Private domain live broadcast peak hot spot prediction and content scheduling method based on deep learning

Abstract

The invention provides a private domain live broadcast peak hot spot prediction and content scheduling method based on deep learning, and relates to the technical field of private domain live broadcast, and the method comprises the steps: extracting multi-scale time sequence features in private domain live broadcast historical data through employing a hybrid memory enhanced neural network, a live broadcast peak prediction result is generated through a spatial pyramid attention module and a progressive feature distillation network, a dynamic heterogeneous graph neural network is constructed, multi-modal data features and hierarchical comparative learning are combined, a live broadcast content scheduling strategy is output through a multi-agent reinforcement learning framework, a bidirectional interactive evaluation network is constructed, and a live broadcast peak prediction result is obtained. And performing online optimization on the scheduling strategy according to the real-time live broadcast data by using a fast adaptation algorithm of meta reinforcement learning to obtain an optimal content scheduling strategy.

Description

Technical Field

[0001] The present invention relates to the field of private domain live broadcast technology, and in particular to a method for predicting hot spots and scheduling content during peak periods of private domain live broadcast based on deep learning. Background Art

[0002] The rapid development of deep learning technology has brought new opportunities to various industries. In the field of live streaming, how to accurately predict the peak period of live streaming and effectively schedule content is crucial to improving user experience and enhancing live streaming results. Traditional methods for predicting peak live broadcast periods rely primarily on statistical analysis and rule matching, which struggles to capture complex and changing user behavior patterns and results in limited prediction accuracy. Furthermore, traditional content scheduling methods lack intelligence and personalization, making it impossible to dynamically adjust live broadcast content based on real-time conditions and unable to meet diverse user needs. Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the Invention

[0003] The embodiments of the present invention provide a method for predicting hot spots and scheduling content during peak periods of private domain live broadcasts based on deep learning, which can at least solve some of the problems existing in the prior art.

[0004] A first aspect of an embodiment of the present invention provides a method for predicting hot spots and scheduling content during peak periods of private domain live broadcasts based on deep learning, comprising: Collect historical private domain live broadcast data and construct a hybrid memory-enhanced neural network containing a fast memory pathway and a slow memory pathway. The fast memory pathway models short-term dependencies through a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies through a hierarchical sparse attention structure. The private domain live broadcast historical data is input into the hybrid memory-enhanced neural network, and multi-scale temporal features are extracted through the fast and slow memory pathways. A spatial pyramid attention module is constructed, and the multi-scale temporal features are reorganized based on the parallel attention branches in the spatial pyramid attention module. The feature receptive field is adjusted using deformable convolution to obtain a hierarchical feature representation. The hierarchical feature representation is input into a progressive feature distillation network, which performs multi-level knowledge transfer and feature reorganization operations to generate a live broadcast peak period prediction result. Collect live broadcast-related multimodal data and perform feature extraction and feature encoding to construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network. Combined with the topology structure generator, the node association relationship is constructed. The cross-modal feature fusion is performed through the feature interaction engine. The hierarchical contrast learning training method is applied to the dynamic heterogeneous graph neural network. Sample pairs are constructed at different semantic levels and feature comparison is performed. The feature comparison results are input into the hierarchical strategy optimization network. The global goal is divided into multiple sub-goals through the goal decomposition module. The sub-goal realization path is planned according to the execution planning module. The hierarchical strategy optimization network is input into the multi-agent reinforcement learning framework of non-cooperative game. The strategy space is optimized through multi-agent competition and collaboration, and the live broadcast content scheduling strategy is output. Real-time live broadcast data is collected and added to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features according to a cross-modal fusion method. The effect evaluator establishes a mapping relationship between predicted effects and actual effects based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Collection and adjustment are repeated to obtain the optimal content scheduling strategy.

[0005] In an optional embodiment, Collecting historical private domain live broadcast data, constructing a hybrid memory-enhanced neural network including a fast memory pathway and a slow memory pathway, wherein the fast memory pathway performs short-term dependency modeling through a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies through a hierarchical sparse attention structure, inputting the historical private domain live broadcast data into the hybrid memory-enhanced neural network, extracting multi-scale temporal features through the fast memory pathway and the slow memory pathway, constructing a spatial pyramid attention module, and performing feature recombination on the multi-scale temporal features according to the parallel attention branches in the spatial pyramid attention module, including: Sampling private domain live broadcast historical data every minute, collecting viewer data, like data, comment data, and reward amount data, and performing weighted average calculation on the viewer data, like data, comment data, and reward amount data to obtain historical trend characteristics; A fast memory path and a slow memory path of a hybrid memory-enhanced neural network are set up, a bidirectional gated recurrent unit network is configured in the fast memory path, a candidate state generation unit is set up in the bidirectional gated recurrent unit network, and a nonlinear combination of a current input and a historical state is calculated by the candidate state generation unit to generate a candidate hidden state; the historical state is randomly forgotten through a reset gate, and the candidate hidden state is updated through an update gate; A converter network is configured in the slow memory path, a multi-head self-attention layer is combined with a feedforward neural network layer to construct a computational block, a query vector, a key vector, and a value vector are projected into multiple feature subspaces in the multi-head self-attention layer, a scaled dot product calculation is performed in each feature subspace to obtain an attention weight, dropout regularization is performed on the attention weight, and two layers of perceptrons are connected in series in the feedforward neural network layer, wherein the first layer of perceptrons performs dimensionality expansion and the second layer of perceptrons performs dimensionality compression, and a normalization layer is added between the layers; Performing an adaptive batch normalization operation on the private domain live broadcast historical data and the historical trend features to obtain normalized data, reorganizing the normalized data into a three-dimensional array comprising a day dimension, a minute dimension, and an indicator dimension, inputting the three-dimensional array into the fast memory path, bidirectionally scanning along the day dimension to obtain local time series features, inputting the three-dimensional array into the slow memory path, and adding time series position information through position encoding to obtain global time series features; Configure the first attention branch, second attention branch and third attention branch of the spatial pyramid attention module, set a sparse attention mechanism in the first attention branch, generate an adaptive attention mask, perform information aggregation of key temporal positions through Top-k selection, set a dense attention mechanism in the second attention branch, perform feature extraction of local areas through a deformable sliding window, set a hybrid attention mechanism in the third attention branch, cascade channel attention and spatial attention, extract cross-channel and cross-spatial dependencies, align the local temporal features with the global temporal features in the time dimension and input them into the spatial pyramid attention module, and perform weighted fusion of the output features of the three attention branches to obtain multi-scale fusion features.

[0006] In an optional embodiment, By combining deformable convolution with feature receptive field adjustment, a hierarchical feature representation is obtained. This hierarchical feature representation is input into a progressive feature distillation network, which performs multi-level knowledge transfer and feature recombination operations to generate live broadcast peak period prediction results, including: Configuring a multi-layer deformable convolutional network with an offset prediction layer, a modulation factor prediction layer, and a feature sampling layer, calculating a two-dimensional offset of a sampling position of a pre-acquired multi-scale fusion feature in the offset prediction layer, generating a modulation factor of the sampling position in the modulation factor prediction layer, adaptively adjusting the feature response, performing bilinear interpolation sampling on the feature map in the feature sampling layer in combination with the two-dimensional offset and the modulation factor, sequentially inputting the multi-scale fusion feature into each layer of the multi-layer deformable convolutional network, adding cross-layer connections between adjacent layers, and generating a hierarchical feature representation; A teacher network and a multi-level student network of a progressive feature distillation network are configured, parameters of the teacher network are initialized according to the hierarchical feature representation, each level of the student network is associated with a feature layer of different depths, features of adjacent levels of the student network are reused through dense residual connections, an attention distillation module is introduced into the teacher network, an attention map is used as an intermediate supervision signal, the hierarchical feature representation is input into the progressive feature distillation network, the L2 loss and attention map loss of the feature maps of the teacher network and the student network are calculated, the KL divergence loss corresponding to the soft label distribution is calculated, multiple loss terms are weightedly combined to obtain a total loss, network parameters are optimized based on the total loss, and a distilled feature representation is generated; The distilled feature representation is input into a multi-layer perceptron, in which the feature dimension is reduced layer by layer, and the output dimension is set to the number of discrete time periods in the prediction time window in the last layer; the output is normalized by the Softmax function to obtain the peak probability of each time period, and the optimal threshold is determined by combining the dynamic probability threshold selection algorithm, and the time period above the optimal threshold is marked as the predicted peak period.

[0007] In an optional embodiment, Collect live broadcast-related multimodal data and perform feature extraction and feature encoding to construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network. Combined with the topology structure generator, the node association relationship is constructed. The cross-modal feature fusion is performed through the feature interaction engine. The hierarchical contrast learning training method is applied to the dynamic heterogeneous graph neural network. Sample pairs are constructed at different semantic levels and feature comparison is performed. The feature comparison results are input into the hierarchical strategy optimization network. The global goal is divided into multiple sub-goals through the goal decomposition module. The sub-goal realization path is planned according to the execution planning module. The hierarchical strategy optimization network is input into the multi-agent reinforcement learning framework of non-cooperative game. The strategy space is optimized through multi-agent competition and collaboration. The output live content scheduling strategy includes: Perform feature extraction on each frame of the video stream data through a feature extraction network, the feature extraction network includes a feature pyramid network and an attention enhancement module, extract multi-scale visual features through the feature pyramid network, and perform feature enhancement in the channel dimension and spatial dimension through the attention enhancement module. Combined with the multi-head self-attention network, a short-time Fourier transform is performed on the audio data to obtain a spectrogram, and the spectrogram is input into a deep network to extract audio features. A bidirectional encoding network is used to perform sequence modeling on the barrage text and comment text, and the word vector sequence is mapped into a text feature vector; the visual features, the audio features, and the text feature vector are input into a temporal encoding network, and temporal encoding is performed to obtain a multimodal feature tensor; Construct a dynamic heterogeneous graph neural network, using anchor nodes, audience nodes, video clip nodes, audio clip nodes, bullet comment nodes, and comment nodes as graph nodes, assigning the multimodal feature tensor to the corresponding graph nodes, calculating the similarity between node pairs using a topology generator, performing a normalization transformation on the similarity to obtain edge weights, constructing a node adjacency matrix using a sparse matrix representation, and using a feature matrix to store node attributes to obtain a feature representation of the node; Dynamically updating the dynamic heterogeneous graph neural network, extracting corresponding modal features and creating new nodes when new data arrives, calculating feature similarity between the newly added nodes and existing nodes corresponding to the new data, adding connecting edges between node pairs whose similarity is greater than a preset first threshold, calculating the product of the degree centrality of the existing nodes and the time decay factor, deleting nodes whose product is less than a preset second threshold, calculating the exponential sliding average of the edge weights, deleting edges whose sliding average is less than a preset third threshold, and obtaining an optimized graph structure; The feature matrix and the node adjacency matrix are input into the multi-layer graph neural network through the feature interaction engine, wherein the feature matrix is ​​composed of the multimodal feature tensor of the node, which represents the initial feature of each node. A learnable parameter matrix is ​​constructed according to the input feature dimension and the hidden layer dimension. First, the input dimension and output dimension of each layer of the graph neural network are set, and the learnable parameter matrix is ​​initialized using uniform distribution or normal distribution. The learnable parameter matrix is ​​used as the weight matrix of each layer of the graph neural network, and the node adjacency matrix is ​​degree-normalized to obtain a symmetric normalized matrix. The normalized matrix is ​​multiplied by the node feature matrix of the current layer, and then multiplied by the learnable parameter matrix. The nonlinear activation function is used. Obtain the node features of the next layer, apply the hierarchical contrast learning method at each layer, construct positive sample pairs and negative sample pairs, where positive sample pairs are sampled from adjacent nodes and negative sample pairs are sampled from non-adjacent nodes at different semantic levels, calculate the feature contrast loss, and perform weighted average of the contrast losses at different levels to obtain the total loss. Input the node representation output by the graph neural network into the hierarchical strategy optimization network, set the interaction quality improvement sub-goal, content form optimization sub-goal, and user experience optimization sub-goal in the goal decomposition module, calculate the sub-goal weights, and set the resource scheduling sub-network in the execution planning module, wherein the resource scheduling sub-network includes a multi-layer encoder, and the output of the encoder is mapped to obtain the scheduling action space; The nodes in a dynamic heterogeneous graph neural network are mapped into multiple agents. Each agent maintains a local state representation and a policy network. The policy network consists of an action generation network and a state evaluation network. The action generation network outputs an action probability distribution, and the state evaluation network evaluates the state value function. The agents exchange information through a competitive collaboration mechanism and optimize strategies based on a non-cooperative game. Rewards are calculated based on the joint actions of the agents. The reward function comprehensively considers multiple indicators such as interactivity, content quality, and user feedback to obtain a comprehensive evaluation score. Each agent generates a probability distribution of live content scheduling strategies based on its policy network, performs parameterized sampling on the probability distribution to obtain discrete actions, and combines the discrete actions of all agents into a live content scheduling strategy, including a sequence of host interaction instructions, a sequence of audience activity push, and a sequence of platform resource allocation. The structure and node features of the dynamic heterogeneous graph neural network are updated based on the feedback after executing the scheduling strategy.

[0008] In an optional embodiment, The agents exchange information through a competitive cooperation mechanism, optimize strategies based on non-cooperative games, and calculate rewards based on the joint actions of the agents, including: An agent is deployed at each node of a dynamic heterogeneous graph. The nodes of the dynamic heterogeneous graph include video clip nodes, audio clip nodes, and bullet comment nodes. The state space of the agent is composed of the attribute feature vector of the corresponding node and the attribute feature vector of the adjacent nodes. The state vector of the video clip node contains the visual features of the video frame, the acoustic features of the audio clip, and the text features of the surrounding bullet comment nodes. The action space of the agent includes picture clarity adjustment, interactive question insertion, and related video push. The agent broadcasts its own action probability distribution to neighboring nodes at each decision time step, receives the action probability distribution of neighboring nodes, and uses the received action probability distribution as an additional input signal to update the corresponding policy network, which includes an action generation network and a value evaluation network; Strategy optimization is performed based on a non-cooperative game framework. The objective functions of different node types in the non-cooperative game include audience engagement for anchor nodes, viewing time for video nodes, and number of interactions for bullet comments nodes. The agent takes actions based on observed states and the actions of other agents to obtain immediate rewards. It also receives trajectory data of other agents' interactions with the environment to estimate the joint action value function. Iteratively update the joint action value function, calculate the value function of the current agent state and the joint action, and take the weighted average of the difference between the value function and the sum of the immediate reward and the discounted value of the next state value function to obtain the updated value function, where the immediate reward is obtained according to the environment and the weight of the weighted average is the learning rate; The policy function is optimized based on the policy gradient algorithm, and the Monte Carlo estimation method is used to calculate the policy gradient of each agent's objective function. The product of the policy gradient and the learning rate is accumulated and added to the parameters of the agent's policy function to obtain the updated policy function parameters. The agent generates a scheduling strategy based on the optimized policy function, and performs picture clarity adjustment, interactive question insertion, and related video push according to the scheduling strategy.

[0009] In an optional embodiment, Real-time live broadcast data is collected and added to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features based on a cross-modal fusion method. The effect evaluator establishes a mapping relationship between predicted effects and actual effects based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak period prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Repeated collection and adjustment are performed to obtain the optimal content scheduling strategy, including: Adding the collected real-time video stream, real-time audio stream, and real-time text stream data to a pre-built two-way interactive evaluation network, wherein the two-way interactive evaluation network includes a state encoder and an effect evaluator, wherein the state encoder includes a video feature extraction module, an audio feature extraction module, and a text feature extraction module, and the effect evaluator includes a causal graph construction module, a structural equation model, and a multi-layer perceptron network; Extracting frame-level visual features from a real-time video stream using a convolutional neural network in the video feature extraction module, extracting audio spectrogram features from a real-time audio stream using a convolutional neural network in the audio feature extraction module, and extracting semantic features from a real-time text stream using a recurrent neural network in the text feature extraction module. Inputting the frame-level visual features, the audio spectrogram features, and the semantic features into an attention network to generate a feature weight matrix. Based on the feature weight matrix, weighted fusion of the three features is performed to obtain a multimodal representation vector. The causal graph construction module constructs a live broadcast effect causal graph based on historical data. The live broadcast effect causal graph includes a scheduling strategy node, a live broadcast status node, an external factor node, and an effect indicator node. The edges between the nodes represent causal relationships. The structural equation model defines the functional relationship between each node. The multimodal representation vector and the preset scheduling strategy vector are spliced ​​to form an input vector. Causal reasoning is performed based on the causal graph and the structural equation model to obtain a policy intervention effect. The input vector and the policy intervention effect are input into the multi-layer perceptron network. The multi-layer perceptron network calculates the predicted number of viewers, the predicted number of likes, and the predicted reward amount, counts the real-time number of viewers, the real-time number of likes, and the real-time reward amount, and calculates the weighted average deviation between the predicted value and the actual value to obtain a real-time performance evaluation index. Calculating a deviation between the real-time performance evaluation index and a stored live broadcast peak period prediction result, comparing the deviation with a preset deviation threshold, and triggering a policy adjustment when the deviation is greater than the deviation threshold; The live content scheduling task is decomposed into content recommendation subtask, anchor interaction subtask, and audience activity subtask. Each subtask is configured with an independent policy network and value function network. The policy network generates an action probability distribution based on the current state, and the value function network evaluates the value of the state-action pair. A gradient meta-learner is trained for each subtask. The gradient meta-learner adopts a model-independent meta-learning algorithm. In the inner loop, the policy gradient method is used to update the network parameters based on the latest state data, action data, and reward data. In the outer loop, the meta-parameters are updated by minimizing the meta-objective loss across tasks. The action probability distributions output by the updated policy networks corresponding to the three subtasks are normalized and fused through a soft voting mechanism to obtain the action probability distribution of the global scheduling strategy. Based on the action probability distribution, live content scheduling actions are sampled and selected, and the optimal content scheduling strategy is obtained by combination. While executing the scheduling action, new live broadcast data is continuously collected, and the newly collected data is repeatedly input into the two-way interactive evaluation network for evaluation and optimization until the deviation value between the real-time performance evaluation index and the live broadcast peak period prediction result is less than the deviation threshold.

[0010] In an optional embodiment, The gradient meta-learner adopts a model-independent meta-learning algorithm. In the inner loop, the policy gradient method is used to update the network parameters based on the latest state data, action data, and reward data. In the outer loop, the meta-parameters are updated by minimizing the cross-task meta-objective loss. The update includes: Construct a meta-learner network structure, set up a policy network to generate action probability distributions, configure a meta-optimizer to adjust policy network parameters, implement the policy network as a deep neural network structure, and initialize meta-parameters shared by the policy network and meta-optimizer; For the pre-obtained subtasks, collect historical status data, action data, reward data, and effect data for each subtask to construct a training task, and obtain new status data, action data, reward data, and effect data from the real-time live broadcast scene to construct a test task; Sampling training tasks to obtain state transition data, copying meta-parameters as initial values ​​of task-specific parameters, starting the policy network based on the task-specific parameters, inputting the current state to generate an action probability distribution and sampling actions, executing the actions to obtain immediate rewards and next states, calculating the state value functions of the immediate rewards and next states, calculating the temporal difference error using the Bellman equation to obtain the action value function, taking the negative product of the action value function and the logarithm of the action probability output by the policy network to obtain the policy loss, calculating the gradient of the policy loss with respect to the task-specific parameters, updating the task-specific parameters along the gradient direction, and repeating the policy network forward propagation and gradient update until the inner loop training is completed; Calculate the cumulative reward difference of the task-specific parameters before and after the update of each training task on the corresponding task as the fitness metric, construct a meta-objective loss function by taking the weighted average of the fitness metrics of all tasks, calculate the second-order derivative of the meta-objective loss function with respect to the meta-parameters, and drive the meta-optimizer to update the meta-parameters based on the second-order derivatives to minimize the meta-objective loss function; Sample the test task, copy the trained meta-parameters as the initial values ​​of the task-specific parameters, and execute the inner loop update to obtain the policy network adapted to the current task.

[0011] A second aspect of an embodiment of the present invention provides a deep learning-based system for predicting hotspots and scheduling content during peak periods of private domain live broadcasts, including: The first unit is used to collect historical private domain live broadcast data and construct a hybrid memory-enhanced neural network including a fast memory pathway and a slow memory pathway. The fast memory pathway performs short-term dependency modeling through a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies through a hierarchical sparse attention structure. The private domain live broadcast historical data is input into the hybrid memory-enhanced neural network, multi-scale temporal features are extracted through the fast and slow memory pathways, a spatial pyramid attention module is constructed, and feature recombinations are performed on the multi-scale temporal features according to the parallel attention branches in the spatial pyramid attention module. The feature receptive field is adjusted in combination with deformable convolution to obtain a hierarchical feature representation. The hierarchical feature representation is input into a progressive feature distillation network, which performs multi-level knowledge transfer and feature recombinations to generate a live broadcast peak period prediction result. The second unit is used to collect multimodal data related to live broadcast and perform feature extraction and feature encoding, construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network, combine the topology structure generator to build node association relationships, perform cross-modal feature fusion through the feature interaction engine, apply a hierarchical contrast learning training method to the dynamic heterogeneous graph neural network, construct sample pairs at different semantic levels and perform feature comparison, input the feature comparison results into a hierarchical strategy optimization network, divide the global goal into multiple sub-goals through the goal decomposition module, plan the sub-goal realization path according to the execution planning module, input the hierarchical strategy optimization network into a non-cooperative game multi-agent reinforcement learning framework, optimize the strategy space through multi-agent competition and collaboration, and output the live content scheduling strategy; The third unit is used to collect real-time live broadcast data and add it to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features according to a cross-modal fusion method. The effect evaluator establishes a mapping relationship between predicted effects and actual effects based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Collection and adjustment are repeated to obtain the optimal content scheduling strategy.

[0012] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0013] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0014] In the present invention, by combining fast and slow memory pathways and a spatial pyramid attention module, the short-term and long-term dependencies in live broadcast data can be more effectively captured, thereby more accurately predicting the peak period of live broadcast and providing a reliable basis for subsequent content scheduling. By utilizing multimodal data, dynamic heterogeneous graph neural networks and hierarchical comparative learning, the live broadcast content can be more finely scheduled, and optimized in the strategy space through a multi-agent reinforcement learning framework, thereby improving the attractiveness of the live broadcast content and user engagement. A two-way interactive evaluation network is used to evaluate the live broadcast effect in real time, and a meta-reinforcement learning algorithm is used to perform online optimization of the content scheduling strategy, so that the live broadcast content can be dynamically adjusted according to actual conditions to ensure that the live broadcast effect reaches the best state. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flowchart of a method for predicting hot spots and scheduling content during peak periods of private domain live broadcasts based on deep learning according to an embodiment of the present invention; Figure 2 This is a structural diagram of a deep learning-based private domain live broadcast peak period hotspot prediction and content scheduling system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0017] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0018] Figure 1 This is a flow chart of a method for predicting hot spots and scheduling content during peak periods of private domain live broadcasts based on deep learning according to an embodiment of the present invention. Figure 1 As shown, the method includes: S1. Collect historical private domain live broadcast data and construct a hybrid memory-enhanced neural network consisting of a fast memory pathway and a slow memory pathway. The fast memory pathway models short-term dependencies using a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies using a hierarchical sparse attention structure. Input the historical private domain live broadcast data into the hybrid memory-enhanced neural network, extract multi-scale temporal features through the fast and slow memory pathways, construct a spatial pyramid attention module, and perform feature recombination on the multi-scale temporal features using the parallel attention branches in the spatial pyramid attention module. Adjust the feature receptive field using deformable convolution to obtain a hierarchical feature representation. Input the hierarchical feature representation into a progressive feature distillation network, perform multi-level knowledge transfer and feature recombination operations, and generate a live broadcast peak period prediction result. In an optional embodiment, Collecting historical private domain live broadcast data, constructing a hybrid memory-enhanced neural network including a fast memory pathway and a slow memory pathway, wherein the fast memory pathway performs short-term dependency modeling through a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies through a hierarchical sparse attention structure, inputting the historical private domain live broadcast data into the hybrid memory-enhanced neural network, extracting multi-scale temporal features through the fast memory pathway and the slow memory pathway, constructing a spatial pyramid attention module, and performing feature recombination on the multi-scale temporal features according to the parallel attention branches in the spatial pyramid attention module, including: Sampling private domain live broadcast historical data every minute, collecting viewer data, like data, comment data, and reward amount data, and performing weighted average calculation on the viewer data, like data, comment data, and reward amount data to obtain historical trend characteristics; A fast memory path and a slow memory path of a hybrid memory-enhanced neural network are set up, a bidirectional gated recurrent unit network is configured in the fast memory path, a candidate state generation unit is set up in the bidirectional gated recurrent unit network, and a nonlinear combination of a current input and a historical state is calculated by the candidate state generation unit to generate a candidate hidden state; the historical state is randomly forgotten through a reset gate, and the candidate hidden state is updated through an update gate; A converter network is configured in the slow memory path, a multi-head self-attention layer is combined with a feedforward neural network layer to construct a computational block, a query vector, a key vector, and a value vector are projected into multiple feature subspaces in the multi-head self-attention layer, a scaled dot product calculation is performed in each feature subspace to obtain an attention weight, dropout regularization is performed on the attention weight, and two layers of perceptrons are connected in series in the feedforward neural network layer, wherein the first layer of perceptrons performs dimensionality expansion and the second layer of perceptrons performs dimensionality compression, and a normalization layer is added between the layers; Performing an adaptive batch normalization operation on the private domain live broadcast historical data and the historical trend features to obtain normalized data, reorganizing the normalized data into a three-dimensional array comprising a day dimension, a minute dimension, and an indicator dimension, inputting the three-dimensional array into the fast memory path, bidirectionally scanning along the day dimension to obtain local time series features, inputting the three-dimensional array into the slow memory path, and adding time series position information through position encoding to obtain global time series features; Configure the first attention branch, second attention branch and third attention branch of the spatial pyramid attention module, set a sparse attention mechanism in the first attention branch, generate an adaptive attention mask, perform information aggregation of key temporal positions through Top-k selection, set a dense attention mechanism in the second attention branch, perform feature extraction of local areas through a deformable sliding window, set a hybrid attention mechanism in the third attention branch, cascade channel attention and spatial attention, extract cross-channel and cross-spatial dependencies, align the local temporal features with the global temporal features in the time dimension and input them into the spatial pyramid attention module, and perform weighted fusion of the output features of the three attention branches to obtain multi-scale fusion features.

[0019] In an optional embodiment, Get historical data from private live streaming platforms, including minute-by-minute viewership, likes, comments, and tip amounts. For example, within 10 minutes of a live stream, 100, 120, 150, 180, 200, 220, 250, 280, 300, and 320 people watched it, with 10, 12, 15, 18, 20, 22, 25, 28, 30, and 32 likes, 5, 6, 7, 8, 9, 10, 11, 12, 13, and 14 comments, and 10, 12, 15, 18, 20, 22, 25, 28, 30, and 32 tip amounts.

[0020] Perform a weighted average calculation to obtain the historical trend characteristics. For example, set the weight of the number of viewers to 0.5, the weight of the number of likes to 0.2, the weight of the number of comments to 0.2, and the weight of the amount of rewards to 0.1. The historical trend characteristic value at the 10th minute is (320×0.5+32×0.2+14×0.2+32×0.1)=172.4.

[0021] A hybrid memory-enhanced neural network is constructed, consisting of a fast memory pathway and a slow memory pathway. The fast memory pathway uses a bidirectional gated recurrent unit network, with reset and update gates controlling the flow and forgetting of information. For example, the reset gate determines how much of the previous memory is retained, while the update gate determines how much of the current input information is added to the memory.

[0022] The slow memory pathway uses a transformer network consisting of a multi-head self-attention layer and a feedforward neural network layer. The multi-head self-attention layer projects the input data into multiple feature subspaces and calculates attention weights in each subspace to capture long-term dependencies. The feedforward neural network layer consists of two perceptron layers and an inter-layer normalization layer for nonlinear feature transformation.

[0023] Adaptive batch normalization is performed on the collected private domain live broadcast historical data and the calculated historical trend features. For example, the number of viewers per minute, the number of likes, comments, and reward amounts, as well as the historical trend feature values, are normalized so that their values ​​fall within a reasonable range, such as between 0 and 1.

[0024] Reorganize the normalized data into a three-dimensional array, consisting of day, minute, and metric dimensions. For example, organize seven days of live streaming data, including minute-by-minute viewership, likes, comments, reward amounts, and historical trend features, into a 7x1440x5 three-dimensional array (assuming 24-hour live streaming per day).

[0025] The three-dimensional array is input into the fast and slow memory pathways. The fast memory pathway scans bidirectionally along the day dimension to extract local temporal features. The slow memory pathway adds temporal position information through positional encoding to extract global temporal features.

[0026] A spatial pyramid attention module is constructed, consisting of three parallel attention branches: the first branch uses a sparse attention mechanism, the second branch uses a dense attention mechanism, and the third branch uses a hybrid attention mechanism. Local and global temporal features are aligned in the temporal dimension and input into the spatial pyramid attention module. The output features of the three attention branches are weightedly fused to generate multi-scale fused features for predicting live broadcast performance.

[0027] In this embodiment, by hybridizing the memory pathway and the spatial pyramid attention module, multi-scale temporal features can be extracted more effectively, thereby improving prediction accuracy. The application of technologies such as adaptive batch normalization and dropout regularization enhances the robustness of the model, making it more resistant to noise and abnormal data. Through the different attention branches of the spatial pyramid attention module, the impact of different time scales and spatial positions on the live broadcast effect can be analyzed, thereby improving the interpretability of the model.

[0028] In an optional embodiment, By combining deformable convolution with feature receptive field adjustment, a hierarchical feature representation is obtained. This hierarchical feature representation is input into a progressive feature distillation network, which performs multi-level knowledge transfer and feature recombination operations to generate live broadcast peak period prediction results, including: Configuring a multi-layer deformable convolutional network with an offset prediction layer, a modulation factor prediction layer, and a feature sampling layer, calculating a two-dimensional offset of a sampling position of a pre-acquired multi-scale fusion feature in the offset prediction layer, generating a modulation factor of the sampling position in the modulation factor prediction layer, adaptively adjusting the feature response, performing bilinear interpolation sampling on the feature map in the feature sampling layer in combination with the two-dimensional offset and the modulation factor, sequentially inputting the multi-scale fusion feature into each layer of the multi-layer deformable convolutional network, adding cross-layer connections between adjacent layers, and generating a hierarchical feature representation; A teacher network and a multi-level student network of a progressive feature distillation network are configured, parameters of the teacher network are initialized according to the hierarchical feature representation, each level of the student network is associated with a feature layer of different depths, features of adjacent levels of the student network are reused through dense residual connections, an attention distillation module is introduced into the teacher network, an attention map is used as an intermediate supervision signal, the hierarchical feature representation is input into the progressive feature distillation network, the L2 loss and attention map loss of the feature maps of the teacher network and the student network are calculated, the KL divergence loss corresponding to the soft label distribution is calculated, multiple loss terms are weightedly combined to obtain a total loss, network parameters are optimized based on the total loss, and a distilled feature representation is generated; The distilled feature representation is input into a multi-layer perceptron, in which the feature dimension is reduced layer by layer, and the output dimension is set to the number of discrete time periods in the prediction time window in the last layer; the output is normalized by the Softmax function to obtain the peak probability of each time period, and the optimal threshold is determined by combining the dynamic probability threshold selection algorithm, and the time period above the optimal threshold is marked as the predicted peak period.

[0029] In an optional embodiment, Prepare training data by collecting historical live broadcast data, including multi-dimensional data such as the number of viewers, number of comments, number of likes, number of shares, and corresponding timestamps. Divide the data into predefined time windows (for example, 15-minute windows), and count the various metrics within each time window. Set a threshold based on the peak number of viewers in each time window in the historical data. Time windows above the threshold are marked as peak hours, and those below the threshold are marked as off-peak hours. This constructs a training dataset containing multi-dimensional feature data and corresponding labels (peak or off-peak).

[0030] A multi-layer deformable convolutional network is constructed, consisting of multiple deformable convolutional layers. Each layer consists of an offset prediction layer, a modulation factor prediction layer, and a feature sampling layer. First, the multi-scale fused features are input to the first deformable convolutional layer. In the offset prediction layer, the network calculates a two-dimensional offset for each sampling point based on the input multi-scale fused features. The offset represents the offset position of the sampling point on the feature map. In the modulation factor prediction layer, the network generates a modulation factor for each sampling position, which is used to adjust the strength of the feature response. The modulation factor ranges from 0 to 1, with a larger value indicating a higher importance of the sampling point. In the feature sampling layer, the network performs bilinear interpolation sampling on the feature map based on the calculated two-dimensional offset and modulation factor. Specifically, the network finds new sampling positions based on the offsets and performs a weighted sum of the sampled values ​​according to the modulation factor. The output of the first deformable convolutional layer is input to the next deformable convolutional layer, and the above process is repeated. Cross-layer connections are added between adjacent deformable convolutional layers to fuse the output of the previous layer with the input of the current layer to generate a hierarchical feature representation. For example, assuming the input feature map size is 256x256, the offset output by the offset prediction layer is (0.5, -0.2), and the modulation factor is 0.8, the actual sampling position is (256×0.5, 256×(-0.2)) = (128, -51.2). Since the offset can be a decimal, bilinear interpolation is required to calculate the new sampling value.

[0031] A progressive feature distillation network is constructed, consisting of a teacher network and multiple student networks. The parameters of the teacher network are initialized using the previously generated hierarchical feature representation. Each student network corresponds to a feature layer of different depth. For example, the first student network corresponds to shallow features, the second student network corresponds to intermediate features, and so on. Adjacent student networks reuse features through dense residual connections, with the output of the previous student network serving as part of the input to the next student network. An attention distillation module is introduced into the teacher network, using the attention map generated by the teacher network as an intermediate supervisory signal to guide the learning of the student network. The hierarchical feature representation is input into the progressive feature distillation network. The sum of squared differences between the feature maps of the teacher network and each student network, as well as the sum of squared differences between the attention maps, is calculated. The KL divergence between the soft label distributions output by the teacher network and those of the student network is also calculated. These loss terms are weighted to form a total loss. For example, the feature map loss has a weight of 0.5, the attention map loss has a weight of 0.3, and the soft label loss has a weight of 0.2. Based on the total loss, the network parameters are optimized using the backpropagation algorithm to generate the distilled feature representation.

[0032] The distilled feature representation is input into a multilayer perceptron. Within the multilayer perceptron, the feature dimension is reduced layer by layer. For example, the first layer reduces the feature dimension to 256, the second layer to 128, and so on. In the final layer, the output dimension is set to the number of discrete time periods in the prediction window. For example, if a day is divided into 96 15-minute time windows, the output dimension is 96. The output is normalized using the Softmax function to obtain the peak probability for each time window. A dynamic probability threshold selection algorithm is used, for example, by adaptively determining the optimal threshold based on the distribution of peak probability statistics from historical data. Time windows above the optimal threshold are marked as predicted peak periods. For example, if the prediction results are [0.1, 0.2, 0.8, 0.3, ..., 0.9] and the optimal threshold is 0.7, then the predicted peak period is the third and final time window.

[0033] In this embodiment, deformable convolution can adaptively adjust the receptive field to capture the dynamic change patterns in live broadcast data. Progressive feature distillation can effectively transfer knowledge and improve the generalization ability of the model, thereby improving the accuracy of peak period prediction. Multi-scale feature fusion can better cope with the diversity and complexity of live broadcast data. The attention distillation module can guide the model to focus on important feature information and enhance the robustness of the model. The progressive feature distillation network can reduce the computational complexity of the model while ensuring prediction performance and improve prediction efficiency.

[0034] S2. Collect multimodal data related to live broadcasts and perform feature extraction and encoding to construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network. Node associations are constructed using a topology generator. Cross-modal feature fusion is performed through a feature interaction engine. A hierarchical contrastive learning training method is applied to the dynamic heterogeneous graph neural network. Sample pairs are constructed at different semantic levels and feature comparisons are performed. The feature comparison results are input into a hierarchical strategy optimization network. The global goal is divided into multiple sub-goals through a goal decomposition module. The sub-goal achievement paths are planned according to the execution planning module. The hierarchical strategy optimization network is input into a non-cooperative multi-agent reinforcement learning framework. The strategy space is optimized through multi-agent competition and collaboration to output a live broadcast content scheduling strategy. In an optional embodiment, Collect live broadcast-related multimodal data and perform feature extraction and feature encoding to construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network. Combined with the topology structure generator, the node association relationship is constructed. The cross-modal feature fusion is performed through the feature interaction engine. The hierarchical contrast learning training method is applied to the dynamic heterogeneous graph neural network. Sample pairs are constructed at different semantic levels and feature comparison is performed. The feature comparison results are input into the hierarchical strategy optimization network. The global goal is divided into multiple sub-goals through the goal decomposition module. The sub-goal realization path is planned according to the execution planning module. The hierarchical strategy optimization network is input into the multi-agent reinforcement learning framework of non-cooperative game. The strategy space is optimized through multi-agent competition and collaboration. The output live content scheduling strategy includes: Perform feature extraction on each frame of the video stream data through a feature extraction network, the feature extraction network includes a feature pyramid network and an attention enhancement module, extract multi-scale visual features through the feature pyramid network, and perform feature enhancement in the channel dimension and spatial dimension through the attention enhancement module. Combined with the multi-head self-attention network, a short-time Fourier transform is performed on the audio data to obtain a spectrogram, and the spectrogram is input into a deep network to extract audio features. A bidirectional encoding network is used to perform sequence modeling on the barrage text and comment text, and the word vector sequence is mapped into a text feature vector; the visual features, the audio features, and the text feature vector are input into a temporal encoding network, and temporal encoding is performed to obtain a multimodal feature tensor; Construct a dynamic heterogeneous graph neural network, using anchor nodes, audience nodes, video clip nodes, audio clip nodes, bullet comment nodes, and comment nodes as graph nodes, assigning the multimodal feature tensor to the corresponding graph nodes, calculating the similarity between node pairs using a topology generator, performing a normalization transformation on the similarity to obtain edge weights, constructing a node adjacency matrix using a sparse matrix representation, and using a feature matrix to store node attributes to obtain a feature representation of the node; Dynamically updating the dynamic heterogeneous graph neural network, extracting corresponding modal features and creating new nodes when new data arrives, calculating feature similarity between the newly added nodes and existing nodes corresponding to the new data, adding connecting edges between node pairs whose similarity is greater than a preset first threshold, calculating the product of the degree centrality of the existing nodes and the time decay factor, deleting nodes whose product is less than a preset second threshold, calculating the exponential sliding average of the edge weights, deleting edges whose sliding average is less than a preset third threshold, and obtaining an optimized graph structure; The feature matrix and the node adjacency matrix are input into the multi-layer graph neural network through the feature interaction engine, wherein the feature matrix is ​​composed of the multimodal feature tensor of the node, which represents the initial feature of each node. A learnable parameter matrix is ​​constructed according to the input feature dimension and the hidden layer dimension. First, the input dimension and output dimension of each layer of the graph neural network are set, and the learnable parameter matrix is ​​initialized using uniform distribution or normal distribution. The learnable parameter matrix is ​​used as the weight matrix of each layer of the graph neural network, and the node adjacency matrix is ​​degree-normalized to obtain a symmetric normalized matrix. The normalized matrix is ​​multiplied by the node feature matrix of the current layer, and then multiplied by the learnable parameter matrix. The nonlinear activation function is used. Obtain the node features of the next layer, apply the hierarchical contrast learning method at each layer, construct positive sample pairs and negative sample pairs, where positive sample pairs are sampled from adjacent nodes and negative sample pairs are sampled from non-adjacent nodes at different semantic levels, calculate the feature contrast loss, and perform weighted average of the contrast losses at different levels to obtain the total loss. Input the node representation output by the graph neural network into the hierarchical strategy optimization network, set the interaction quality improvement sub-goal, content form optimization sub-goal, and user experience optimization sub-goal in the goal decomposition module, calculate the sub-goal weights, and set the resource scheduling sub-network in the execution planning module, wherein the resource scheduling sub-network includes a multi-layer encoder, and the output of the encoder is mapped to obtain the scheduling action space; The nodes in a dynamic heterogeneous graph neural network are mapped into multiple agents. Each agent maintains a local state representation and a policy network. The policy network consists of an action generation network and a state evaluation network. The action generation network outputs an action probability distribution, and the state evaluation network evaluates the state value function. The agents exchange information through a competitive collaboration mechanism and optimize strategies based on a non-cooperative game. Rewards are calculated based on the joint actions of the agents. The reward function comprehensively considers multiple indicators such as interactivity, content quality, and user feedback to obtain a comprehensive evaluation score. Each agent generates a probability distribution of live content scheduling strategies based on its policy network, performs parameterized sampling on the probability distribution to obtain discrete actions, and combines the discrete actions of all agents into a live content scheduling strategy, including a sequence of host interaction instructions, a sequence of audience activity push, and a sequence of platform resource allocation. The structure and node features of the dynamic heterogeneous graph neural network are updated based on the feedback after executing the scheduling strategy.

[0035] In an optional embodiment, Multimodal live streaming data is collected, including video streams, audio streams, bullet comments, and comment text. For video streams, a feature pyramid network is used to extract multi-scale visual features. For example, for a 1080p video frame, feature maps of different scales, such as 256x144, 512x288, and 1024x576, can be extracted to capture visual information at different granularities. An attention enhancement module is then used to enhance features in the channel and spatial dimensions. For example, an attention mechanism such as SE-Net can be used to enhance features of important channels and regions. For audio streams, a short-time Fourier transform is performed to obtain a spectrogram, which is then fed into a deep network (such as a CNN) to extract audio features. For example, an audio clip can be converted into a mel-spectrogram and then fed into a deep network to extract feature vectors. For bullet comments and comment text, a bidirectional encoder network (such as BERT) is used for sequence modeling, mapping word vector sequences to text feature vectors. For example, after tokenizing each bullet comment, each word is converted into a word vector, which is then fed into a bidirectional encoder network model to obtain a feature representation of the text. The extracted visual, audio, and text features are fed into a temporal encoding network, which performs temporal encoding to generate a multimodal feature tensor. For example, concatenating 5 seconds of video, audio, and text features and feeding them into the temporal encoding network yields a feature tensor that incorporates temporal information.

[0036] Construct a dynamic heterogeneous graph neural network. Take elements such as anchors, viewers, video clips, audio clips, bullet comments, and comments as graph nodes, and assign the extracted multimodal feature tensors to the corresponding graph nodes. For example, the feature of a video clip node is the multimodal feature tensor of the clip. Calculate the similarity between node pairs through the topology generator. For example, calculate the cosine similarity of visual features between two video clip nodes. Perform a normalization transformation on the similarity to obtain the edge weight. For example, normalize the similarity value to between 0 and 1. Construct a node adjacency matrix using a sparse matrix representation. For example, if the similarity between two nodes is greater than 0.8, the value of the corresponding position in the adjacency matrix is ​​set to 1, otherwise it is set to 0. Use the feature matrix to store node attributes to obtain the feature representation of the node.

[0037] Dynamically update the graph neural network structure. When new data arrives, extract the corresponding modal features and create new nodes. Calculate the feature similarity between the newly added nodes and the existing nodes. For example, calculate the similarity between a new video clip and an existing video clip. Add connecting edges between node pairs whose similarity is greater than the preset first threshold (for example, 0.7). Calculate the product of the degree centrality of existing nodes and the time decay factor. For example, if a node has many connecting edges and the time is relatively long, the product will be relatively small. Delete nodes whose product is less than the preset second threshold (for example, 0.1). Calculate the exponential sliding average of edge weights. For example, record the historical weight of each edge and calculate its weighted average. Delete edges whose sliding average is less than the preset third threshold (for example, 0.2) to obtain the optimized graph structure.

[0038] The feature matrix and node adjacency matrix are input into the multi-layer graph neural network through the feature interaction engine. The feature matrix consists of the multimodal feature tensors of the nodes, representing the initial features of each node. A learnable parameter matrix is ​​constructed based on the input feature dimension and the hidden layer dimension. For example, if the input feature dimension is 1024 and the hidden layer dimension is 512, the parameter matrix has dimensions of 1024x512. The parameter matrix is ​​initialized and used as the weight matrix for each layer of the graph neural network. The node adjacency matrix is ​​degree-normalized. The normalized matrix is ​​multiplied by the node feature matrix of the current layer, and then multiplied by the learnable parameter matrix. A nonlinear activation function (such as ReLU) is used to obtain the node features of the next layer. A hierarchical contrastive learning method is applied to each layer. Positive pairs are sampled from adjacent nodes, and negative pairs are sampled from non-adjacent nodes at different semantic levels. Feature contrastive loss is calculated. The weighted average of the contrastive losses at different layers is used to obtain the total loss.

[0039] The node representations output by the graph neural network are input into the hierarchical strategy optimization network. In the goal decomposition module, sub-goals such as interaction quality improvement, content format optimization, and user experience optimization are set, and the sub-goal weights are calculated. For example, based on the actual situation of the current live broadcast, the weight for interaction quality improvement can be set to 0.5, the weight for content format optimization to 0.3, and the weight for user experience optimization to 0.2. In the execution planning module, a resource scheduling sub-network is set, consisting of multiple layers of encoders. The encoder outputs are mapped to the scheduling action space.

[0040] Nodes in a graph neural network are mapped to multiple agents. Each agent maintains a local state representation and a policy network. The policy network consists of an action generation network and a state evaluation network. The action generation network outputs an action probability distribution, while the state evaluation network evaluates the state value function. Agents exchange information through a competitive and collaborative mechanism, optimizing their strategies based on a non-cooperative game. Rewards are calculated based on the joint actions of the agents. The reward function comprehensively considers multiple metrics, such as engagement, content quality, and user feedback, to produce a comprehensive evaluation score. For example, higher engagement, improved content quality, and positive user feedback will result in higher reward scores.

[0041] Each agent generates a probability distribution for its live content scheduling strategy based on its policy network, performing parameterized sampling to generate discrete actions. The discrete actions of all agents are combined into a live content scheduling strategy, including a sequence of interactive commands for the host, a sequence of viewer activity notifications, and a sequence of platform resource allocations. For example, a strategy can generate command sequences such as "the host asks questions to the audience," "send coupons to active viewers," or "recommend the live broadcast to the homepage." The structure and node features of the graph neural network are updated based on feedback from executing the scheduling strategy.

[0042] In this embodiment, through multimodal data analysis and multi-agent reinforcement learning, user interests and behavior patterns can be captured more accurately, thereby formulating more effective interaction strategies and improving the quality of live interaction. By optimizing the network through dynamic heterogeneous graph neural networks and hierarchical strategies, the live content arrangement can be dynamically adjusted according to the real-time status of the live broadcast and user feedback, making the live content more attractive and interesting. Through personalized content recommendations and interaction methods, the needs of different users can be met, and user experience and satisfaction can be improved.

[0043] In an optional embodiment, The agents exchange information through a competitive cooperation mechanism, optimize strategies based on non-cooperative games, and calculate rewards based on the joint actions of the agents, including: An agent is deployed at each node of a dynamic heterogeneous graph. The nodes of the dynamic heterogeneous graph include video clip nodes, audio clip nodes, and bullet comment nodes. The state space of the agent is composed of the attribute feature vector of the corresponding node and the attribute feature vector of the adjacent nodes. The state vector of the video clip node contains the visual features of the video frame, the acoustic features of the audio clip, and the text features of the surrounding bullet comment nodes. The action space of the agent includes picture clarity adjustment, interactive question insertion, and related video push. The agent broadcasts its own action probability distribution to neighboring nodes at each decision time step, receives the action probability distribution of neighboring nodes, and uses the received action probability distribution as an additional input signal to update the corresponding policy network, which includes an action generation network and a value evaluation network; Strategy optimization is performed based on a non-cooperative game framework. The objective functions of different node types in the non-cooperative game include audience engagement for anchor nodes, viewing time for video nodes, and number of interactions for bullet comments nodes. The agent takes actions based on observed states and the actions of other agents to obtain immediate rewards. It also receives trajectory data of other agents' interactions with the environment to estimate the joint action value function. Iteratively update the joint action value function, calculate the value function of the current agent state and the joint action, and take the weighted average of the difference between the value function and the sum of the immediate reward and the discounted value of the next state value function to obtain the updated value function, where the immediate reward is obtained according to the environment and the weight of the weighted average is the learning rate; The policy function is optimized based on the policy gradient algorithm, and the Monte Carlo estimation method is used to calculate the policy gradient of each agent's objective function. The product of the policy gradient and the learning rate is accumulated and added to the parameters of the agent's policy function to obtain the updated policy function parameters. The agent generates a scheduling strategy based on the optimized policy function, and performs picture clarity adjustment, interactive question insertion, and related video push according to the scheduling strategy.

[0044] In an optional embodiment, Construct a dynamic heterogeneous graph. The node types of this graph include video clip nodes, audio clip nodes, and barrage nodes. An agent is deployed at each node. The state vector of the video clip node is composed of the visual features of the video frame (for example, brightness, color histogram, motion vector, etc.), the acoustic features of the audio clip (for example, volume, pitch, MFCC coefficients, etc.), and the text features of the surrounding barrage nodes (for example, sentiment, keywords, word embedding vectors, etc.). The state vectors of the audio clip node and the barrage node are constructed in a similar way, each containing its own attribute features and the attribute features of the adjacent nodes. The action space of the agent includes picture clarity adjustment (for example, improving clarity, reducing clarity), interactive question insertion (for example, inserting questions related to the video content), and related video push (for example, pushing other videos related to the current video topic).

[0045] Agents exchange information and update policies. At each decision time step, each agent broadcasts its own action probability distribution to neighboring nodes and receives action probability distributions from them. These received action probability distributions serve as additional input signals to update the agent's own policy network. The policy network consists of an action generation network and a value evaluation network. The action generation network outputs the agent's own action probability distribution based on the agent's state and the action probability distributions of neighboring nodes. The value evaluation network assesses the long-term value of taking a specific action in the current state.

[0046] Strategy optimization is performed based on a non-cooperative game framework. Different types of nodes have different objective functions. For example, the objective function of the anchor node is to maximize audience engagement (e.g., the number of likes, comments, gifts, etc.), the objective function of the video node is to maximize viewing time, and the objective function of the barrage node is to maximize the number of interactions (e.g., the number of replies, likes, etc.). Each agent takes its own actions based on the observed state and the actions of other agents and receives immediate rewards. At the same time, the agent also receives trajectory data of other agents interacting with the environment to estimate the joint action value function. For example, a video clip node can observe the user's behavior after watching the clip (e.g., whether to continue watching, whether to send barrage, etc.), as well as the state and actions of other related nodes (e.g., audio clip nodes, barrage nodes), to estimate the joint action value function.

[0047] Iteratively update the joint action value function. Specifically, calculate the value function of the current agent state and the joint action, and then perform a weighted average of the difference between the value function and the sum of the immediate reward and the discounted value of the next state to obtain the updated value function. The weight of the weighted average is the learning rate, for example, 0.01. Assuming the value function of the current state is 10, the immediate reward is 2, the value function of the next state is 12, the discount factor is 0.9, and the learning rate is 0.01, the updated value function is 10 + 0.01 × (2 + 0.9 × 12 - 10) = 10.128.

[0048] The policy function is optimized using a policy gradient algorithm, and the Monte Carlo estimation method is used to calculate the policy gradient of each agent's objective function. The product of this policy gradient and the learning rate is accumulated and added to the parameters of the agent's policy function to obtain the updated policy function parameters. For example, if the policy gradient is 0.5 and the learning rate is 0.01, the policy function parameters will increase by 0.005. Based on the optimized policy function, the agent generates a scheduling policy and executes actions such as adjusting image clarity, inserting interactive questions, and pushing relevant videos according to this scheduling policy. For example, if the policy function indicates that image clarity should be improved, the agent will perform actions to improve image clarity.

[0049] In this embodiment, by intelligently adjusting the picture clarity, inserting interactive questions and pushing related videos, the user's viewing experience can be improved and user stickiness can be increased. Personalized recommendations based on user behavior and preferences can be made to improve content distribution efficiency and push more suitable content to more interested users. By increasing audience engagement, viewing time and the number of interactions, the platform operation can be optimized and the overall efficiency of the platform can be improved.

[0050] S3. Collect real-time live broadcast data and add it to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features according to a cross-modal fusion method. The effect evaluator establishes a mapping relationship between the predicted effect and the actual effect based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Collection and adjustment are repeated to obtain the optimal content scheduling strategy.

[0051] In an optional embodiment, Real-time live broadcast data is collected and added to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features based on a cross-modal fusion method. The effect evaluator establishes a mapping relationship between predicted effects and actual effects based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak period prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Repeated collection and adjustment are performed to obtain the optimal content scheduling strategy, including: Adding the collected real-time video stream, real-time audio stream, and real-time text stream data to a pre-built two-way interactive evaluation network, wherein the two-way interactive evaluation network includes a state encoder and an effect evaluator, wherein the state encoder includes a video feature extraction module, an audio feature extraction module, and a text feature extraction module, and the effect evaluator includes a causal graph construction module, a structural equation model, and a multi-layer perceptron network; Extracting frame-level visual features from a real-time video stream using a convolutional neural network in the video feature extraction module, extracting audio spectrogram features from a real-time audio stream using a convolutional neural network in the audio feature extraction module, and extracting semantic features from a real-time text stream using a recurrent neural network in the text feature extraction module. Inputting the frame-level visual features, the audio spectrogram features, and the semantic features into an attention network to generate a feature weight matrix. Based on the feature weight matrix, weighted fusion of the three features is performed to obtain a multimodal representation vector. The causal graph construction module constructs a live broadcast effect causal graph based on historical data. The live broadcast effect causal graph includes a scheduling strategy node, a live broadcast status node, an external factor node, and an effect indicator node. The edges between the nodes represent causal relationships. The structural equation model defines the functional relationship between each node. The multimodal representation vector and the preset scheduling strategy vector are spliced ​​to form an input vector. Causal reasoning is performed based on the causal graph and the structural equation model to obtain a policy intervention effect. The input vector and the policy intervention effect are input into the multi-layer perceptron network. The multi-layer perceptron network calculates the predicted number of viewers, the predicted number of likes, and the predicted reward amount, counts the real-time number of viewers, the real-time number of likes, and the real-time reward amount, and calculates the weighted average deviation between the predicted value and the actual value to obtain a real-time performance evaluation index. Calculating a deviation between the real-time performance evaluation index and a stored live broadcast peak period prediction result, comparing the deviation with a preset deviation threshold, and triggering a policy adjustment when the deviation is greater than the deviation threshold; The live content scheduling task is decomposed into content recommendation subtask, anchor interaction subtask, and audience activity subtask. Each subtask is configured with an independent policy network and value function network. The policy network generates an action probability distribution based on the current state, and the value function network evaluates the value of the state-action pair. A gradient meta-learner is trained for each subtask. The gradient meta-learner adopts a model-independent meta-learning algorithm. In the inner loop, the policy gradient method is used to update the network parameters based on the latest state data, action data, and reward data. In the outer loop, the meta-parameters are updated by minimizing the meta-objective loss across tasks. The action probability distributions output by the updated policy networks corresponding to the three subtasks are normalized and fused through a soft voting mechanism to obtain the action probability distribution of the global scheduling strategy. Based on the action probability distribution, live content scheduling actions are sampled and selected, and the optimal content scheduling strategy is obtained by combination. While executing the scheduling action, new live broadcast data is continuously collected, and the newly collected data is repeatedly input into the two-way interactive evaluation network for evaluation and optimization until the deviation value between the real-time performance evaluation index and the live broadcast peak period prediction result is less than the deviation threshold.

[0052] In an optional embodiment, A bidirectional interactive evaluation network is built, consisting of a state encoder and an effect evaluator. The state encoder is responsible for extracting features from multi-source data. A convolutional neural network is used to extract visual features of each frame from the real-time video stream, such as color, texture, and edges. Similarly, a convolutional neural network is used to extract audio spectrogram features, such as pitch, loudness, and rhythm, from the real-time audio stream. A recurrent neural network is used to extract semantic features, such as user sentiment and topic keywords, from real-time text streams (such as bullet comments and comments). The extracted visual, audio, and semantic features are input into an attention network. The attention network generates a feature weight matrix based on the importance of different features. Based on this weight matrix, the three features are weighted and fused to obtain a multimodal representation vector that comprehensively reflects the status of the current live broadcast.

[0053] The effect evaluator is responsible for evaluating the effectiveness of the current live content scheduling strategy. It constructs a live broadcast effect causal graph based on historical live broadcast data. This causal graph includes scheduling strategy nodes (e.g., game segments, talent shows, audience interactions), live broadcast status nodes (e.g., number of online users, number of likes, number of comments), external factor nodes (e.g., competitor live content, holiday effects), and performance indicator nodes (e.g., number of viewers, number of likes, and amount of reward). Edges between nodes represent causal relationships. For example, the scheduling strategy node "game segments" affects the live broadcast status node "number of online users," while the external factor node "holiday effects" also affects "number of online users." A structural equation model is used to define the functional relationships between nodes and quantify the degree of causal influence. The multimodal representation vector output by the state encoder is concatenated with the preset scheduling strategy vector to form an input vector. Based on the constructed causal graph and structural equation model, causal inference is performed to derive the effect of the policy intervention, that is, to predict the impact of the policy on the performance indicators. The input vector and the policy intervention effect are input into a multi-layer perceptron network, which outputs predicted number of viewers, predicted number of likes, and predicted amount of reward. Simultaneously, real-time statistics of the number of viewers, number of likes, and amount of reward are generated. The real-time performance evaluation index is obtained by calculating the weighted average deviation between the predicted value and the actual value.

[0054] The real-time performance evaluation indicator is compared with the deviation value of the pre-stored live peak period prediction result. The pre-stored live peak period prediction result is a live peak period performance indicator predicted based on historical data. If the deviation value between the real-time performance evaluation indicator and the live peak period prediction result exceeds the pre-set deviation threshold, the policy adjustment mechanism is triggered.

[0055] The policy adjustment mechanism, based on a fast-adaptive algorithm using meta-reinforcement learning, decomposes the live content scheduling task into content recommendation, host interaction, and audience activity subtasks. Each subtask is configured with an independent policy network and value function network. The policy network generates an action probability distribution based on the current state. For example, in the content recommendation subtask, the policy network outputs the probability of recommending different content. The value function network evaluates the value of state-action pairs, predicting the long-term benefit of taking a particular action. A gradient meta-learner is trained for each subtask. The gradient meta-learner utilizes a model-agnostic meta-learning algorithm. In the inner loop, a policy gradient method is used to update the parameters of the policy and value function networks based on the latest state data, action data, and reward data (e.g., number of viewers, number of likes, and amount of tips). In the outer loop, the cross-task meta-objective loss is minimized and the meta-parameters are updated, enabling the model to quickly adapt to new tasks. The action probability distributions output by the updated policy networks for the three subtasks are normalized. The policies of the three subtasks are integrated through a soft voting mechanism to obtain the action probability distribution of the global scheduling policy. Based on this probability distribution, we sample and select live content scheduling actions, and combine them to form the optimal content scheduling strategy. For example, we can recommend a specific game content based on the global strategy, cooperate with the anchor to interact with it, and carry out relevant audience activities.

[0056] While scheduling actions are being executed, new live broadcast data is continuously collected. This newly collected data is repeatedly fed into the bidirectional interactive evaluation network for evaluation and optimization until the deviation between the real-time performance evaluation indicator and the live broadcast peak forecast is less than a preset deviation threshold. For example, during a certain time period, the real-time number of viewers is 1,000, the real-time number of likes is 5,000, and the real-time reward amount is 1,000 yuan. The predicted number of viewers is 1,200, the predicted number of likes is 6,000, and the predicted reward amount is 1,200 yuan. Assuming the weights are all 0.33, the deviation of the real-time performance evaluation indicator is (|1,000 - 1,200| + |5,000 - 6,000| + |1,000 - 1,200|) / 3 × 0.33 = 133.33. Assuming the live broadcast peak forecast results are 1,500, 10,000, and 2,000, and the deviation threshold is 200, the current deviation is less than the threshold, and no strategy adjustment is required.

[0057] In this embodiment, through cross-modal information fusion and causal reasoning, the factors affecting the live broadcast effect are considered more comprehensively, thereby improving the accuracy of the live broadcast effect prediction. The meta-reinforcement learning algorithm can quickly adjust the strategy according to real-time data, so that the live broadcast content scheduling plan can adapt to the ever-changing live broadcast environment and user preferences. By dynamically optimizing the live broadcast content scheduling strategy, it can better meet user needs, increase the attractiveness of the live broadcast content and user participation, and ultimately improve the overall effect of the live broadcast.

[0058] In an optional embodiment, The gradient meta-learner adopts a model-independent meta-learning algorithm. In the inner loop, the policy gradient method is used to update the network parameters based on the latest state data, action data, and reward data. In the outer loop, the meta-parameters are updated by minimizing the cross-task meta-objective loss. The update includes: Construct a meta-learner network structure, set up a policy network to generate action probability distributions, configure a meta-optimizer to adjust policy network parameters, implement the policy network as a deep neural network structure, and initialize meta-parameters shared by the policy network and meta-optimizer; For the pre-obtained subtasks, collect historical status data, action data, reward data, and effect data for each subtask to construct a training task, and obtain new status data, action data, reward data, and effect data from the real-time live broadcast scene to construct a test task; Sampling training tasks to obtain state transition data, copying meta-parameters as initial values ​​of task-specific parameters, starting the policy network based on the task-specific parameters, inputting the current state to generate an action probability distribution and sampling actions, executing the actions to obtain immediate rewards and next states, calculating the state value functions of the immediate rewards and next states, calculating the temporal difference error using the Bellman equation to obtain the action value function, taking the negative product of the action value function and the logarithm of the action probability output by the policy network to obtain the policy loss, calculating the gradient of the policy loss with respect to the task-specific parameters, updating the task-specific parameters along the gradient direction, and repeating the policy network forward propagation and gradient update until the inner loop training is completed; Calculate the cumulative reward difference of the task-specific parameters before and after the update of each training task on the corresponding task as the fitness metric, construct a meta-objective loss function by taking the weighted average of the fitness metrics of all tasks, calculate the second-order derivative of the meta-objective loss function with respect to the meta-parameters, and drive the meta-optimizer to update the meta-parameters based on the second-order derivatives to minimize the meta-objective loss function; Sample the test task, copy the trained meta-parameters as the initial values ​​of the task-specific parameters, and execute the inner loop update to obtain the policy network adapted to the current task.

[0059] In an optional embodiment, Build the meta-learner network structure. The policy network uses a deep neural network structure to output action probability distributions. The meta-optimizer is used to adjust the policy network parameters. The policy network and meta-optimizer share a set of initial meta-parameters, such as network weights and biases.

[0060] Prepare training and testing tasks. For a specific type of decision-making problem, such as game control or robot navigation, collect historical data for multiple subtasks, including state, action, reward, and effect data, to build a training task set. Simultaneously, acquire new state, action, reward, and effect data from live broadcast scenes to build a test task set. For example, in a game scenario, the state can be pixels on the game screen, the action can be the direction of controlling the character's movement, the reward can be the game score, and the effect can be the change in the character's health. Suppose we have collected historical data for 100 subtasks as a training set, and each subtask contains 1,000 state transition data.

[0061] Perform meta-learning training by randomly sampling a task from the training task set and copying the current meta-parameters as the task-specific initial parameters. Initialize the policy network using the task-specific parameters. Input the current state into the policy network to obtain an action probability distribution, and sample an action based on this distribution. After executing the action, the environment returns the immediate reward and the next state. Use a value network to evaluate the value of the current and next states, and calculate the temporal difference error using the Bellman equation to obtain the value of the action. Multiply the value of the action by the logarithm of the corresponding action probability output by the policy network and take the negative value to obtain the policy loss. Calculate the gradient of the policy loss with respect to the task-specific parameters and update the task-specific parameters along the gradient direction. Repeat the forward propagation and parameter update process of the policy network for, for example, 10 iterations to complete the inner loop training. Assume that the learning rate is set to 0.001.

[0062] After completing the inner loop update for a training task, the difference in the cumulative reward for the task-specific parameters before and after the update is calculated as the fitness metric for that task. A weighted average of the fitness metrics for all sampled training tasks is taken to construct a meta-objective loss function. The second-order derivative of the meta-objective loss function with respect to the meta-parameters is calculated, for example, using the Hessian vector product method. A meta-optimizer, such as the Adam optimizer, is driven to update the meta-parameters based on the calculated second-order derivatives to minimize the meta-objective loss function.

[0063] To test, randomly sample a task from the test task set and copy the trained meta-parameters as the task-specific initial parameters. Perform the same inner loop update process as in the training phase to obtain a policy network adapted to the current test task.

[0064] This embodiment enables the model to quickly adapt to new tasks and learn effective strategies with only a small amount of data. For example, when faced with a new game level, the model can quickly learn a strategy for clearing it. Through the meta-learning process, the model can learn cross-task knowledge, thereby improving generalization performance on unseen tasks. It can effectively learn from a small number of samples, reducing data collection and training costs.

[0065] Figure 2 This is a structural diagram of a system for predicting hot spots during peak periods of private domain live broadcasts and content scheduling based on deep learning according to an embodiment of the present invention. Figure 2 As shown, the system includes: The first unit is used to collect historical private domain live broadcast data and construct a hybrid memory-enhanced neural network including a fast memory pathway and a slow memory pathway. The fast memory pathway performs short-term dependency modeling through a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies through a hierarchical sparse attention structure. The private domain live broadcast historical data is input into the hybrid memory-enhanced neural network, multi-scale temporal features are extracted through the fast and slow memory pathways, a spatial pyramid attention module is constructed, and feature recombinations are performed on the multi-scale temporal features according to the parallel attention branches in the spatial pyramid attention module. The feature receptive field is adjusted in combination with deformable convolution to obtain a hierarchical feature representation. The hierarchical feature representation is input into a progressive feature distillation network, which performs multi-level knowledge transfer and feature recombinations to generate a live broadcast peak period prediction result. The second unit is used to collect multimodal data related to live broadcast and perform feature extraction and feature encoding, construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network, combine the topology structure generator to build node association relationships, perform cross-modal feature fusion through the feature interaction engine, apply a hierarchical contrast learning training method to the dynamic heterogeneous graph neural network, construct sample pairs at different semantic levels and perform feature comparison, input the feature comparison results into a hierarchical strategy optimization network, divide the global goal into multiple sub-goals through the goal decomposition module, plan the sub-goal realization path according to the execution planning module, input the hierarchical strategy optimization network into a non-cooperative game multi-agent reinforcement learning framework, optimize the strategy space through multi-agent competition and collaboration, and output the live content scheduling strategy; The third unit is used to collect real-time live broadcast data and add it to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features according to a cross-modal fusion method. The effect evaluator establishes a mapping relationship between predicted effects and actual effects based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Collection and adjustment are repeated to obtain the optimal content scheduling strategy.

[0066] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0067] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0068] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting hot spots and scheduling content during peak periods of private domain live broadcasts based on deep learning, characterized in that: include: Collect private domain live broadcast history data, build a hybrid memory enhanced neural network including a fast memory pathway and a slow memory pathway, the fast memory pathway performs short-term dependency modeling through a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies through a hierarchical sparse attention structure, input the private domain live broadcast history data into the hybrid memory enhanced neural network, extract multi-scale temporal features through the fast memory pathway and the slow memory pathway, build a spatial pyramid attention module, perform feature reorganization on the multi-scale temporal features according to the parallel attention branches in the spatial pyramid attention module, adjust the feature receptive field in combination with deformable convolution, obtain a hierarchical feature representation, input the hierarchical feature representation into a progressive feature distillation network, perform multi-level knowledge transfer and feature reorganization operations, and generate a live broadcast peak prediction result; Collect multimodal data related to live broadcast and perform feature extraction and feature encoding, construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network, combine the topology structure generator to build node association relationships, perform cross-modal feature fusion through the feature interaction engine, apply a hierarchical contrast learning training method to the dynamic heterogeneous graph neural network, construct sample pairs at different semantic levels and perform feature comparison, input the feature comparison results into a hierarchical strategy optimization network, divide the global goal into multiple sub-goals through a goal decomposition module, plan the sub-goal realization path according to the execution planning module, input the hierarchical strategy optimization network into a non-cooperative game multi-agent reinforcement learning framework, optimize the strategy space through multi-agent competition and collaboration, and output a live content scheduling strategy; Real-time live broadcast data is collected and added to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features according to a cross-modal fusion method. The effect evaluator establishes a mapping relationship between predicted effect and actual effect based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Collection and adjustment are repeated to obtain the optimal content scheduling strategy.

2. The method according to claim 1, characterized in that Collecting private domain live broadcast history data, constructing a hybrid memory enhanced neural network including a fast memory pathway and a slow memory pathway, wherein the fast memory pathway performs short-term dependency modeling through a dynamic gating mechanism, and the slow memory pathway captures long-term dependencies through a hierarchical sparse attention structure, inputting the private domain live broadcast history data into the hybrid memory enhanced neural network, extracting multi-scale temporal features through the fast memory pathway and the slow memory pathway, constructing a spatial pyramid attention module, and reorganizing the multi-scale temporal features according to the parallel attention branches in the spatial pyramid attention module, including: The historical data of private domain live broadcast is sampled every minute, and the number of viewers, likes, comments, and reward amount data are collected. The weighted average calculation is performed on the number of viewers, likes, comments, and reward amount data to obtain historical trend characteristics; A fast memory path and a slow memory path of a hybrid memory-enhanced neural network are set, a bidirectional gated recurrent unit network is configured in the fast memory path, a candidate state generation unit is set in the bidirectional gated recurrent unit network, and a nonlinear combination of a current input and a historical state is calculated by the candidate state generation unit to generate a candidate hidden state; the historical state is randomly forgotten by a reset gate, and the candidate hidden state is updated by an update gate; A converter network is configured in the slow memory path, a multi-head self-attention layer is combined with a feedforward neural network layer to construct a computing block, a query vector, a key vector, and a value vector are projected into multiple feature subspaces in the multi-head self-attention layer, a scaled dot product calculation is performed in each feature subspace to obtain an attention weight, dropout regularization is performed on the attention weight, two layers of perceptrons are connected in series in the feedforward neural network layer, wherein the first layer of perceptrons performs dimensional expansion, and the second layer of perceptrons performs dimensional compression, and a normalization layer is added between the layers; Performing an adaptive batch normalization operation on the private domain live broadcast history data and the historical trend features to obtain normalized data, reorganizing the normalized data into a three-dimensional array including a day dimension, a minute dimension, and an indicator dimension, inputting the three-dimensional array into the fast memory path, bidirectionally scanning along the day dimension to obtain local time series features, inputting the three-dimensional array into the slow memory path, adding time series position information through position coding, and obtaining global time series features; Configure the first attention branch, the second attention branch and the third attention branch of the spatial pyramid attention module, set a sparse attention mechanism in the first attention branch, generate an adaptive attention mask, perform information aggregation of key temporal positions through Top-k selection, set a dense attention mechanism in the second attention branch, perform feature extraction of local areas through a deformable sliding window, set a hybrid attention mechanism in the third attention branch, cascade channel attention and spatial attention, extract cross-channel and cross-spatial dimension dependencies, align the local temporal features with the global temporal features in the time dimension and input them into the spatial pyramid attention module, and perform weighted fusion on the output features of the three attention branches to obtain multi-scale fusion features.

3. The method according to claim 1, characterized in that Combined with deformable convolution to adjust the feature receptive field, a hierarchical feature representation is obtained, and the hierarchical feature representation is input into the progressive feature distillation network to perform multi-level knowledge transfer and feature reorganization operations to generate the live broadcast peak period prediction results including: The offset prediction layer, the modulation factor prediction layer and the feature sampling layer of the multi-layer deformable convolutional network are configured, the two-dimensional offset of the sampling position of the pre-acquired multi-scale fusion feature is calculated in the offset prediction layer, the modulation factor of the sampling position is generated in the modulation factor prediction layer, the feature response is adaptively adjusted, the bilinear interpolation sampling is performed on the feature map in combination with the two-dimensional offset and the modulation factor in the feature sampling layer, the multi-scale fusion feature is sequentially input into each layer of the multi-layer deformable convolutional network, cross-layer connections are added between adjacent layers, and a hierarchical feature representation is generated; A teacher network and a multi-level student network of a progressive feature distillation network are configured, parameters of the teacher network are initialized according to the hierarchical feature representation, each level of the student network corresponds to a feature layer of different depths, features of the student networks at adjacent levels are reused through dense residual connections, an attention distillation module is introduced into the teacher network, an attention map is used as an intermediate supervision signal, the hierarchical feature representation is input into the progressive feature distillation network, the L2 loss and the attention map loss of the feature maps of the teacher network and the student network are calculated, the KL divergence loss corresponding to the soft label distribution is calculated, a total loss is obtained by weighted combination of multiple loss items, network parameters are optimized based on the total loss, and a distilled feature representation is generated; The distilled feature representation is input into a multilayer perceptron, in which the feature dimension is reduced layer by layer, and in the last layer the output dimension is set to the number of discrete time periods in the prediction time window; the output is normalized by the Softmax function to obtain the peak probability of each time period, and the optimal threshold is determined by combining the dynamic probability threshold selection algorithm, and the time period above the optimal threshold is marked as the predicted peak period.

4. The method according to claim 1, characterized in that: Collect multimodal data related to live broadcast and perform feature extraction and feature encoding, construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network, combine the topology structure generator to build node associations, perform cross-modal feature fusion through the feature interaction engine, apply a hierarchical contrast learning training method to the dynamic heterogeneous graph neural network, construct sample pairs at different semantic levels and perform feature comparison, input the feature comparison results into a hierarchical strategy optimization network, divide the global goal into multiple sub-goals through the goal decomposition module, plan the sub-goal realization path according to the execution planning module, input the hierarchical strategy optimization network into a non-cooperative game multi-agent reinforcement learning framework, optimize the strategy space through multi-agent competition and collaboration, and output the live content scheduling strategy, including: Perform feature extraction on each frame in the video stream data through a feature extraction network, the feature extraction network includes a feature pyramid network and an attention enhancement module, extract multi-scale visual features through the feature pyramid network, perform feature enhancement in the channel dimension and the spatial dimension through the attention enhancement module, perform short-time Fourier transform on the audio data in combination with the multi-head self-attention network to obtain a spectrogram, input the spectrogram into a deep network to extract audio features, use a bidirectional encoding network to perform sequence modeling on the barrage text and the comment text, and map the word vector sequence into a text feature vector; input the visual features, the audio features, and the text feature vector into a temporal encoding network, perform temporal encoding to obtain a multimodal feature tensor; Construct a dynamic heterogeneous graph neural network, take anchor nodes, audience nodes, video clip nodes, audio clip nodes, bullet screen nodes, and comment nodes as graph nodes, assign the multimodal feature tensor to the corresponding graph nodes, calculate the similarity between node pairs through a topology generator, perform a normalization transformation on the similarity to obtain edge weights, construct a node adjacency matrix in combination with a sparse matrix representation, and use a feature matrix to store node attributes to obtain a feature representation of the node; The dynamic heterogeneous graph neural network is dynamically updated. When new data arrives, the corresponding modal features are extracted and new nodes are created. The feature similarity between the newly added nodes and the existing nodes corresponding to the new data is calculated. Connecting edges are added between node pairs whose similarity is greater than a preset first threshold. The product of the degree centrality of the existing nodes and the time decay factor is calculated. Nodes whose product is less than a preset second threshold are deleted. The exponential sliding average of the edge weights is calculated. The edges whose sliding average is less than a preset third threshold are deleted to obtain an optimized graph structure. The feature matrix and the node adjacency matrix are input into a multi-layer graph neural network through a feature interaction engine, wherein the feature matrix is ​​composed of a multimodal feature tensor of nodes, representing the initial features of each node, and a learnable parameter matrix is ​​constructed according to the input feature dimension and the hidden layer dimension. First, the input dimension and output dimension of each layer of the graph neural network are set, and the learnable parameter matrix is ​​initialized using a uniform distribution or a normal distribution. The learnable parameter matrix is ​​used as the weight matrix of each layer of the graph neural network, and the node adjacency matrix is ​​degree-normalized to obtain a symmetric normalized matrix. The normalized matrix is ​​multiplied by the node feature matrix of the current layer, and then multiplied by the learnable parameter matrix. A nonlinear activation function is used to obtain the symmetric normalized matrix. Get the node features of the next layer, apply the hierarchical contrast learning method in each layer, construct positive sample pairs and negative sample pairs, the positive sample pairs are sampled from adjacent nodes, and the negative sample pairs are sampled from non-adjacent nodes at different semantic levels, calculate the feature contrast loss, and perform weighted average of the contrast losses at different levels to obtain the total loss. Input the node representation output by the graph neural network into the hierarchical strategy optimization network, set the interaction quality improvement sub-goal, content form optimization sub-goal, and user experience optimization sub-goal in the goal decomposition module, calculate the sub-goal weights, and set the resource scheduling sub-network in the execution planning module, wherein the resource scheduling sub-network includes a multi-layer encoder, and the output of the encoder is mapped to obtain the scheduling action space; The nodes in the dynamic heterogeneous graph neural network are mapped to multiple agents. Each agent maintains a local state representation and a policy network. The policy network consists of an action generation network and a state evaluation network. The action generation network outputs the action probability distribution, and the state evaluation network evaluates the state value function. The agents exchange information through a competitive cooperation mechanism, optimize strategies based on non-cooperative games, and calculate rewards based on the joint actions of the agents. The reward function comprehensively considers multiple indicators such as interactivity, content quality, and user feedback to obtain a comprehensive evaluation score. Each agent generates a probability distribution of live content scheduling strategy based on its strategy network, and performs parameterized sampling on the probability distribution to obtain discrete actions. The discrete actions of all agents are combined into a live content scheduling strategy, including the host interaction instruction sequence, the audience activity push sequence, and the platform resource allocation sequence. The structure and node characteristics of the dynamic heterogeneous graph neural network are updated according to the feedback after executing the scheduling strategy.

5. The method according to claim 4, characterized in that The agents exchange information through a competitive cooperation mechanism, optimize strategies based on non-cooperative games, and calculate rewards based on the joint actions of the agents, including: An agent is deployed at each node of a dynamic heterogeneous graph, wherein the nodes of the dynamic heterogeneous graph include video clip nodes, audio clip nodes, and bullet-screen nodes. The state space of the agent is composed of the attribute feature vector of the corresponding node and the attribute feature vector of the adjacent node. The state vector of the video clip node includes the visual features of the video frame, the acoustic features of the audio clip, and the text features of the surrounding bullet-screen nodes. The action space of the agent includes picture clarity adjustment, interactive question insertion, and related video push. The agent broadcasts its own action probability distribution to neighboring nodes at each decision time step, receives the action probability distribution of neighboring nodes, and uses the received action probability distribution as an additional input signal to update the corresponding strategy network, wherein the strategy network includes an action generation network and a value evaluation network; Strategy optimization is performed based on a non-cooperative game framework. The objective functions of different types of nodes in the non-cooperative game include the audience participation of the anchor node, the viewing time of the video node, and the number of interactions of the bullet comment node. The agent takes its own actions to obtain immediate rewards based on the observed state and the actions of other agents, and receives trajectory data of the interaction between other agents and the environment to estimate the joint action value function; Iteratively update the joint action value function, calculate the value function of the current agent state and the joint action, and perform weighted average of the difference between the value function and the sum of the instant reward and the discounted value of the state value function at the next moment to obtain the updated value function, wherein the instant reward is obtained according to the environment, and the weight of the weighted average is the learning rate; The policy function is optimized based on the policy gradient algorithm, and the Monte Carlo estimation method is used to calculate the policy gradient of each agent's objective function. The product of the policy gradient and the learning rate is added to the parameters of the agent's policy function to obtain the updated policy function parameters. The agent generates a scheduling strategy based on the optimized policy function, and performs picture clarity adjustment, interactive question insertion, and related video push according to the scheduling strategy.

6. The method according to claim 1, characterized in that Collect real-time live broadcast data and add it to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features according to a cross-modal fusion method. The effect evaluator establishes a mapping relationship between the predicted effect and the actual effect based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak prediction result is greater than a preset deviation threshold, the adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy. Repeated collection and adjustment to obtain the optimal content scheduling strategy includes: Adding the collected real-time video stream, real-time audio stream, and real-time text stream data to a pre-built two-way interactive evaluation network, wherein the two-way interactive evaluation network includes a state encoder and an effect evaluator, wherein the state encoder includes a video feature extraction module, an audio feature extraction module, and a text feature extraction module, and the effect evaluator includes a causal graph construction module, a structural equation model, and a multi-layer perceptron network; Extracting frame-level visual features from a real-time video stream through a convolutional neural network in the video feature extraction module, extracting audio spectrogram features from a real-time audio stream through a convolutional neural network in the audio feature extraction module, extracting semantic features from a real-time text stream through a recurrent neural network in the text feature extraction module, inputting the frame-level visual features, the audio spectrogram features, and the semantic features into an attention network to generate a feature weight matrix, and performing weighted fusion on the three features based on the feature weight matrix to obtain a multimodal representation vector; The causal graph construction module constructs a live broadcast effect causal graph based on historical data, the live broadcast effect causal graph includes a scheduling strategy node, a live broadcast status node, an external factor node and an effect indicator node, the edges between the nodes represent causal relationships, the structural equation model defines the functional relationship between the nodes, the multimodal representation vector is spliced ​​with the preset scheduling strategy vector to form an input vector, causal reasoning is performed based on the causal graph and the structural equation model, the policy intervention effect is obtained, the input vector and the policy intervention effect are input into the multi-layer perceptron network, the multi-layer perceptron network calculates the predicted number of viewers, the predicted number of likes, and the predicted reward amount, counts the real-time number of viewers, the real-time number of likes, and the real-time reward amount, and calculates the weighted average deviation between the predicted value and the actual value to obtain the real-time performance evaluation index; Calculate the deviation value between the real-time performance evaluation index and the stored live broadcast peak period prediction result, compare the deviation value with a preset deviation threshold, and trigger strategy adjustment when the deviation value is greater than the deviation threshold; The live content scheduling task is decomposed into content recommendation subtask, anchor interaction subtask, and audience activity subtask. Each subtask is configured with an independent policy network and value function network. The policy network generates an action probability distribution based on the current state, and the value function network evaluates the value of the state-action pair. A gradient meta-learner is trained for each subtask. The gradient meta-learner adopts a model-independent meta-learning algorithm. In the inner loop, the policy gradient method is used to update the network parameters based on the latest state data, action data, and reward data. In the outer loop, the meta-parameters are updated by minimizing the meta-objective loss across tasks. The action probability distributions output by the policy network corresponding to the three updated subtasks are normalized, and the action probability distribution of the global scheduling strategy is obtained by fusion through a soft voting mechanism. The live content scheduling action is sampled and selected based on the action probability distribution, and the optimal content scheduling strategy is obtained by combination. While executing the scheduling action, new live broadcast data is continuously collected, and the newly collected data is repeatedly input into the two-way interactive evaluation network for evaluation and optimization until the deviation value between the real-time performance evaluation index and the live broadcast peak period prediction result is less than the deviation threshold.

7. The method according to claim 6, characterized in that The gradient meta-learner adopts a model-independent meta-learning algorithm, uses a policy gradient method in the inner loop to update network parameters based on the latest state data, action data, and reward data, and minimizes the cross-task meta-objective loss in the outer loop to update the meta-parameters, including: Construct a meta-learner network structure, set a policy network to generate action probability distribution, configure a meta-optimizer to adjust policy network parameters, implement the policy network as a deep neural network structure, and initialize meta-parameters shared by the policy network and the meta-optimizer; For the pre-obtained subtasks, collect the historical status data, action data, reward data, and effect data of each type of subtask to construct a training task, and obtain new status data, action data, reward data, and effect data from the real-time live broadcast scene to construct a test task; Sampling training tasks to obtain state transition data, copying meta-parameters as initial values ​​of task-specific parameters, starting the policy network based on the task-specific parameters, inputting the current state to generate an action probability distribution and sampling actions, executing the actions to obtain immediate rewards and next states, calculating the state value functions of the immediate rewards and next states, using the Bellman equation to calculate the temporal difference error to obtain the action value function, multiplying the action value function with the logarithm of the action probability output by the policy network and taking the negative value to obtain the policy loss, calculating the gradient of the policy loss with respect to the task-specific parameters, updating the task-specific parameters along the gradient direction, and repeatedly executing the policy network forward propagation and gradient update until the inner loop training is completed; Calculate the cumulative reward difference of the task-specific parameters before and after each training task is updated on the corresponding task as a fitness metric, construct a meta-objective loss function by weighted average of the fitness metrics of all tasks, calculate the second-order derivative of the meta-objective loss function with respect to the meta-parameters, and drive the meta-optimizer to update the meta-parameters based on the second-order derivatives to minimize the meta-objective loss function; Sample the test task, copy the trained meta-parameters as the initial values ​​of the task-specific parameters, and execute the inner loop update to obtain the policy network that adapts to the current task.

8. A deep learning-based private domain live broadcast peak hotspot prediction and content scheduling system, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect private domain live broadcast history data, build a hybrid memory enhanced neural network including a fast memory path and a slow memory path, wherein the fast memory path performs short-term dependency modeling through a dynamic gating mechanism, and the slow memory path captures long-term dependency through a hierarchical sparse attention structure, input the private domain live broadcast history data into the hybrid memory enhanced neural network, extract multi-scale temporal features through the fast memory path and the slow memory path, build a spatial pyramid attention module, perform feature recombination on the multi-scale temporal features according to the parallel attention branches in the spatial pyramid attention module, adjust the feature receptive field in combination with deformable convolution, obtain a hierarchical feature representation, input the hierarchical feature representation into a progressive feature distillation network, perform multi-level knowledge transfer and feature recombination operations, and generate a live broadcast peak prediction result; The second unit is used to collect multimodal data related to live broadcast and perform feature extraction and feature encoding, construct a multimodal feature tensor and add it to a pre-set dynamic heterogeneous graph neural network, combine the topology structure generator to build node association relationships, perform cross-modal feature fusion through the feature interaction engine, apply a hierarchical contrast learning training method to the dynamic heterogeneous graph neural network, construct sample pairs at different semantic levels and perform feature comparison, input the feature comparison results into a hierarchical strategy optimization network, divide the global goal into multiple sub-goals through a goal decomposition module, plan the sub-goal realization path according to the execution planning module, input the hierarchical strategy optimization network into a non-cooperative game multi-agent reinforcement learning framework, optimize the strategy space through multi-agent competition and collaboration, and output the live content scheduling strategy; The third unit is used to collect real-time live broadcast data and add it to a pre-built two-way interactive evaluation network including a state encoder and an effect evaluator. The state encoder extracts multi-source information features according to a cross-modal fusion method. The effect evaluator establishes a mapping relationship between predicted effect and actual effect based on a causal reasoning mechanism to obtain a real-time performance evaluation index. If the deviation value between the real-time performance evaluation index and the live peak prediction result is greater than a preset deviation threshold, an adjustment mechanism is triggered. According to a fast adaptation algorithm based on meta-reinforcement learning, the live content scheduling strategy is optimized online through task decomposition and gradient meta-learning. The live content scheduling plan is dynamically adjusted based on the optimized strategy, and the collection and adjustment are repeated to obtain the optimal content scheduling strategy.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images