A multimedia video stream management system and method based on artificial intelligence

Abstract

The present invention discloses a multimedia video stream management system and method based on artificial intelligence, and relates to the technical field of video stream label processing. The present invention combs the reference element information based on which the AI ​​module divides the video label types of any video stream, and extracts a number of neighboring labels based on data analysis for the phenomenon of neighboring judgment when the AI ​​module performs differential classification. The present invention monitors whether there is data deviation in the classification result output by the AI ​​module within the range of video labels that are mutually neighboring labels, thereby providing a judgment basis for whether the AI ​​module needs to be optimized. The present invention can realize adaptive monitoring of the performance of the AI ​​module in the field of multimedia video stream label processing, and effectively improves the role of AI in the discoverability of multimedia video content and user experience.

Description

Technical Field

[0001] The present invention relates to the technical field of video stream label processing, and in particular to a multimedia video stream management system and method based on artificial intelligence. Background Art

[0002] ‌AI labeling technology in multimedia video streams‌ refers to the process of automatically labeling and classifying video content using artificial intelligence and machine learning algorithms. Specifically, in the process of labeling multimedia video streams, AI first analyzes the video content through computer vision and deep learning technology, identifies key elements in the video (such as people, objects, scenes, etc.), and then generates labels based on these elements. These labels can cover multiple aspects such as video content description, emotional color, and theme classification.

[0003] The application scenarios of AI labeling technology in multimedia video streams are very extensive. First, in terms of personalized recommendations, by analyzing video content to generate labels and combining them with user interests and preferences, accurate personalized recommendations can be achieved. Secondly, in the field of content review, AI can help detect illegal content and ensure the security and compliance of the platform. Finally, in terms of media asset management, automatically generated labels can greatly improve the retrieval efficiency of media assets and facilitate users to quickly find the required video materials. For example, AI technology can be used to identify and search for repeated or similar clips in videos, which is suitable for scenarios such as original identification and video duplication detection.

[0004] Therefore, it can be seen that the accuracy of AI's attribute label judgment on video streams will directly affect the effective use of AI processing results in reality. However, some data bias problems often occur in the process of AI labeling video streams. This is mainly because the data set used in training the AI ​​model is unbalanced or incomplete in some specific aspects, resulting in errors in the model's prediction results for certain specific groups or situations. The above-mentioned imbalance or incompleteness is also a reflection of the lack of diversity in sample data. Summary of the invention

[0005] The purpose of the present invention is to provide a multimedia video stream management system and method based on artificial intelligence to solve the problems raised in the prior art.

[0006] To achieve the above object, the present invention provides the following technical solution: a multimedia video stream management method based on artificial intelligence, the method comprising:

[0007] Step S1: The multimedia video stream management cloud platform calls the AI ​​module to collect process data generated when each video stream segment received from the multimedia port generates a corresponding video tag, and sorts out the reference element information based on which the AI ​​module classifies the video tag types of any video stream segment;

[0008] Step S2: By comparing the deviation distribution of the reference element information presented between any two different video labels, any two video labels that have adjacent judgments when the AI ​​module performs differential classification are judged and identified, and several pairs of video label groups that are adjacent to each other are extracted;

[0009] Step S3: extracting the distribution of the distinguishing reference element information presented in the process of the AI ​​module judging the video label of the corresponding video stream segment within the range of the corresponding mutually adjacent labels, and calculating the mutual proximity value for each pair of video label groups of mutually adjacent labels;

[0010] Step S4: In the multimedia video stream management cloud platform, the call rate distribution of users for the video stream segments stored in each storage area is monitored, and the video tags of the video stream segments with abnormal call rates are adjusted by referring to the corresponding adjacent tags. According to the corresponding call rate changes brought about by the video tag adjustment, it is determined whether to send an early warning prompt that the AI ​​module needs to be optimized.

[0011] Preferably, step S1 comprises:

[0012] Step S1-1: The multimedia video stream management cloud platform collects all feature element information extracted after image recognition and audio analysis for each video stream segment before calling the AI ​​module to generate a corresponding video tag for each video stream segment received from the multimedia port, and collects and generates a feature element information set corresponding to each video stream segment;

[0013] According to the above content, the characteristic element information set corresponding to each video stream segment is the information set that the AI ​​module finally refers to when analyzing and generating a certain exact video label for the video stream segment;

[0014] Step S1-2: Collect the characteristic element information sets of all video stream segments with the same corresponding video tags stored in the multimedia video stream management cloud platform, and obtain several reference element information sets based on which the AI ​​module classifies the video tag types of any video stream segments; wherein the video stream segments with the same video tags are placed in a same storage area in the multimedia video stream management cloud platform for data storage; wherein one reference element information set corresponds to one video tag.

[0015] Preferably: step S2 comprises:

[0016] Step S2-1: extracting overlapping feature element information from reference element information sets corresponding to any two different video tags one by one, and generating overlapping reference element information sets corresponding to any two different video tags respectively;

[0017] Step S2-2: Obtain the total value of the characteristic element information contained in each overlapping reference element information set corresponding to each type of video label and other types of video labels; if the total value of the characteristic element information contained in the overlapping reference element information set between any two different video labels is greater than the total threshold, it is determined that the AI ​​module has a proximity judgment when distinguishing and classifying any two different video labels, and it is determined that the any two different video labels are a pair of video label groups that are adjacent to each other.

[0018] According to the above content, the video labels of adjacent labels satisfy a certain number of overlapping feature element information, that is to say, when the AI ​​module determines which specific video label of a certain video stream segment belongs to, it uses relatively insignificant distinguishing feature elements to determine the final output label division result, that is, when judging which category of adjacent labels a video stream segment containing corresponding overlapping feature elements belongs to, the recognition accuracy requirement of the AI ​​module is relatively high. If the AI ​​module has recognition data deviation, there is a high possibility of error in distinguishing and classifying video labels of adjacent labels.

[0019] Preferably, step S3 includes:

[0020] Step S3-1: If video tag A and video tag B are adjacent tags, extract the reference element information set P(A) of video tag A, the reference element information set P(B) of video tag B, and the overlapping reference element information set U between video tag A and video tag B. a,b ; Extract the video tag A to obtain the distinguishing reference element information set C1=P(A)-U a,b , extract the video tag B to obtain the distinguishing reference element information set C2=P(B)-U a,b ;

[0021] The above-mentioned distinguishing reference element information sets obtained for the video label A and the video label B respectively are often the feature element information used by the AI ​​module to narrow the matching degree between the video label A or the video label B in the process of judging whether the video stream segment contains the overlapping reference element information of the video label A and the video label B in the corresponding feature element information set. It can also be said that in the overlapping reference element information set U a,bThe characteristic element information contained in represents the common element information between video tag A and video tag B, the characteristic element information contained in the distinguishing reference element information set obtained for video tag A represents the individual element information of video tag A, and the characteristic element information contained in the distinguishing reference element information set obtained for video tag B represents the individual element information of video tag B;

[0022] Step S3-2: Collect all video stream segments marked with video tag A from the multimedia video stream management cloud platform to obtain a first video stream segment set Y1, and collect all video stream segments marked with video tag B to obtain a second video stream segment set Y2; if a certain video stream segment exists in the first video stream segment set Y1 or the second video stream segment set Y2, and the feature element information set R extracted from the certain video stream segment satisfies R∩U a,b =Q≠ , where Q represents the feature element information set R and the overlapping reference element information set U a,b If the intersection between them is , a certain video stream segment is marked as a feature, and at the same time, a target distinguishing element information set Q'=RQ is extracted from a certain video stream segment;

[0023] Step S3-3: In the first video stream segment set Y1, the target distinguishing element information set extracted from all the feature-marked video stream segments is collected, and the number of types of feature element information appearing is accumulated. 1 In the second video stream segment set Y2, the target distinguishing element information set extracted from all the feature-marked video stream segments is collected, and the number of types of feature element information appearing is accumulated. 2 ; Calculate the first proximity index β between video label A and video label B 1 =[η 1 / card(C1)+η 2 / card(C2)] / 2;

[0024] η 1 / card(C1) value or η 2The larger the value of / card(C2), the more evenly the reference element information involved is distributed when the video stream segment containing the overlapping reference element information is judged as video label A or video label B, that is, the lower the difficulty of the AI ​​module in making a distinction and adjacent division in the process of excluding the conclusion belonging to video label B or video label A based on the reference element information of video label A or video label B, that is, the video stream segment with the reference element information of video label A or video label B is sufficiently different from the video feature gap between the video stream segment corresponding to video label B or video label A, that is, when the AI ​​module performs the classification work of video label A and video label B on the video stream segment, the actual proximity degree at the data judgment and analysis level is not high, and the possibility of error in the AI ​​module making a distinction and classification between video label A and video label B is small;

[0025] Step S3-4: Obtain the number of video stream segments marked as features in the first video stream segment set Y1, and the number of video stream segments marked as features in the second video stream segment set Y2. The proportion value α1; calculate the second proximity index β between the video label A and the video label B 2 =(α1+α2) / 2;

[0026] The higher the ratio α1 or the ratio α2, the higher the ratio of the video stream segments containing the overlapping reference element information between the video label A and the video label B in the video stream segments determined to correspond to the video label A or the video label B, and the more frequently the AI ​​module makes a distinction between the video label A and the video label B; the more likely it is that the AI ​​module will make an error when distinguishing and classifying between the video label A and the video label B;

[0027] Step S3-5: Calculate the similarity of the feature element information set corresponding to each video stream segment in the first video stream segment set Y1 with the feature element information set corresponding to each video stream segment in the second video stream segment set Y2, and capture the highest similarity value δ; calculate the proximity value ξ=(1 / β 1 +β 2 )×δ.

[0028] Preferably, step S4 comprises:

[0029] Step S4-1: In each storage area, the average call rate of all stored video stream segments by users is monitored; if in a storage area corresponding to a video tag a, there is a video stream segment whose corresponding call rate is lower than the average call rate, the video tag b with the highest proximity value to the video tag a is captured, and the video tag of the video stream segment is adjusted from a to b;

[0030] Step S4-2: monitor the call rate of a certain video stream segment presented by the user within a unit period of time after the video tag of a certain video stream segment is adjusted from a to b. If the call rate shows a numerical increase compared with that before the adjustment, generate an early warning signal for the optimization of the AI ​​module. If the call rate does not show a numerical increase compared with that before the adjustment, cancel the adjustment.

[0031] Step S4-3: When the cumulative number of warning signals generated for the AI ​​module optimization is greater than the number threshold, a warning prompt that the AI ​​module needs to be optimized is sent to the management terminal.

[0032] In order to better implement the above method, a multimedia video stream management system is also proposed, which includes: a video stream label processing data management module, a neighboring label judgment management module, a neighboring degree value calculation management module, and an AI module optimization prompt management module;

[0033] The video stream label processing data management module is used to collect the process data generated when the multimedia video stream management cloud platform generates corresponding video labels for each video stream segment received from the multimedia port by calling the AI ​​module, and to sort out the reference element information based on which the AI ​​module classifies the video labels of any video stream segment;

[0034] The adjacent label judgment management module is used to judge and identify any two video labels that have adjacent judgments when the AI ​​module performs differential classification by comparing the deviation distribution of the reference element information presented between any two different video labels, and extract several pairs of video label groups that are adjacent to each other;

[0035] A proximity value calculation management module is used to extract the distribution of the distinguishing reference element information presented in the process of the AI ​​module judging the video label of the corresponding video stream segment within the range of the corresponding mutually adjacent labels, and calculate the mutual proximity value between each pair of video label groups of mutually adjacent labels;

[0036] The AI ​​module optimization prompt management module is used to monitor the distribution of users' call rates of video stream segments stored in each storage area in the multimedia video stream management cloud platform, and adjust the video tags of video stream segments with abnormal call rates with reference to corresponding adjacent tags. According to the corresponding call rate changes brought about by the video tag adjustment, it is determined whether to send an early warning prompt that the AI ​​module needs to be optimized.

[0037] Preferably, the neighboring tag judgment management module includes: a feature element information combing unit and a neighboring tag judgment unit;

[0038] A characteristic element information combing unit, used to compare the deviation distribution of reference element information presented between any two different video tags;

[0039] The adjacent label judgment unit is used to judge and identify any two video labels that have adjacent judgments when the AI ​​module performs differential classification, and extract several pairs of video label groups that are adjacent to each other.

[0040] Preferably, the AI ​​module optimization prompt management module includes a label adjustment monitoring management unit and an optimization early warning prompt management unit;

[0041] The label adjustment monitoring management unit is used to monitor the distribution of the calling rate of the video stream segments stored in each storage area by users in the multimedia video stream management cloud platform, and adjust the video labels of the video stream segments with abnormal calling rate by referring to the corresponding adjacent labels;

[0042] The optimization early warning prompt management unit is used to determine whether to send an early warning prompt that the AI ​​module needs to be optimized based on the corresponding call rate changes brought about by the adjustment of video tags.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention combs through the reference element information that the AI ​​module relies on when classifying video labels for any video stream, and extracts several neighboring labels based on data analysis for the phenomenon that the AI ​​module has neighboring judgments when performing differential classification. The present invention monitors whether there is data deviation in the classification results output by the AI ​​module within the range of video labels that are mutually adjacent labels, thereby providing a basis for judging whether the AI ​​module needs to be optimized. The present invention can realize adaptive monitoring of the performance of the AI ​​module in the field of multimedia video stream label processing, and effectively improves the role of AI in the discoverability of multimedia video content and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The present invention is a flowchart of a multimedia video stream management method based on artificial intelligence.

[0045] Figure 2 The present invention is a structural schematic diagram of an artificial intelligence-based multimedia video stream management system. DETAILED DESCRIPTION

[0046] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0047] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution, a multimedia video stream management method based on artificial intelligence, the method comprising:

[0048] Step S1: The multimedia video stream management cloud platform calls the AI ​​module to collect process data generated when each video stream segment received from the multimedia port generates a corresponding video tag, and sorts out the reference element information based on which the AI ​​module classifies the video tag types of any video stream segment;

[0049] Wherein, step S1 comprises:

[0050] Step S1-1: The multimedia video stream management cloud platform collects all feature element information extracted after image recognition and audio analysis for each video stream segment before calling the AI ​​module to generate a corresponding video tag for each video stream segment received from the multimedia port, and collects and generates a feature element information set corresponding to each video stream segment;

[0051] Step S1-2: Collecting the characteristic element information sets of all video stream segments with the same corresponding video tags stored in the multimedia video stream management cloud platform, respectively, to obtain several reference element information sets based on which the AI ​​module classifies the video tags of any video stream segments; wherein the video stream segments with the same video tags are placed in a same storage area in the multimedia video stream management cloud platform for data storage; wherein one reference element information set corresponds to one video tag;

[0052] Step S2: By comparing the deviation distribution of the reference element information presented between any two different video labels, any two video labels that have adjacent judgments when the AI ​​module performs differential classification are judged and identified, and several pairs of video label groups that are adjacent to each other are extracted;

[0053] Preferably: step S2 comprises:

[0054] Step S2-1: extracting overlapping feature element information from reference element information sets corresponding to any two different video tags one by one, and generating overlapping reference element information sets corresponding to any two different video tags respectively;

[0055] For example, the reference element information set corresponding to the first video tag includes {characteristic element information r1, characteristic element information r2, characteristic element information r4, characteristic element information r7}; the reference element information set corresponding to the second video tag includes {characteristic element information r1, characteristic element information r2, characteristic element information r3, characteristic element information r6};

[0056] In summary, the feature element information that overlaps between the first video tag and the second video tag includes feature element information r1 and feature element information r2; therefore, the reference element information set that overlaps between the first video tag and the second video tag is {feature element information r1, feature element information r2};

[0057] Step S2-2: Obtain the total value of the characteristic element information contained in each overlapping reference element information set corresponding to each type of video label and other types of video labels; if the total value of the characteristic element information contained in the overlapping reference element information set between any two different video labels is greater than the total threshold, it is determined that the AI ​​module has a neighboring judgment when distinguishing and classifying any two different video labels, and it is determined that the any two different video labels are a pair of video label groups that are adjacent labels to each other;

[0058] Step S3: extracting the distribution of the distinguishing reference element information presented in the process of the AI ​​module judging the video label of the corresponding video stream segment within the range of the corresponding mutually adjacent labels, and calculating the mutual proximity value for each pair of video label groups of mutually adjacent labels;

[0059] Wherein, step S3 comprises:

[0060] Step S3-1: If video tag A and video tag B are adjacent tags, extract the reference element information set P(A) of video tag A, the reference element information set P(B) of video tag B, and the overlapping reference element information set U between video tag A and video tag B. a,b ; Extract the video tag A to obtain the distinguishing reference element information set C1=P(A)-U a,b , extract the video tag B to obtain the distinguishing reference element information set C2=P(B)-U a,b ;

[0061] Step S3-2: Collect all video stream segments marked with video tag A from the multimedia video stream management cloud platform to obtain a first video stream segment set Y1, and collect all video stream segments marked with video tag B to obtain a second video stream segment set Y2; if a certain video stream segment exists in the first video stream segment set Y1 or the second video stream segment set Y2, and the feature element information set R extracted from the certain video stream segment satisfies R∩U a,b =Q≠ , where Q represents the feature element information set R and the overlapping reference element information set U a,b If the intersection between them is , a certain video stream segment is marked as a feature, and at the same time, a target distinguishing element information set Q'=RQ is extracted from a certain video stream segment;

[0062] Step S3-3: In the first video stream segment set Y1, the target distinguishing element information set extracted from all the feature-marked video stream segments is collected, and the number of types of feature element information appearing is accumulated. 1 In the second video stream segment set Y2, the target distinguishing element information set extracted from all the feature-marked video stream segments is collected, and the number of types of feature element information appearing is accumulated. 2 ; Calculate the first proximity index β between video label A and video label B 1 =[η 1 / card(C1)+η 2 / card(C2)] / 2;

[0063] Step S3-4: Obtain the number of video stream segments marked as features in the first video stream segment set Y1, and the number of video stream segments marked as features in the second video stream segment set Y2. The proportion value α1; calculate the second proximity index β between the video label A and the video label B 2 =(α1+α2) / 2;

[0064] The higher the ratio α1 or the ratio α2, the higher the ratio of the video stream segments containing the overlapping reference element information between the video label A and the video label B in the video stream segments determined to correspond to the video label A or the video label B, and the more frequently the AI ​​module makes a distinction between the video label A and the video label B; the more likely it is that the AI ​​module will make an error when distinguishing and classifying between the video label A and the video label B;

[0065] Step S3-5: Calculate the similarity of the feature element information set corresponding to each video stream segment in the first video stream segment set Y1 with the feature element information set corresponding to each video stream segment in the second video stream segment set Y2, and capture the highest similarity value δ; calculate the proximity value ξ=(1 / β 1 +β 2 )×δ.

[0066] For example, the first video stream segment set Y1 includes video stream segment w1, video stream segment w2, and video stream segment w3;

[0067] The second video stream segment set Y2 includes the video stream segment d1 and the video stream segment d2;

[0068] The similarity between the feature element information set corresponding to the video stream segment w1 and the feature element information set corresponding to the video stream segment d1 is Q1;

[0069] The similarity between the feature element information set corresponding to the video stream segment w1 and the feature element information set corresponding to the video stream segment d2 is Q2;

[0070] The similarity between the feature element information set corresponding to the video stream segment w2 and the feature element information set corresponding to the video stream segment d1 is Q3;

[0071] The similarity between the feature element information set corresponding to the video stream segment w2 and the feature element information set corresponding to the video stream segment d2 is Q4;

[0072] The similarity between the feature element information set corresponding to the video stream segment w3 and the feature element information set corresponding to the video stream segment d1 is Q5;

[0073] The similarity between the feature element information set corresponding to the video stream segment w3 and the feature element information set corresponding to the video stream segment d2 is Q6;

[0074] If Q6>Q3>Q2>Q4>Q61>Q5 is satisfied, then the highest similarity value captured is δ=Q6;

[0075] Step S4: In the multimedia video stream management cloud platform, the call rate distribution of the video stream segments stored in each storage area by the user is monitored, and the video tags of the video stream segments with abnormal call rates are adjusted by referring to the corresponding adjacent tags. According to the corresponding call rate changes brought about by the video tag adjustment, it is determined whether to send an early warning prompt that the AI ​​module needs to be optimized;

[0076] Wherein, step S4 comprises:

[0077] Step S4-1: In each storage area, the average call rate of all stored video stream segments by users is monitored; if in a storage area corresponding to a video tag a, there is a video stream segment whose corresponding call rate is lower than the average call rate, the video tag b with the highest proximity value to the video tag a is captured, and the video tag of the video stream segment is adjusted from a to b;

[0078] Step S4-2: monitor the call rate of a certain video stream segment presented by the user within a unit period of time after the video tag of a certain video stream segment is adjusted from a to b. If the call rate shows a numerical increase compared with that before the adjustment, generate an early warning signal for the optimization of the AI ​​module. If the call rate does not show a numerical increase compared with that before the adjustment, cancel the adjustment.

[0079] Step S4-3: When the cumulative number of warning signals generated for the AI ​​module optimization is greater than the number threshold, a warning prompt that the AI ​​module needs to be optimized is sent to the management terminal.

[0080] In order to better implement the above method, a multimedia video stream management system is also proposed, which includes: a video stream label processing data management module, a neighboring label judgment management module, a neighboring degree value calculation management module, and an AI module optimization prompt management module;

[0081] The video stream label processing data management module is used to collect the process data generated when the multimedia video stream management cloud platform generates corresponding video labels for each video stream segment received from the multimedia port by calling the AI ​​module, and to sort out the reference element information based on which the AI ​​module classifies the video labels of any video stream segment;

[0082] The adjacent label judgment management module is used to judge and identify any two video labels that have adjacent judgments when the AI ​​module performs differential classification by comparing the deviation distribution of the reference element information presented between any two different video labels, and extract several pairs of video label groups that are adjacent to each other;

[0083] Among them, the neighboring tag judgment management module includes: a feature element information combing unit and a neighboring tag judgment unit;

[0084] A characteristic element information combing unit, used to compare the deviation distribution of reference element information presented between any two different video tags;

[0085] The adjacent label judgment unit is used to judge and identify any two video labels that are adjacent to each other when the AI ​​module performs distinction classification, and extract several pairs of video label groups that are adjacent to each other;

[0086] A proximity value calculation management module is used to extract the distribution of the distinguishing reference element information presented in the process of the AI ​​module judging the video label of the corresponding video stream segment within the range of the corresponding mutually adjacent labels, and calculate the mutual proximity value between each pair of video label groups of mutually adjacent labels;

[0087] The AI ​​module optimization prompt management module is used to monitor the distribution of user call rates of video stream segments stored in each storage area in the multimedia video stream management cloud platform, and adjust the video tags of video stream segments with abnormal call rates by referring to the corresponding adjacent tags. According to the corresponding call rate changes brought about by the video tag adjustment, it is determined whether to send an early warning prompt that the AI ​​module needs to be optimized;

[0088] Among them, the AI ​​module optimization prompt management module includes a label adjustment monitoring management unit and an optimization early warning prompt management unit;

[0089] The label adjustment monitoring management unit is used to monitor the distribution of the calling rate of the video stream segments stored in each storage area by users in the multimedia video stream management cloud platform, and adjust the video labels of the video stream segments with abnormal calling rate by referring to the corresponding adjacent labels;

[0090] The optimization early warning prompt management unit is used to determine whether to send an early warning prompt that the AI ​​module needs to be optimized based on the corresponding call rate changes brought about by the adjustment of video tags.

[0091] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multimedia video stream management method based on artificial intelligence, characterized in that: The method comprises: Step S1: The multimedia video stream management cloud platform calls the AI ​​module to collect process data generated when each video stream segment received from the multimedia port generates a corresponding video tag, and sorts out the reference element information based on which the AI ​​module classifies the video tag types of any video stream segment; Step S2: By comparing the deviation distribution of the reference element information presented between any two different video labels, any two video labels that have adjacent judgments when the AI ​​module performs differential classification are judged and identified, and several pairs of video label groups that are adjacent to each other are extracted; Step S3: extracting the distribution of the distinguishing reference element information presented in the process of the AI ​​module judging the video label of the corresponding video stream segment within the range of the corresponding mutually adjacent labels, and calculating the mutual proximity value for each pair of video label groups of mutually adjacent labels; Step S4: In the multimedia video stream management cloud platform, the call rate distribution of the video stream segments stored in each storage area by the user is monitored, and the video tags of the video stream segments with abnormal call rates are adjusted by referring to the corresponding adjacent tags. According to the corresponding call rate changes brought about by the video tag adjustment, it is determined whether to send an early warning prompt that the AI ​​module needs to be optimized; The step S4 comprises: Step S4-1: In each storage area, the average call rate of all stored video stream segments by users is monitored; if in a storage area corresponding to a video tag a, there is a video stream segment whose corresponding call rate is lower than the average call rate, the video tag b with the highest proximity value to the video tag a is captured, and the video tag of the video stream segment is adjusted from a to b; Step S4-2: monitoring the call rate of the video stream segment presented by the user within a unit period of time after the video tag of the video stream segment is adjusted from a to b, and if the call rate shows a numerical increase compared with before the adjustment, generating a warning signal for AI module optimization; if the call rate does not show a numerical increase compared with before the adjustment, canceling the adjustment; Step S4-3: When the cumulative number of warning signals generated for the AI ​​module optimization is greater than the number threshold, a warning prompt that the AI ​​module needs to be optimized is sent to the management terminal.

2. The method for managing multimedia video streams based on artificial intelligence according to claim 1, characterized in that: The step S1 comprises: Step S1-1: The multimedia video stream management cloud platform collects all feature element information extracted after image recognition and audio analysis for each video stream segment before calling the AI ​​module to generate a corresponding video tag for each video stream segment received from the multimedia port, and collects and generates a feature element information set corresponding to each video stream segment; Step S1-2: Collect the characteristic element information sets of all video stream segments with the same corresponding video tags stored in the multimedia video stream management cloud platform, and obtain several reference element information sets based on which the AI ​​module classifies the video tag types of any video stream segments; wherein the video stream segments with the same video tags are placed in a same storage area in the multimedia video stream management cloud platform for data storage; wherein one reference element information set corresponds to one video tag.

3. The method for managing multimedia video streams based on artificial intelligence according to claim 2, characterized in that: The step S2 comprises: Step S2-1: extracting overlapping feature element information from reference element information sets corresponding to any two different video tags one by one, and generating overlapping reference element information sets corresponding to any two different video tags respectively; Step S2-2: Obtain the total value of the characteristic element information contained in each overlapping reference element information set corresponding to each type of video label and other types of video labels; if the total value of the characteristic element information contained in the overlapping reference element information set between any two different video labels is greater than the total threshold, it is determined that the AI ​​module has a proximity judgment when distinguishing and classifying the any two different video labels, and it is determined that the any two different video labels are a pair of video label groups that are adjacent to each other.

4. The method for managing multimedia video streams based on artificial intelligence according to claim 3, characterized in that: The step S3 comprises: Step S3-1: If video tag A and video tag B are adjacent tags, extract the reference element information set P(A) of video tag A, the reference element information set P(B) of video tag B, and the overlapping reference element information set U between video tag A and video tag B. a,b ; Extract the video tag A to obtain the distinguishing reference element information set C1=P(A)-U a,b , extract the video tag B to obtain the distinguishing reference element information set C2=P(B)-U a,b ; Step S3-2: Collect all video stream segments marked with video tag A from the multimedia video stream management cloud platform to obtain a first video stream segment set Y1, and collect all video stream segments marked with video tag B to obtain a second video stream segment set Y2; if a certain video stream segment exists in the first video stream segment set Y1 or the second video stream segment set Y2, and the feature element information set R extracted from the certain video stream segment satisfies R∩U a,b =Q≠ , where Q represents the feature element information set R and the overlapping reference element information set U a,b If the intersection between the two is a certain video stream segment, the certain video stream segment is marked as a feature, and at the same time, the certain video stream segment is extracted to obtain a target distinguishing element information set Q'=RQ; Step S3-3: In the first video stream segment set Y1, the target distinguishing element information set extracted from all the feature-marked video stream segments is collected, and the number of types of feature element information appearing is accumulated. 1 In the second video stream segment set Y2, the target distinguishing element information set extracted from all the feature-marked video stream segments is collected, and the number of types of feature element information appearing is accumulated. 2 ; Calculate the first proximity index β between video label A and video label B 1 =[η 1 / card(C1)+η 2 / card(C2)] / 2; Step S3-4: Obtain the number of video stream segments marked as features in the first video stream segment set Y1, and the number of video stream segments marked as features in the second video stream segment set Y2. The proportion value α1; calculate the second proximity index β between the video label A and the video label B 2 =(α1+α2) / 2; Step S3-5: Calculate the similarity of the feature element information set corresponding to each video stream segment in the first video stream segment set Y1 with the feature element information set corresponding to each video stream segment in the second video stream segment set Y2, and capture the highest similarity value δ; calculate the proximity value ξ=(1 / β 1 +β 2 )×δ.

5. A multimedia video stream management system, used to execute the multimedia video stream management method based on artificial intelligence as described in any one of claims 1 to 4, characterized in that: The system includes: a video stream label processing data management module, a neighboring label judgment management module, a neighboring degree value calculation management module, and an AI module optimization prompt management module; The video stream label processing data management module is used to collect process data generated when the multimedia video stream management cloud platform generates corresponding video labels for each video stream segment received from the multimedia port by calling the AI ​​module, and to sort out the reference element information based on which the AI ​​module classifies the video labels of any video stream segment; The adjacent label judgment management module is used to judge and identify any two video labels that have adjacent judgments when the AI ​​module performs differential classification by comparing the deviation distribution of reference element information presented between any two different video labels, and extract several pairs of video label groups that are adjacent to each other; The proximity value calculation management module is used to extract the distribution of the distinguishing reference element information presented in the process of the AI ​​module judging the video label of the corresponding video stream segment within the range of the corresponding mutually adjacent labels, and calculate the mutual proximity value between each pair of video label groups of mutually adjacent labels; The AI ​​module optimization prompt management module is used to monitor the distribution of user call rates of video stream segments stored in each storage area in the multimedia video stream management cloud platform, adjust the video tags of video stream segments with abnormal call rates with reference to corresponding adjacent tags, and determine whether to send an early warning prompt that the AI ​​module needs to be optimized based on the corresponding call rate changes brought about by the video tag adjustment.

6. A multimedia video stream management system according to claim 5, characterized in that: The adjacent tag judgment management module includes: a feature element information combing unit and an adjacent tag judgment unit; The characteristic element information combing unit is used to compare the deviation distribution of the reference element information presented between any two different video tags; The adjacent label judgment unit is used to judge and identify any two video labels that have adjacent judgments when the AI ​​module performs differential classification, and extract several pairs of video label groups that are adjacent labels to each other.

7. A multimedia video stream management system according to claim 5, characterized in that: The AI ​​module optimization prompt management module includes a label adjustment monitoring management unit and an optimization early warning prompt management unit; The label adjustment monitoring management unit is used to monitor the distribution of the calling rate of the video stream segments stored in each storage area by users in the multimedia video stream management cloud platform, and to adjust the video labels of the video stream segments with abnormal calling rates by referring to the corresponding adjacent labels; The optimization warning prompt management unit is used to determine whether to send a warning prompt that the AI ​​module needs to be optimized based on the corresponding call rate change caused by the adjustment of the video tag.

Images