Live broadcast stream hotlinking real-time tracing method and system based on multi-source heterogeneous signaling fusion

Abstract

The invention discloses a live streaming hotlinking real-time traceability method and system based on multi-source heterogeneous signaling fusion. The method comprises the steps of multi-source heterogeneous signaling collection, multi-source heterogeneous signaling feature extraction, hotlinking adaptation loss function construction, hotlinking traceability model design, parameter dynamic optimization and live streaming hotlinking real-time traceability judgment. The invention belongs to the field of data processing, and particularly relates to a live stream hotlinking real-time traceability method and system based on multi-source heterogeneous signaling fusion, and the method comprises the steps: constructing a hotlinking adaptive loss function based on the arc tangent equation power fallback characteristic, and adapting to multi-signaling modal noise fluctuation; a risk probability site inverted penalty term is embedded, so that the live streaming hotlinking real-time traceability reliability is effectively improved; and constructing a whole-process prevention and control system: designing a parameter dynamic optimization strategy driven by a reverse order rate, dynamically adapting data fluctuation, establishing a structure-gradient two-dimensional reverse order control mechanism, and avoiding reverse order risks from model structure design to parameter gradient optimization in a whole link, so that the hotlinking traceability effect is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a live streaming pirate link real-time tracing method and system based on multi-source heterogeneous signaling fusion. BACKGROUND

[0002] The live streaming pirate link real-time tracing method refers to a technical means for collecting and analyzing multi-source data related to live streaming, identifying abnormal features and pirate link behavior patterns, and locating the source of pirate links and determining the risk of pirate links. However, the general live streaming pirate link real-time tracing method is difficult to cope with the noise fluctuation of multi-signaling data of live streaming, and high-value signaling information is easily covered by noise, resulting in poor reliability of pirate link tracing results. The general live streaming pirate link real-time tracing method has weak adaptability to live streaming data fluctuations and is prone to logical conflicts, which exacerbates the risk of reverse order of probability points, resulting in poor tracing effect. SUMMARY

[0003] In view of the above problems, in order to overcome the defects of the prior art, the present application provides a live streaming pirate link real-time tracing method and system based on multi-source heterogeneous signaling fusion. In view of the problem that the general live streaming pirate link real-time tracing method is difficult to cope with the noise fluctuation of multi-signaling data of live streaming, and high-value signaling information is easily covered by noise, resulting in poor reliability of pirate link tracing results, the present application constructs a pirate link adaptation loss function based on the arc-cut algorithm power type rollback characteristics, adapts to the noise fluctuation of multi-signaling modal to ensure the optimization stability; calculates the signaling adaptation coefficient according to the signal-to-noise ratio of each modal, distinguishes the information value of the modal and balances the error contribution; embeds a reverse order penalty term of risk probability points, constrains the monotonicity of risk level, and effectively improves the real-time tracing reliability of live streaming pirate links. In view of the problem that the general live streaming pirate link real-time tracing method has weak adaptability to live streaming data fluctuations and is prone to logical conflicts, which exacerbates the risk of reverse order of probability points, resulting in poor tracing effect, the present application constructs a whole-process prevention and control system: designs a reverse order rate driven parameter dynamic optimization strategy, dynamically adapts to data fluctuations, and considers the convergence efficiency and training stability; establishes a structure-gradient dual-dimensional reverse order control mechanism to avoid reverse order risk from model structure design to parameter gradient optimization, and significantly improves the pirate link tracing effect.

[0004] The technical scheme adopted by the present application is as follows: the present application provides a live streaming pirate link real-time tracing method based on multi-source heterogeneous signaling fusion, which comprises the following steps:

[0005] Step S1: multi-source heterogeneous signaling acquisition;

[0006] Step S2: multi-source heterogeneous signaling feature extraction;

[0007] Step S3: pirate link adaptation loss function construction;

[0008] Step S4: pirate link tracing model design;

[0009] Step S5: parameter dynamic optimization;

[0010] Step S6: live stream pirate link real-time tracing determination.

[0011] Further, in step S1, the multi-source heterogeneous signaling collection is to obtain historical multi-source heterogeneous signaling data, including stream signaling modal data, device fingerprint signaling modal data and network behavior signaling modal data; multi-source data space-time logical alignment and missing value processing are performed, and labels are labeled; and a multi-source heterogeneous signaling set is obtained.

[0012] Further, in step S2, the multi-source heterogeneous signaling feature extraction specifically includes:

[0013] Single-source signaling feature extraction: for network behavior signaling modal data, a convolutional neural network is used to extract deep visual features to capture abnormal patterns of network transmission; for stream signaling modal, a Transformer encoder is used to extract semantic features to identify abnormal signaling fields; for device fingerprint signaling modal, dimension differences are eliminated through standardization, and then quantitative features are extracted through a fully connected layer to capture abnormal device fingerprints;

[0014] Multi-source signaling feature weighted fusion: heterogeneous features are mapped to a unified feature space to realize multi-source signaling feature weighted fusion.

[0015] Further, in step S3, the pirate link adaptive loss function construction specifically includes:

[0016] Define the pirate link risk prediction error: the error between the pirate link risk value predicted by the multi-source signaling fusion feature calculation model and the true label is calculated;

[0017] Pirate link adaptive loss function definition: using the power-type decreasing property of arc tangent formula, combined with the signaling adaptation coefficient, a loss function is constructed to construct the pirate link adaptive loss function;

[0018] Risk level reverse order penalty term design: the reverse order penalty term is embedded into the total loss function to measure the degree of monotonicity violation;

[0019] Total loss function construction: the pirate link adaptive loss and the reverse order penalty term are fused to construct the total loss function.

[0020] Further, in step S4, the pirate link tracing model design specifically includes:

[0021] Feature enhancement unit design: a self-attention mechanism is introduced to perform feature enhancement on the fused features;

[0022] Multi-probability site parallel prediction unit design: based on the enhanced features, a deep fully connected network is constructed to realize parallel output of multiple confidence pirate link risk values.

[0023] Further, in step S5, the parameter dynamic optimization is to calculate the risk probability site reverse order rate in real time during the training process, and to adjust the single-step parameter update amplitude.

[0024] Further, in step S6, the live streaming link stealing real-time tracing judgment specifically includes:

[0025] The two-dimensional reverse active control includes: structure layer constraint, all risk probability sites share the same set of feature enhancement and full connection network parameters, and different sites are predicted only by the difference of output layer weight; gradient layer constraint, in the reverse propagation stage, the parameter update amplitude of different probability sites is balanced based on the first-order and second-order derivatives;

[0026] Regarding model establishment: the obtained multi-source heterogeneous signaling set is divided into a training set, a validation set and a test set; the model is trained based on the training set, and the training is stopped when any condition is met: the training round number reaches the upper limit, the validation set loss is continuously stable or the reverse order rate is lower than the threshold; the test set data is used to verify the model performance;

[0027] Real-time tracing judgment, based on the established link stealing tracing model, the live streaming multi-source signaling data is traced and judged.

[0028] The live streaming link stealing real-time tracing system based on multi-source heterogeneous signaling fusion provided by the application includes a multi-source heterogeneous signaling acquisition module, a multi-source heterogeneous signaling feature extraction module, a link stealing adaptive loss function construction module, a link stealing tracing model design module, a parameter dynamic optimization module and a live streaming link stealing real-time tracing judgment module.

[0029] The multi-source heterogeneous signaling acquisition module acquires historical multi-signaling modal data and pre-processes to obtain a multi-source heterogeneous signaling set;

[0030] The multi-source heterogeneous signaling feature extraction module extracts features from the multi-source heterogeneous signaling set and performs weighted fusion;

[0031] The link stealing adaptive loss function construction module calculates the prediction error based on the fused features, defines the link stealing adaptive loss function, and superimposes the risk probability site reverse order penalty term to obtain the total loss function;

[0032] The link stealing tracing model design module designs a link stealing tracing model based on the multi-source heterogeneous signaling set and the total loss function, which is a self-attention feature enhancement-multi-probability site parallel prediction model;

[0033] The parameter dynamic optimization module dynamically adjusts the single-step parameter update amplitude based on the risk probability site reverse order rate to optimize the model parameters;

[0034] The live streaming link stealing real-time tracing judgment module controls the risk level reverse order, establishes the link stealing tracing model, and performs live streaming link stealing real-time tracing judgment.

[0035] The beneficial effects achieved by the application using the above scheme are as follows:

[0036] (1) In view of the problem that the general live stream pirate link real-time tracing method is difficult to cope with the noise fluctuation of live stream multi-signaling data, and high-value signaling information is easily covered by noise, resulting in poor reliability of pirate link tracing results, the scheme constructs a pirate link adaptation loss function based on the power-type back-off characteristics of arc cutting formula, adapts to the noise fluctuation of multi-signaling modal to ensure the optimization stability; calculates the signaling adaptation coefficient according to the signal-to-noise ratio of each modal, distinguishes the modal information value and balances the error contribution; embeds a risk probability site reverse order penalty term, which constrains the monotonicity of risk level and effectively improves the reliability of live stream pirate link real-time tracing.

[0037] (2) In view of the problem that the general live stream pirate link real-time tracing method has weak adaptability to live stream data fluctuation, is easy to appear logical conflict, aggravates the reverse order risk of probability site, and leads to poor tracing effect, the scheme constructs a whole-process prevention and control system: designs a reverse order rate driven parameter dynamic optimization strategy, dynamically adapts to data fluctuation, and considers the convergence efficiency and training stability; establishes a structure-gradient dual-dimensional reverse order control mechanism, which avoids reverse order risk from model structure design to parameter gradient optimization whole link, and significantly improves the pirate link tracing effect. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 A live stream pirate link real-time tracing method based on multi-source heterogeneous signaling fusion is provided for the application;

[0039] Figure 2 A live stream pirate link real-time tracing system based on multi-source heterogeneous signaling fusion is provided for the application.

[0040] The accompanying drawings are used to provide a further understanding of the application, and constitute a part of the specification, together with the embodiments of the application, to explain the application, and do not constitute a limitation on the application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments of the application; based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0042] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the systems or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0043] Embodiment one, refer to Figure 1 The present application provides a live stream piracy real-time tracing method based on multi-source heterogeneous signaling fusion, which comprises the following steps:

[0044] Step S1: Multi-source heterogeneous signaling acquisition, obtaining historical multi-signaling modal data and pre-processing to obtain a multi-source heterogeneous signaling set;

[0045] Step S2: Multi-source heterogeneous signaling feature extraction, feature extraction and weighted fusion are performed on the multi-source heterogeneous signaling set;

[0046] Step S3: Piracy adaptation loss function construction, based on the fusion features, the prediction error is calculated, the piracy adaptation loss function is defined, and the total loss function is obtained by superimposing the risk probability site inverse order penalty term;

[0047] Step S4: Piracy tracing model design, based on the multi-source heterogeneous signaling set and the total loss function, a piracy tracing model with self-attention feature enhancement and multi-probability site parallel prediction is designed;

[0048] Step S5: Parameter dynamic optimization, based on the risk probability site inverse order rate, the single step parameter update amplitude is dynamically adjusted to optimize the model parameters;

[0049] Step S6: Live stream piracy real-time tracing determination, constraint control risk level inverse order, establish the piracy tracing model, and make live stream piracy real-time tracing determination.

[0050] Embodiment two, refer to Figure 1, the embodiment based on the above embodiment, in step S1, multi-source heterogeneous signaling collection is to obtain historical multi-source heterogeneous signaling data, including stream signaling modal data, device fingerprint signaling modal data and network behavior signaling modal data; the stream signaling modal data includes live server push stream / pull stream request signaling, session ID, timestamp, stream address, and text type signaling data; the device fingerprint signaling modal data includes IP address, UA header, hardware model, and numerical value type characteristic data of terminal device; the network behavior signaling modal data includes network transmission packet size, transmission rate, request frequency, TCP connection duration, and time sequence image data, which converts time sequence features into two-dimensional heat maps; multi-source data space-time logical alignment and missing value processing are performed, and labels are labeled at the same time, including normal pull stream and pirate pull stream; the multi-source data space-time logical alignment is to realize one-to-one correspondence of stream signaling, device fingerprint and network behavior data according to the double keys of session ID+timestamp; the missing value processing is to use bilinear interpolation to complete the missing time sequence heat map pixels for image data, fill in empty strings for text data, delete invalid random code signaling, and use mean value to complete the missing device characteristic value for numerical value data; a multi-source heterogeneous signaling set is obtained.

[0051] Embodiment three, refer to Figure 1 , the embodiment based on the above embodiment, in step S2, multi-source heterogeneous signaling feature extraction is to design a special feature encoder according to the characteristics of the three types of heterogeneous signaling data in the multi-source heterogeneous signaling set, extract deep features, and map them to a unified feature space through weighted fusion, strengthen the key features related to pirate link behavior, and weaken the redundant interference features; specifically including:

[0052] Single-source signaling feature extraction, for network behavior signaling modal data, a convolutional neural network is used to extract deep visual features to capture abnormal patterns of network transmission, and the feature is represented as: ; for stream signaling modal, a Transformer encoder is used to extract semantic features to identify abnormal signaling fields, and the feature is represented as: ; for device fingerprint signaling modal, first, the dimension difference is eliminated through standardization, and then the quantized features are extracted through a fully connected layer to capture abnormal device fingerprints, and the feature is represented as: ;

[0053] wherein, , and are network behavior deep feature vector, stream signaling semantic feature vector and device fingerprint quantized feature vector respectively; is the feature extraction operation of the convolutional neural network; , and are time sequence heat map data, stream signaling text data and device fingerprint numerical value data respectively; is a semantic feature extraction operation of the Transformer encoder; is a nonlinear activation function using Sigmoid; is a normalization operation; is a full-connection layer bias vector of the numerical signaling modality;

[0054] Multi-source signaling feature weighted fusion, mapping heterogeneous features to a unified feature space, realizing multi-source signaling feature weighted fusion, is expressed as: ; wherein, is a multi-source signaling fusion feature vector; , and are fusion weight matrices of the three signaling modalities; b is a fusion bias vector.

[0055] Embodiment four, see Figure 1 , this embodiment is based on the above embodiment, in step S3, the construction of the pirate link adaptation loss function is to convert the live stream pirate link tracing into a pirate link risk probability quantile prediction task, and output the pirate link risk value under different confidence degrees; in order to avoid the loss of non-derivative and gradient mutation, the pirate link adaptation loss function is designed, the signaling adaptation coefficient is introduced to balance the error contribution of different signaling modalities, and the risk level inverse sequence penalty term is embedded to realize the integration of loss optimization and risk level rationality; specifically including:

[0056] Define the pirate link risk prediction error, which is the error between the pirate link risk value predicted by the multi-source signaling fusion feature calculation model and the real label, expressed as: ; wherein, y is the real label of the pirate link behavior, 1 represents pirate link, and 0 represents normal; is the pirate link risk prediction value of the model based on the fusion feature; u is the prediction error;

[0057] Pirate link adaptation loss function definition, using the power type decreasing attribute of arc tangent formula, combined with signaling adaptation coefficient to construct loss function, to construct a globally derivable pirate link adaptation loss function; the pirate link adaptation loss function is expressed as: ; wherein, is the pirate link adaptation loss function; is the target risk probability quantile; s is a softening parameter, taking a value of 0.05~1.0, the greater the noise of the pirate link behavior, the greater the value; is the signaling adaptation coefficient, calculated by the signal-to-noise ratio of each modality, expressed as , , and are the adaptation coefficients of the three signaling modalities;

[0058] The risk level reverse order penalty term is designed, the risk level of the link stealing needs to meet the monotonicity (high confidence risk value ≥ low confidence risk value), the reverse order penalty term is embedded into the total loss function, the degree of violation of the monotonicity is measured, and is expressed as: ; wherein, is the reverse order penalty value of the ith sample, if , the reverse order occurs, a positive penalty value is generated, and the greater the reverse order degree, the higher the penalty value; M is the number of risk probability points, and m is the index of the risk probability point; and are the predicted values of the mth risk probability point and the m+1th risk probability point of the ith sample respectively;

[0059] The total loss function is constructed by fusing the link stealing adaptive loss and the reverse order penalty term, and the total loss function is constructed , and is expressed as: ; wherein, N is the number of samples in the current training batch; is a regularization intensity coefficient, which is calculated based on the feature dimension proportion of the signaling, ; wherein, is a basic penalty intensity, and the value is 0.1-1.0; and are the feature dimension proportion and the feature vector dimension of the fth signaling respectively.

[0060] By performing the above operation, in order to solve the problem that the general live stream link stealing real-time tracing method is difficult to deal with the noise fluctuation of the live stream multi-signaling data, the high-value signaling information is easily covered by the noise, and the reliability of the link stealing tracing result is poor, the scheme constructs a link stealing adaptive loss function based on the arc cutting algorithm power type rollback characteristic, adapts to the noise fluctuation of the multi-signaling mode to ensure the optimization stability; according to the signal-to-noise ratio of each mode, a signaling adaptive coefficient is calculated, the value of the mode information is distinguished, and the error contribution is balanced; the reverse order penalty term of the risk probability point is embedded, the monotonicity of the risk level is constrained, and the reliability of the live stream link stealing real-time tracing is effectively improved.

[0061] Embodiment five, refer to Figure 1 , this embodiment is based on the above embodiment, in step S4, the link stealing tracing model design is based on multi-source heterogeneous signaling feature extraction and total loss function, a link stealing tracing model of self-attention feature enhancement-multi-probability point parallel prediction is constructed, the key information of the link stealing behavior in the fused feature is strengthened, and the parameter update amplitude is dynamically adjusted based on the reverse order rate of the link stealing risk, to ensure the stability and real-time performance of the model training, specifically including:

[0062] The feature enhancement unit is designed, the self-attention mechanism is introduced, the key information strongly related to the link stealing behavior in the fused feature is automatically selected and fused, and the redundant information is weakened, and the feature enhancement formula is expressed as: ; wherein, is the enhanced link stealing feature vector; is the self-attention feature enhancement operation; , and are the query, key, value matrices of the attention mechanism, respectively, obtained by linear transformation; is the dimension of the key matrix, used for scale normalization of the attention score;

[0063] The multi-probability site parallel prediction unit is designed, a deep fully connected network is constructed based on the enhanced features, and parallel output of the multi-confidence link stealing risk value is realized. The model prediction formula is represented as: ; wherein, and are the weight matrix and bias vector of the hidden layer, respectively; and are the weight matrix and bias vector of the output layer, and the dimension of the output layer is equal to the number of target risk probability sites; is the ReLU activation function.

[0064] Embodiment six, see Figure 1 , this embodiment is based on the above embodiment, in step S5, the parameter dynamic optimization is to avoid the reverse order of the risk probability site prediction result, by monitoring the reverse order rate of the risk probability site in real time, the single step parameter update amplitude of the model is dynamically adjusted; The specific operation is:

[0065] The reverse order rate CR of the risk probability site is calculated in real time during the training process, and the rationality of the risk level prediction is quantified, which is represented as: ; When CR>0.1, it means that the reverse order of the risk probability site is serious, and the single step parameter update amplitude is dynamically reduced (reduced by 0.1 times each time), to suppress the prediction confusion caused by the too large parameter update amplitude; Otherwise, it means that the prediction result is reasonable, and the single step parameter update amplitude is appropriately increased (increased by 0.02 times each time), to improve the model convergence speed; is the indicator function, the value in the parentheses is equal to 1 when the condition is true, otherwise it is equal to 0; The parameters involved include model structure hyperparameters, training hyperparameters and loss function hyperparameters; The optimizer selects Adam optimizer; The parameter update process is: Adam optimizer iterative update- calculate the reverse rate- dynamically adjust the single step update amplitude- adjust the update amplitude of each site according to the gradient layer constraint during back propagation- enter the next round of iteration.

[0066] Embodiment seven, see Figure 1 , this embodiment is based on the above embodiment, in step S6, the live stream link stealing real-time tracing judgment is to actively control the risk level reverse order from the model structure and gradient optimization, and the model trained is combined to realize the real-time tracing of the live stream link stealing, and the link stealing risk probability interval and the tracing result are output; The specific operation is:

[0067] The two-dimensional reverse order active control includes: structure layer constraint, all risk probability sites share the same set of feature enhancement and fully connected network parameters, and different sites are predicted only by the difference of output layer weight, avoiding the reverse order caused by different feature screening logic; gradient layer constraint, in the backward propagation stage, the parameter update amplitude of different probability sites is balanced based on the first and second derivatives, and the parameter update amplitude is represented as: ; wherein, is the prediction value update amplitude of the mth risk probability site; and are the first and second derivatives of the ith sample at the mth risk probability site, respectively; avoiding the case that the update amplitude of the low risk probability site is too large to exceed the high risk probability site; is the learning rate, and the value is 1e -5 ~0.1; is the smoothing coefficient, and the value is 0.01~0.1;

[0068] Regarding model establishment: the obtained multi-source heterogeneous signaling set is divided into training set, validation set and test set; the model is trained based on the training set, and the training is stopped when any condition is met: the training round reaches the upper limit (the value is 50~200), the validation set loss is continuously stable or the reverse order rate is lower than the threshold value (the value is 0.03~0.05); the test set data is used to verify the performance of the model, and the traceability accuracy corresponding to the quantile loss is required to be not less than 0.97;

[0069] Real-time traceability judgment: based on the established pirate link traceability model, the live stream multi-source signaling data is judged for traceability; the process is: inputting new live stream multi-source signaling data, sequentially passing through single signaling modal feature extraction, multi-source feature weighted fusion, self-attention feature enhancement, and multi-probability site parallel prediction, outputting M pirate link risk values of probability sites, forming a pirate link risk probability interval; when the risk value of the high confidence site (τ=0.9) is ≥0.8, it is judged as a pirate link behavior, and the corresponding traceability information including device fingerprint and stream address is output.

[0070] By performing the above operation, in view of the problems that the general live stream pirate link real-time traceability method has weak data fluctuation adaptability, is prone to logical conflict, aggravates the reverse order risk of probability sites, and leads to poor traceability effect, the present scheme constructs a whole-process prevention and control system: a reverse order rate driven parameter dynamic optimization strategy is designed, which dynamically adapts to data fluctuation, and takes into account the convergence efficiency and training stability; a structure-gradient two-dimensional reverse order control mechanism is established, which avoids the reverse order risk from the whole link of model structure design to parameter gradient optimization, and significantly improves the pirate link traceability effect.

[0071] Embodiment eight, refer to Figure 2The embodiment is based on the above-mentioned embodiment, and the live stream pirate link real-time tracing system based on multi-source heterogeneous signaling fusion provided by the application comprises a multi-source heterogeneous signaling acquisition module, a multi-source heterogeneous signaling feature extraction module, a pirate link adaptive loss function construction module, a pirate link tracing model design module, a parameter dynamic optimization module and a live stream pirate link real-time tracing judgment module.

[0072] The multi-source heterogeneous signaling acquisition module acquires historical multi-signaling modal data and performs preprocessing to obtain a multi-source heterogeneous signaling set.

[0073] The multi-source heterogeneous signaling feature extraction module performs feature extraction on the multi-source heterogeneous signaling set and performs weighted fusion.

[0074] The pirate link adaptive loss function construction module calculates a prediction error based on the fused features, defines a pirate link adaptive loss function, superimposes a risk probability site reverse order penalty term to obtain a total loss function.

[0075] The pirate link tracing model design module designs a pirate link tracing model based on the multi-source heterogeneous signaling set and the total loss function, and the pirate link tracing model is a self-attention feature enhancement-multi-probability site parallel prediction pirate link tracing model.

[0076] The parameter dynamic optimization module dynamically adjusts the single-step parameter update amplitude based on the risk probability site reverse order rate to optimize the model parameters.

[0077] The live stream pirate link real-time tracing judgment module constrains and controls the risk level reverse order, establishes the pirate link tracing model, and performs live stream pirate link real-time tracing judgment.

[0078] It should be noted that in this document, the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles or devices that include a series of elements not only include those elements, but also include other elements not explicitly listed, or inherent to such processes, methods, articles or devices.

[0079] Although the embodiments of the application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the application.

[0080] The above describes the application and its embodiments, which are not restrictive, and the drawings only show one of the embodiments of the application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the application, without creative design, similar structure and embodiments of the technical solution can be designed, which should belong to the protection scope of the application.

Claims

1. A live streaming piracy real-time tracing method based on multi-source heterogeneous signaling fusion, characterized in that: The method comprises the following steps: Step S1: multi-source heterogeneous signaling collection, obtaining historical multi-signaling modal data, and preprocessing to obtain a multi-source heterogeneous signaling set; Step S2: multi-source heterogeneous signaling feature extraction, performing feature extraction on the multi-source heterogeneous signaling set and performing weighted fusion; Step S3: construction of a pirate link adaptation loss function, calculating a prediction error based on the fused features, defining a pirate link adaptation loss function, superimposing a risk probability site reverse order penalty term to obtain a total loss function; Step S4: design of a pirate link tracing model, based on the multi-source heterogeneous signaling set and the total loss function, designing a pirate link tracing model with self-attention feature enhancement and multi-probability site parallel prediction; Step S5: dynamic optimization of parameters, dynamically adjusting the single-step parameter update amplitude based on the risk probability site reverse order rate to optimize the model parameters; Step S6: real-time tracing and judgment of live streaming pirate links, constraining and controlling the reverse order of the risk level, establishing a pirate link tracing model, and performing real-time tracing and judgment of live streaming pirate links.

2. The live streaming piracy tracing method based on multi-source heterogeneous signaling fusion according to claim 1, characterized in that: In step S3, the pirate link adaptation loss function construction specifically comprises: Defining a pirate link risk prediction error, calculating the error between the pirate link risk value predicted by the multi-source signaling fusion feature calculation model and the true label; Pirate link adaptation loss function definition, using the power type decreasing attribute of the arc tangent formula to construct the loss function combined with the signaling adaptation coefficient to construct the pirate link adaptation loss function; Risk level reverse order penalty term design, embedding the reverse order penalty term into the total loss function to measure the degree of monotonicity violation; Total loss function construction, fusing the pirate link adaptation loss and the reverse order penalty term to construct the total loss function.

3. The live streaming piracy tracing method based on multi-source heterogeneous signaling fusion according to claim 2, characterized in that: In step S4, the pirate link tracing model design specifically comprises: Feature enhancement unit design, introducing a self-attention mechanism to perform feature enhancement on the fused features; Multi-probability site parallel prediction unit design, constructing a deep fully connected network based on the enhanced features to realize parallel output of multiple confidence pirate link risk values.

4. The live streaming piracy tracing method based on multi-source heterogeneous signaling fusion according to claim 3, characterized in that: In step S5, the parameter dynamic optimization is to calculate the risk probability site reverse order rate in real time during the training process and adjust the single-step parameter update amplitude.

5. The live streaming piracy tracing method based on multi-source heterogeneous signaling fusion according to claim 4, characterized in that: In step S6, the real-time tracing and judgment of live streaming pirate links specifically comprises: Double-dimensional reverse order active control, including: structure layer constraint, all risk probability sites share the same set of feature enhancement and fully connected network parameters, and only the output layer weight difference is used to realize prediction of different sites; gradient layer constraint, based on the first and second derivatives, the parameter update amplitude of different probability sites is balanced during the back propagation phase; Regarding model establishment: divide the obtained multi-source heterogeneous signaling set into a training set, a validation set, and a test set; train the model based on the training set, and stop training when any of the following conditions is met: the training round number reaches the upper limit, the validation set loss is continuously stable, or the reverse order rate is lower than a threshold; use the test set data to verify the model performance; Real-time tracing and judgment, based on the established pirate link tracing model, the live streaming multi-source signaling data is traced and judged.

6. The live streaming piracy tracing method based on multi-source heterogeneous signaling fusion according to claim 5, characterized in that: In step S1, the multi-source heterogeneous signaling collection is to obtain historical multi-source heterogeneous signaling data, including stream signaling modal data, device fingerprint signaling modal data, and network behavior signaling modal data; perform multi-source data space-time logical alignment and missing value processing, and simultaneously perform label labeling; Obtain a multi-source heterogeneous signaling set.

7. The live streaming piracy tracing method based on multi-source heterogeneous signaling fusion according to claim 6, characterized in that: In step S2, the multi-source heterogeneous signaling feature extraction specifically includes: Single-source signaling feature extraction: for network behavior signaling modality data, a convolutional neural network is used to extract deep visual features to capture abnormal patterns of network transmission; for flow signaling modality, a Transformer encoder is used to extract semantic features to identify abnormal signaling fields; for device fingerprint signaling modality, dimension differences are eliminated through standardization, and then quantitative features are extracted through a fully connected layer to capture abnormal device fingerprints; Multi-source signaling feature weighted fusion: mapping heterogeneous features to a unified feature space to realize multi-source signaling feature weighted fusion.

8. A live streaming piracy real-time tracing system based on multi-source heterogeneous signaling fusion, configured to implement the live streaming piracy real-time tracing method based on multi-source heterogeneous signaling fusion according to any one of claims 1-7, characterized in that: It comprises a multi-source heterogeneous signaling collection module, a multi-source heterogeneous signaling feature extraction module, a link theft adaptive loss function construction module, a link theft tracing model design module, a parameter dynamic optimization module, and a live streaming link theft real-time tracing judgment module. The multi-source heterogeneous signaling collection module acquires historical multi-signaling modality data and pre-processes them to obtain a multi-source heterogeneous signaling set. The multi-source heterogeneous signaling feature extraction module extracts features from the multi-source heterogeneous signaling set and performs weighted fusion. The link theft adaptive loss function construction module calculates the prediction error based on the fused features, defines the link theft adaptive loss function, and adds a risk probability site reverse order penalty term to obtain a total loss function. The link theft tracing model design module designs a link theft tracing model based on the multi-source heterogeneous signaling set and the total loss function, which is a self-attention feature enhancement-multi-probability site parallel prediction model. The parameter dynamic optimization module dynamically adjusts the single-step parameter update amplitude based on the risk probability site reverse order rate to optimize the model parameters. The live streaming link theft real-time tracing judgment module constrains and controls the risk level reverse order, establishes the link theft tracing model, and performs live streaming link theft real-time tracing judgment.

Images