation period may be viewed, e.g., 8 score between 30 minutes and 40 minutes, 7 score between 40 minutes and 50 minutes, 8 score between 50 minutes and 60 minutes, etc. The change condition of the scores with time is analyzed, the difference value of the scores of adjacent time periods is calculated, for example, 8 minutes to 7 minutes = 1 minute, 7 minutes to 8 minutes = 1 minute, and the quality fluctuation feature is determined by the size and the change trend of the difference values. If the score difference is large and the change is frequent, the quality fluctuation is large, and if the score difference is small and relatively stable, the quality fluctuation is small.

And step S154, performing percentile calculation on the interactive response delay index to generate delay distribution characteristics.

For example, all of the interactive response delay data may be collected over the observation period, and given that there are 20 of these interactive response delay data, these interactive response delay data may be sorted from small to large, and then the delay value corresponding to the set percentile (e.g., 90% percentile) is calculated. If the 18 th data (20×90% =18, rounded up) has a value of 2 seconds, then the 90% percentile has an interactive response delay of 2 seconds, which is an important manifestation of the delay profile.

Step S155, calculating the difference degree between the characteristic of the cartoon frequency and the historical cartoon baseline data, generating a first loss component, analyzing the deviation degree of the characteristic of the quality fluctuation and the expected quality stability curve, generating a second loss component, and counting the proportion of high-delay samples in the delay distribution characteristic, so as to generate a third loss component.

For example, assuming that the historical stuck baseline data is 1 stuck every 10 minutes, the current stuck frequency is characterized by 1 stuck every 6 minutes. When calculating the degree of difference, the historical stuck baseline data is subtracted from the current stuck frequency, i.e., (1/6-1/10). First, the fractional division is changed from 1/6 to 5/30 and 1/10 to 3/30, and then (5/30-3/30) =2/30=1/15, which is the first loss component. For the second loss component, the expected quality stability curve indicates that the picture quality score should ideally remain relatively stable with minimal fluctuations. If the actual quality fluctuation feature shows that the score fluctuation is large, the deviation from the expected quality stability curve is obtained by comparing the deviation degree between the two (the specific calculation mode is according to the set deviation degree calculation rule, and the deviation degree value is 0.2 by calculating the sum of the square difference value of the fluctuation score and the stability score and the like, which is assumed to be the second loss component). For the third loss component, assuming that the high delay is defined as a delay greater than 3 seconds, in the delay profile calculated above, if there are 3 data greater than 3 seconds for a total of 20 data, the proportion of high delay samples is 3/20=0.15, which is the third loss component.

Step S156, performing weighted summation on the first loss component, the second loss component and the third loss component to generate a comprehensive loss function, calculating a partial derivative of the model weight of the comprehensive loss function through a back propagation algorithm, generating a model parameter adjustment gradient, and updating the weight parameters of the dynamic optimization model based on the model parameter adjustment gradient.

For example, assume that the weight of the first loss component is 0.3, the weight of the second loss component is 0.4, and the weight of the third loss component is 0.3. The calculated composite loss function is (1/15×0.3+0.2×0.4+0.15×0.3). The multiplication section is calculated first, 1/15×0.3= 0.02,0.2 ×0.4= 0.08,0.15 ×0.3=0.045, and then the results are added, 0.02+0.08+0.045=0.145, this 0.145 being the value of the integrated loss function. And calculating the partial derivative of the comprehensive loss function to the model weight through a back propagation algorithm, and generating a model parameter adjustment gradient. The back propagation algorithm calculates the partial derivatives corresponding to each model weight based on the relationship between the integrated loss function and the model weight (the calculation process involves complex mathematical principles and algorithm operations, which are mainly emphasized here on the basis of the previously calculated integrated loss function), which constitute the model parameter adjustment gradient. Finally, the weight parameter of the dynamic optimization model is updated based on the model parameter adjustment gradient, for example, if the partial derivative corresponding to a certain model weight is 0.05, according to a set updating rule (for example, the weight is updated according to a set learning rate multiplied by the partial derivative), and if the learning rate is 0.1, the updating amount of the weight parameter is 0.05x0.1=0.005, and the updating amount is applied to the original weight parameter, so that the updating of the dynamic optimization model weight parameter is completed, and the model can continuously optimize itself according to the real-time live broadcast quality index, so that the optimization effect of the subsequent live broadcast service is improved.

Fig. 2 shows a schematic diagram of exemplary hardware and software components of a big data analysis based theatre live optimization system 100 that may implement the concepts of the present application, provided by some embodiments of the present application. For example, the processor 120 may be used on the theatre live optimization system 100 based on big data analysis and to perform the functions of the present application.

The big data analysis based cinema live broadcast optimization system 100 may be a general purpose server or a special purpose server, both of which may be used to implement the big data analysis based cinema live broadcast optimization method of the present application. Although only one server is shown, the functionality described herein may be implemented in a distributed fashion across multiple similar platforms for convenience to balance processing loads.

For example, the big data analysis based theatre live optimization system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and a different form of storage medium 140, such as a disk, ROM, or RAM, or any combination thereof. Illustratively, the big data analysis based theatre live optimization system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application may be implemented in accordance with these program instructions. The big data analysis based theatre live optimization system 100 also includes an Input/Output (I/O) interface 150 between the computer and other Input/Output devices.

For ease of illustration, only one processor is depicted in the theatre live optimization system 100 based on big data analysis. It should be noted, however, that the theatre live broadcast optimization system 100 based on big data analysis in the present application may also include multiple processors, and thus the steps performed by one processor described in the present application may also be performed jointly or separately by multiple processors. For example, if the processor of the theatre live optimization system 100 based on big data analysis performs steps a and B, it should be understood that steps a and B may also be performed by two different processors together or in one processor alone. For example, the first processor performs step a, the second processor performs step B, or the first processor and the second processor together perform steps a and B.

In addition, the embodiment of the invention also provides a readable storage medium, wherein computer executable instructions are preset in the readable storage medium, and when a processor executes the computer executable instructions, the cinema live broadcast optimization method based on big data analysis is realized.

It should be noted that in order to simplify the presentation of the disclosure and thereby aid in understanding one or more embodiments of the invention, various features are sometimes grouped together in a single embodiment, figure, or description thereof.

Claims (10)

1. A method for optimizing cinema live broadcast based on big data analysis, characterized in that the method comprises: Obtaining historical live broadcast big data of a target cinema, wherein the historical live broadcast big data includes multiple live broadcast data streams, each of which is composed of at least one user viewing behavior record and a corresponding live broadcast service status record; Perform multi-dimensional viewing feature extraction processing on the historical live broadcast big data to obtain real-time viewing features and service stability features of each live broadcast data stream; Based on the pre-trained dynamic optimization model, dynamic strategy matching processing is performed on the real-time viewing feature and the service stability feature to generate an optimized annotation result of the live data stream, and a live service adjustment strategy is generated according to the optimized annotation result; Synchronize the live broadcast service adjustment strategy to the live broadcast service node in real time, and trigger the live broadcast service node to adjust the current live broadcast parameter configuration according to the live broadcast service adjustment strategy; Based on the adjusted live broadcast parameter configuration, a real-time live broadcast quality index is collected, and the parameters of the dynamic optimization model are updated according to the real-time live broadcast quality index. 2. The method for optimizing live streaming of cinemas based on big data analysis according to claim 1 is characterized in that the multi-dimensional viewing feature extraction process is performed on the historical live streaming big data to obtain the real-time viewing feature and service stability feature of each live streaming data stream, including: Performing time window division processing on the user viewing behavior records in the live data stream to obtain viewing behavior segments of multiple continuous time periods; Calling a pre-trained real-time feature encoder to extract real-time viewing features from the viewing behavior segment, and generating a first feature set including user interaction frequency features, screen switching preference features, and content response delay features; Performing abnormal event detection processing on the live broadcast service status record to identify the timestamp of the live broadcast service interruption event and the triggering condition of the service degradation event; Performing a stability quantitative analysis on the live broadcast service status record based on the timestamp and the trigger condition to generate a second feature set including bandwidth fluctuation features, decoding error rate features, and resource occupancy peak features; The first feature set and the second feature set are subjected to feature fusion to obtain the real-time viewing feature and the service stability feature. 3. The method for optimizing live streaming of cinemas based on big data analysis according to claim 2 is characterized in that the calling of a pre-trained real-time feature encoder extracts real-time viewing features from the viewing behavior segment to generate a first feature set including user interaction frequency features, screen switching preference features, and content response delay features, including: Calling a pre-trained real-time feature encoder to perform behavior pattern analysis on the user operation log in the movie-watching behavior segment, extracting the number of user interaction operations and the operation interval duration in a single time window, and generating user interaction frequency features; Identify the timing relationship between the screen switching request time point and the live content key frame in the user operation log, calculate the time deviation between the screen switching request and the content key frame, and generate a screen switching preference feature; The difference between the content loading request and the server response timestamp in the user operation log is monitored, and the number of abnormal events in which the content loading delay exceeds a preset threshold is counted to generate a content response delay feature. 4. The method for optimizing live streaming of cinemas based on big data analysis according to claim 3 is characterized in that the identifying the timing relationship between the screen switching request time point in the user operation log and the key frame of the live streaming content, calculating the time deviation between the screen switching request and the key frame of the content, and generating the screen switching preference feature comprises: Extracting the timestamp data of the screen switching operation triggering time from the user operation log, and generating a screen switching request time point set; Parse the key frame types and corresponding timestamp sequences pre-annotated in the live content metadata, and generate a set of live content key frame time points including scene switching key frames, special effect start key frames, and plot turning key frames; Traversing each screen switching request time point in the screen switching request time point set, searching for a target key frame time point having the smallest time difference with the current screen switching request time point in the live content key frame time point set; Calculating the absolute time difference between the screen switching request time point and the target key frame time point to generate an original time deviation of a single screen switching request time point; The original time deviation is annotated with positive and negative signs, when the screen switching request time point is later than the target key frame time point, it is annotated as a positive value, and when it is earlier than the target key frame time point, it is annotated as a negative value, and a time deviation sequence with a timing direction mark is generated; Counting a first frequency in a preset positive deviation interval, a second frequency in a preset negative deviation interval, and a third frequency in a zero deviation tolerance interval in the time deviation sequence; Generate a user's active switching tendency index for scene switching key frames, a delay follow-up index for special effect start key frames, and a predicted switching density index for plot turning key frames according to the proportional relationship among the first frequency, the second frequency, and the third frequency; The active switching tendency index, the delayed following index and the predicted switching density index are weighted according to the key frame type to generate a screen switching preference feature set including a time sensitivity weight and a content relevance weight. 5. The method for optimizing live streaming of cinemas based on big data analysis according to claim 4 is characterized in that the key frame types and corresponding timestamp sequences pre-annotated in the metadata of the live streaming content are parsed to generate a set of key frame time points of the live streaming content including scene switching key frames, special effect start key frames and plot turning key frames, including: Read the metadata marker block embedded in the live content video stream, and extract the frame number, frame type label and offset relative to the live start time of each key frame; Classifying the key frames into scene switching key frames, special effect start key frames and plot turning key frames according to the frame type labels; Converting the frame number of each key frame into absolute timestamp data based on the offset to generate a set of key frame time points of the live content arranged in chronological order; Verify the timestamp continuity of adjacent key frames in the live content key frame time point set. If it is detected that the timestamp jump exceeds a preset frame interval threshold, insert a virtual key frame time point in the jump interval and mark it as an unclassified key frame type. 6. The method for optimizing live streaming of cinemas based on big data analysis according to claim 2, characterized in that the stability quantitative analysis of the live streaming service status record is performed based on the timestamp and the trigger condition to generate a second feature set including bandwidth fluctuation features, decoding error rate features and resource occupancy peak features, including: Extracting a network bandwidth sampling value sequence associated with the timestamp of the live service interruption event from the live service status record, screening out bandwidth sampling data within a set time range before and after each live service interruption event, calculating a ratio of a standard deviation to a mean of the bandwidth sampling data, and generating a bandwidth fluctuation feature; Parsing the video decoding log segment matching the trigger condition of the service degradation event in the live broadcast service status record, identifying the video frame data transmitted during the period when the trigger condition of the service degradation event is effective, counting the percentage of the number of frames carrying decoding error identifiers in the video frame data to the total number of transmitted frames, and generating a decoding error rate feature; Traversing the hardware resource usage records of the live service status records, locating the CPU occupancy data points that overlap with the timestamp of the service interruption event, extracting the continuous time periods that exceed the preset safety threshold in the data points, calculating the weighted difference between the maximum value and the average value of the CPU occupancy in the continuous time period, and generating resource occupancy peak features; The bandwidth fluctuation characteristics, decoding error rate characteristics and resource occupancy peak characteristics are time-aligned to ensure that the time intervals corresponding to the bandwidth fluctuation characteristics, decoding error rate characteristics and resource occupancy peak characteristics maintain a synchronous mapping relationship with the timestamp of the service interruption event and the triggering condition of the service degradation event, thereby generating a second feature set with a time sequence label. 7. The method for optimizing live streaming of cinemas based on big data analysis according to claim 1, characterized in that the dynamic optimization model based on pre-training performs dynamic strategy matching processing on the real-time viewing features and the service stability features to generate an optimized annotation result of the live streaming data stream, including: Concatenate the real-time viewing feature and the service stability feature into feature vectors to generate a multi-dimensional feature input vector; Obtaining a historical feature weight template stored in a strategy matching layer in the dynamic optimization model, wherein the historical feature weight template includes a benchmark weight value of each feature dimension under different live broadcast scenarios; Calculating the cosine similarity between each feature dimension in the multidimensional feature input vector and the historical feature weight template to generate a scene matching index, dynamically adjusting the reference weight value according to the scene matching index to generate a real-time feature weight value, and normalizing the real-time feature weight value to obtain a feature importance distribution; Performing weighted fusion on the multi-dimensional feature input vector according to the feature importance distribution to generate an optimized decision vector; The optimization decision vector is input into the annotation output layer of the dynamic optimization model to generate an optimization annotation result including a picture quality optimization direction, a bandwidth allocation optimization direction and a decoding priority optimization direction. 8. The method for optimizing live streaming of cinemas based on big data analysis according to claim 1, characterized in that the step of synchronizing the live streaming service adjustment strategy to the live streaming service node in real time and triggering the live streaming service node to adjust the current live streaming parameter configuration according to the live streaming service adjustment strategy comprises: Parsing the picture resolution adjustment parameters, bit rate control parameters and buffer space configuration parameters included in the live broadcast service adjustment strategy; Sending the picture resolution adjustment parameter to the video encoding module to trigger dynamic adjustment of the resolution preset value of the video encoder; Synchronize the bit rate control parameters to the bandwidth allocation module to trigger a dynamic bit rate adaptation algorithm based on the current network status; Deploy the buffer zone configuration parameters to a data cache module, triggering adjustment of an initial buffer threshold according to user equipment performance; The step of synchronizing the rate control parameters to the bandwidth allocation module and triggering a dynamic rate adaptation algorithm based on the current network status includes: Monitor the real-time network throughput and packet loss rate of live broadcast service nodes and generate network status evaluation indicators; Calculate the recommended bit rate range under the current network conditions according to the maximum allowed bit rate and the minimum guaranteed bit rate in the bit rate control parameters; Calling a bit rate adaptive controller to dynamically adjust the video transmission bit rate within the recommended bit rate range so that the actual bit rate and the network throughput maintain a preset proportional relationship; When it is detected that the packet loss rate exceeds the threshold, an emergency bit rate reduction operation is triggered and the forward error correction coding mechanism is enabled. 9. The method for optimizing live streaming of cinemas based on big data analysis according to claim 1, characterized in that the real-time live streaming quality index is collected based on the adjusted live streaming parameter configuration, and the parameters of the dynamic optimization model are updated according to the real-time live streaming quality index, including: Collect the user-side playback fluency index, picture quality score and interactive response delay index within the preset observation period; Performing frequency statistics of freeze events on the playback smoothness index to generate freeze frequency features; Performing time series analysis on the picture quality scores to generate quality fluctuation characteristics; Calculating percentiles of the interactive response delay indicator to generate a delay distribution feature; Calculate the difference between the jam frequency feature and the historical jam baseline data to generate a first loss component, analyze the degree of deviation between the quality fluctuation feature and the expected quality stability curve to generate a second loss component, and count the proportion of high-delay samples in the delay distribution feature to generate a third loss component; Perform a weighted summation on the first loss component, the second loss component and the third loss component to generate a comprehensive loss function, calculate the partial derivative of the comprehensive loss function with respect to the model weight through a back propagation algorithm, generate a model parameter adjustment gradient, and update the weight parameters of the dynamic optimization model based on the model parameter adjustment gradient. 10. A cinema live broadcast optimization system based on big data analysis, characterized in that the cinema live broadcast optimization system based on big data analysis includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the cinema live broadcast optimization method based on big data analysis as described in any one of claims 1 to 9.